Artificial Intelligence – Lesson 18 – Rabbi Michael Abraham

This is an English translation (via GPT-5.4). Read the original Hebrew version.

This transcript was produced automatically using artificial intelligence. There may be inaccuracies in the transcribed content and in speaker identification.

🔗 Link to the original lecture

🔗 Link to the transcript on Sofer.AI

Review of Chain of Thought and training with feedback on the process — the Rabbi explains how a model is trained not only on the final answer but also on the path of reasoning, and how that improves performance.
The fundamental question: why should human-style thinking improve results? — the puzzling issue is discussed whether imitating the order of human thought is really optimal, even though empirically it turns out to help.
Presentation of a post by Alexei Efros about understanding in models — the claim is brought that responding to instructions like “step by step” creates an illusion of understanding and anthropomorphism toward language models.
The “duck duck goose” experiment and the argument against understanding — an experiment is described in which the reasoning instruction was replaced with a gibberish phrase the model was trained on, and the model still improved performance to the same extent.
Sharp criticism of the interpretation of the experiment — the Rabbi argues that the experiment proves nothing about a lack of understanding, because it is equivalent to teaching an alternative language with a slightly different dictionary.
A methodology for testing philosophical arguments about AI — he suggests examining every experiment from the perspective of someone who thinks the model does understand, and asking whether the findings really undermine that view.
The example of Rabbi Aharon Levy and model hallucinations — a video is cited in which a model invents claims about Baal HaSulam, and then apologizes and admits error when asked for sources and support.
Hallucinations, temperature, and non-determinism — it is explained that models do not always give the same answer, because the output is selected statistically from a distribution and not only by the single most likely option.
The difference between a classical computer and a language model regarding “internal information” — the Rabbi emphasizes that information in a model is not stored as memory items but embodied in the organization of weights, without guaranteed retrieval.
The limits of trust in artificial intelligence in sensitive contexts — errors in sources, lawyers who submitted invented precedents, and the need for a human in the loop in medicine, law, and warfare are discussed.
Autonomous cars and the phobia of handing responsibility to a machine — even though AI is statistically safer than human beings, there is public resistance because people need some concrete bearer of responsibility.
Apologies, politeness, and the model’s “fear of Heaven” — the Rabbi analyzes the style of excessive apology as an effect of the wrapper, instructions, and external tools, not as an expression of emotion or consciousness.
A comparison to the question of human apology without emotion — the Rabbi distinguishes between a human apology that stems from understanding and judgment, and a model’s output of words behind which there is no conscious decision.
Analysis of the Baal HaSulam and Zionism case — the claim is that the model may actually have answered correctly to the question “what was he criticized for,” and only the later apology was mistaken because it switched the original question.
A drift into political issues of the Right, security, and Ra’am — at the end of the lesson, a discussion developed about responsibility, utilitarian judgment, Ben Gvir, Abbas, the “change government,” and attitudes toward Arabs in Israel.

Summary

General Overview

The lesson continued the discussion of artificial intelligence, especially language models and neural networks. The Rabbi opened by recalling the idea of Chain of Thought: when you train a model not only to give the correct answer but also to present its route of reasoning, and even give it feedback on the reasoning itself, results improve. From there the philosophical question arose: does the very imitation of human thinking indicate understanding, or only an efficient computational mechanism?

## The “Duck Duck Goose” experiment and the flaw in the conclusion
The Rabbi presented a post by a researcher from Berkeley who claimed that one can show that models do not “understand”: if you replace the instruction “think step by step” with a gibberish phrase like “duck duck goose,” and after appropriate training the model still produces good reasoning, then semantics is not important and there is no understanding here.

The Rabbi rejects this sharply. In his view, if the model was systematically trained to connect the expression “duck duck goose” with staged inference, then that is equivalent to teaching an alternative language. Like a child raised in a language where that expression means “think step by step” — it would be absurd to say that because of this he does not understand. Therefore the experiment does not prove a lack of understanding; it only shows that the model learned a slightly different dictionary.

## How to test arguments about understanding in models
The Rabbi’s methodological innovation is that one should not evaluate an argument from a position that already assumes the model does not understand. One should ask: if I thought it did understand, would this experiment force me to retract that view? If not — then there is no evidence here. In his view, a large part of the public and academic discourse about artificial intelligence suffers from this flaw and rests on anthropomorphism, conceptual confusion, or a failure to understand how the models are trained and operate.

## Hallucinations, temperature, and the absence of “internal information”
Later, a video was discussed in which a model gave false claims about Baal HaSulam and then apologized when asked to provide sources. From there the Rabbi moved to explain the phenomenon of hallucinations. A language model does not answer in a fully deterministic way; it operates on the basis of a probability distribution, and sometimes even samples among possibilities according to “temperature.” Therefore different answers to the same question are possible, including wrong ones.

A particularly important point was the distinction between a classical computer and a language model. In an ordinary computer, information is stored in memory at a defined location, so if it exists, it can be retrieved precisely. In a language model, the information does not “sit” inside as a clear unit of content; it is embodied in the arrangement of weights created during training. Therefore even information that the model has “learned” can be retrieved wrongly, partially, or inconsistently.

## Wrapper, politeness, and apology
The Rabbi also stressed that today’s models are not just a “bare neural network.” They are wrapped in tools, agents, prompts, and system instructions that affect the answers. That explains the politeness, the apologies, and sometimes even a “God-fearing” style in Torah-related contexts. There is no reason to infer emotion, awareness, or real respect from that; it is syntactic imitation or the product of an instructional wrapper.

In that context, the Rabbi compared this to a human apology. In the past he argued that an apology without emotion can actually be elevated if it stems from understanding and judgment. But in a model, he claims, even that is absent: not only is there no emotion there, there is also no conscious decision or intellectual judgment in the human sense.

## The Baal HaSulam case: sometimes the mistake is actually the apology
The Rabbi showed that in the case of the question about Baal HaSulam, the model may not have erred at all in its original answer. If it was asked “what was he criticized for,” it may be correct to answer that he was criticized for closeness to Zionism, even if there is no proof that he was actually a Zionist. When the model apologized, it effectively switched between two different questions: “what was he accused of” and “is the accusation true.” So sometimes the apology itself is the mistake.

## Conclusion
The central conclusion of the lesson is that claims about understanding, awareness, and thinking in language models must be analyzed with great care. One must not be dazzled by their human imitation, but one must also not draw hasty conclusions from shallow experiments. To understand what models do and do not do, one has to get back to the nuts and bolts: training, weights, probabilities, wrappers, and tools. Without that, most of the discussion turns into slogans and confusion.

Full Transcript

[Rabbi Michael Abraham] Okay. Last time we dealt a bit more with neural networks, and after I explained what a neural network is, I demonstrated a little of Chain of Thought, yes, how a neural network reconstructs a human way of thinking, and we saw that this reconstruction improves the result. Meaning, it manages to think better if we guide it to think in the way human beings think, according to the order of thought or the path of thought of human beings. How do you do that? You train the network in such a way that we give it feedback, ask it to output the path of reasoning, and give it feedback on the path of reasoning too, not only on the answer. And if it reaches the right answer with the wrong path of reasoning, negative feedback, you have to coordinate the weights, adjust the weights. Just one second, there’s shouting here. So I said that the training itself directs the network to answer with the path, yes, like on exams where the teacher asks: give me the answer together with the method, not only the final answer. So here too, the same thing. We’re basically telling the model: give me the answer together with the path, how you built it or how you got there, and it gets feedback on the path too, not only on the answer, until little by little it learns to do things the way human beings do them. And then, after we train it for that, now when we give it some question, it gives us both the path and the answer to the question. And when you test the result, it turns out that after the training that causes it to follow the human path of thought, the answers improve. Meaning, the answers it gives are better than what happened when it was trained only on the answer without the path of reasoning. And I talked about this a bit, yes, in the context of game theory. What? I can’t hear.

[Speaker B] Did someone speak? No, a mistake, a mistake.

[Rabbi Michael Abraham] Okay. So I talked about this in the context of game theory, yes, the question whether really thinking like a human being is supposed to improve the result or not, yes? It is not at all clear that our form of thinking is really optimal. Why assume that if it thinks like us it will arrive at better results? But it turns out that yes. Meaning, if you train it to think according to our order of thought, somehow the results come out better. Now, I want to show you — look, here is a post, okay? Someone named Alexei Efros, apparently a well-known person in his field. I don’t know him well enough, but he is well known in his field, a professor at Berkeley, and he basically brings, describes some kind of experiments they did that are supposed to test whether the model understands or does not understand, okay? So look, here he says the following: part of the magic of LLMs, yes, part of the magic of LLMs, is that they seem as though they understand us, yes? As though they understand us on a deep level. For example — I’m just translating — for example, when you tell an LLM, reason step by step, and suddenly, surprisingly, it really does that. Not only does it do that, but as we saw, its answer actually improves too, okay? Therefore, he says, the feeling created among users of the models is that the model really understands us. Meaning, we tell it to work step by step, to follow the form of reasoning, and it responds, it really does that. It doesn’t just give an answer; it also shows the reasoning process. Meaning, it seems that it really operates like us, understands us, and so on. But for my taste, he says, there is too much anthropomorphism of the models, yes, of the chatbots or the models. What does that mean? Humanization, yes. In other words, anthropomorphism means that we relate to them too much as human beings, and the interpretation we give to what we get from them is affected by that anthropomorphism, by that humanization that we impose on them. Meaning, the claim is that the humanization is an assumption at the beginning of the matter and not a result. Because usually, if we were reading the first sentence he wrote, then when we assume the model understands us, or infer that the model understands us, then basically what we got is a result of humanization. Meaning, the result of contact with the model is that our conclusion is apparently that the model probably understands us. But he says, no. We assume a humanization of the model, and therefore we get that impression. Yes, that is basically the claim. And then he says, yes, if one follows Saussure and Wittgenstein — Saussure is a French philosopher, and a French philosopher is more or less an oxymoron, for anyone who read my column about the law permitting summary justice against Frenchmen and their helpers — and Wittgenstein, following them he basically wants to argue that responding to instructions like, yes, infer this step by step, does not necessarily indicate understanding, yes, awareness of the semantics of the matter. It could simply be some mechanical activation of an inference mechanism. It sees the words “reason step by step,” and it basically activates some mechanical mechanism that produces better inference. That still doesn’t mean there is understanding here. All right? So up to this point these are simply two interpretive possibilities. But here is the paragraph important for our purposes. But thanks to two brilliant students, yes, thanks to two brilliant students, something from MIT, now we can already get evidence for the claim that these models do not understand. Meaning, these brilliant students came up with an idea how to test empirically the question whether the model understands or does not understand, okay? So he says, basically what they did was to cause the model, to train it, to do the inference “duck duck goose,” yes, never mind, a set of words, ducks, a set of contextless words, and they managed to get the model to receive this instruction, yes, “reason duck duck goose,” and the results it produced were exactly like the instructions when it receives “reason it step by step.” Meaning, basically, you can condition the operation of the model and improve the result even if you don’t use the words “do the inference step by step,” but also if you tell it blah blah blah blah blah. Meaning, if you trained it on the blah blah blah blah blah, then it too will produce the inference at a good level, and you don’t need to speak to it in words whose actual meaning is “do the inference step by step.” And therefore — after that, yes, he did another experiment, yes, “nn okay,” meaning read line read line and write okay, and then you ask it the question, and again, you can see that the model trained to do this will give you all the reasoning with the best answer. And therefore, therefore, basically his claim is that models do not — this impression that models really possess understanding is a mistaken impression. The moment you tell it mere gibberish, and even then, if you train it on that, it will perform the same action as when you give it the instruction with meaningful semantics to do it step by step. That is, that is basically the claim. Now, I deliberately opened this post for you so that you would see it inside, so that you would believe me that there really is such a post, because I wouldn’t have believed it. If someone had told me there was such a post, and from a talented person prominent in his field, I wouldn’t have believed it. I haven’t seen such nonsense in a long time. The claim he is basically making is that he has proof — not proof, an indication — that the model does not understand. Why? Because if you feed it gibberish, which has no meaning of “make an inference in this way or that way,” still, if you train it on that, it will carry out the inference in the best way just as if you gave it the instruction “do it step by step.” Now, does that really constitute an indication that the model does not understand? To test that systematically, what we basically need to do is ask ourselves: let’s say we hold that the model does understand. Okay? Now let’s ask: do these experimental findings that he describes persuade us that we are mistaken? Or are they an indication that we are mistaken?

[Speaker C] I don’t understand what is wrong with this experiment. After all, understanding language is based on understanding the semantics behind the syntax. If you give it mere syntax with no semantics behind it, and it still manages to interpret some semantics, some meaning behind it, then that shows it is just making things up and doesn’t really understand.

[Rabbi Michael Abraham] Okay, so that is what he claims. So I’m sorry, but what I said earlier applies to this too. I’ll explain in a moment, because I’m about to explain it. So look, let’s think about a child who is born, and in his language I teach him that “do it duck duck goose” means doing it step by step. Those are the words in that language whose meaning is: in Hebrew, work step by step. Do the inference step by step. But I raised this child with Hebrew except for one pathology. Instead of the words “do the inference step by step,” I say to him “do it duck duck goose.” All right? Or whatever you want. That’s what he learned from childhood, those are the words. Now, the child grows up, and he becomes a professor at Berkeley. Okay? And now I tell him “do it duck duck goose,” and he solves the test step by step, hands in marvelous results. Do you think that child is not really endowed with understanding?

[Speaker C] But then, of course he is, but then I simply think you failed in placing him behind the test. You didn’t really test him on meaningless words; those words have meaning.

[Rabbi Michael Abraham] Wait, how? How did they do the test there? Let’s try to think how they did the test there. Suppose we take this model and we want to test whether doing it “duck duck goose” will also give us the same results, in order to show that it does not understand. Now you understand that if you now take ChatGPT, okay, and tell it, find me, I don’t know, how many capital cities in Europe begin with the letter G, and do it in duck duck goose — tell it that, okay? You will get either a worse answer or the same answer as “how many capital cities in Europe begin with G.” Either it will give the same answer or a worse one, certainly not a better one. Why? Because it doesn’t — what is “duck duck goose”? It understands nothing there. No, usually it will ask you. Wait a second. How did they manage to reach the point where the model responds to “duck duck goose” and performs the whole chain of thought? Do you know how? When they trained it, they trained it on very many examples, and in all those examples, when it said “duck duck goose,” after that came a perfect solution in stages of human reasoning. That was its training material, okay? Or in other words, what they did was take the child and teach him Hebrew, except for the expression “do it step by step.” There, instead of that expression, they said “do it duck duck goose.” That’s all. But they taught him the same thing. So now I ask, what on earth does this experiment prove? It’s unbelievable. I don’t know, I would fire him from the university. Really, this stupidity is just unbelievable. And all he did was train the model — he trained the model on a language that is Hebrew with a slight variation, that’s all. Aside from that everything is the same, it’s just that in the model’s dictionary a different word appears. So when you tell it “duck duck goose,” in its language that means “do it step by step,” right? That is the meaning in its language, so it did it step by step. What exactly is the problem? It understands perfectly well. What do I mean by understands perfectly well? I too think the model does not understand. That is why I introduced this by saying: when I want to test whether such an experiment can persuade me or contribute to the persuasion that the model does not understand, I need to have before my eyes someone who thinks the model does understand and ask that person, how would you explain the results of this experiment? And the answer is, he would explain it very easily. He would explain it by saying that the model simply learned prime Hebrew, yes, Hebrew with a small correction, not ordinary Hebrew, and besides that it understands everything and responds to everything. Just as I can speak to it in English, I can speak to it in Hebrew, I can speak to it in prime Hebrew — what difference does it make? I speak to it and it is trained on those words, it understands those words and acts accordingly. The point is, if a person is convinced in advance that the model has no understanding, then of course it is very easy to swallow this nonsense and say, you see, this experiment shows it. But that is why I say: in order to test the quality of his argument, you have to begin from the premise that the model does possess understanding. I do not agree with that, but I’m saying that when I want to test his argument, I have to assume: what would I say as someone who thinks the model does possess understanding? And if I saw this experiment, I would not be impressed in any way. Rabbi?

[Speaker D] Yes. Why don’t you write to him? What? Why don’t you ask him?

[Rabbi Michael Abraham] What do I have to ask him?

[Speaker D] No, ask him about all this. Say, I read your post, it —

[Rabbi Michael Abraham] He has surely already received lots of criticisms like this. I didn’t even look; maybe it’s even there in the comments. It’s obvious. It’s nonsense. I don’t think anyone else fell for such a thing. And with ChatGPT, by the way, I just asked it what it thought of this, and it immediately produced this criticism. It’s obvious. Meaning, even without understanding it managed to understand that there is no proof here that a model does not understand. Yes, it’s just — there’s nothing here to —

[Speaker C] With the Rabbi’s permission, maybe — it really does sound very stupid — maybe there is some kind of —

[Rabbi Michael Abraham] Again, I didn’t come to slander him; that’s not the point. I’m using this example to begin seeing what all that we learned about language models means. So I’m trying to show you that many of the questions and arguments that come up in this context — whether a model understands or does not understand, whether it is a human being or not a human being, whether it has consciousness or not — you have to go back, think what this model means, what this model is, how it is trained, how it achieves the results being presented to you, and then return and ask the question. And when you do that, you will see that ninety-nine percent of what is written in this field is simply garbage, garbage, just people who don’t understand a thing. Now, that’s unbelievable because these are experts in the field, so it is probably their philosophical ability, but I don’t know — you don’t need to be some great philosopher here. It’s not that. It’s just a simple matter.

[Speaker D] He’s a physicist, actually. What? He’s a physicist, actually.

[Rabbi Michael Abraham] I don’t know, maybe.

[Speaker D] No, he’s the son of Efros, of the well-known Professor Efros. He’s a good friend of Perelman’s.

[Rabbi Michael Abraham] And therefore what? I didn’t understand.

[Speaker D] No, it could simply be that this is not his field at all.

[Rabbi Michael Abraham] Okay, this is not my field either, and still he wrote nonsense.

[Speaker D] No, but he wrote something and you’re checking it logically, and it could be that two people are talking about the same thing and you see it differently from him. Fine, it could be.

to be.

[Rabbi Michael Abraham] doesn’t see any other possibility; he just doesn’t see it. He’s simply an idiot.

[Speaker D] Fine, okay, that may not be

[Speaker C] maybe we understood him correctly, maybe

[Rabbi Michael Abraham] when he could be—again, the argument is idiotic. I don’t know, maybe he’s a genius and meant something else, I have no idea. But I’m saying: I’m bringing you an argument that is idiotic. Meaning, again, I spoke with ChatGPT, I also asked people in the company I work with who are experts in the field, everyone agreed: it’s just nonsense. He clearly didn’t mean anything—nobody saw anything meaningful that could be understood from his words. In other words, it’s just nonsense. But fine, maybe he means something hidden that I didn’t understand. Again, my goal is not to accuse him or put him down. I’m using this example to show you—I’m using the chain of thought I presented in the previous class, and through that to return, for example, to a typical argument. By the way, there’s a lot of this kind of nonsense floating around in these discussions, and I’m trying to show you how it’s worth approaching this kind of argument. And try to remember what a model is, how it is trained, what the meaning of the results you get is, and then think about the question. Because if you don’t come with that background, then it’s much easier to fall for some argument of this type, to be convinced by an argument of this type, because for a moment you forget the machinery standing behind this machine we’re talking to here.

[Speaker C] But according to the way you’re presenting it, is it even possible—according to the way you explained to us what a model is—is it even possible to do an experiment like this that shows it has no consciousness?

[Rabbi Michael Abraham] That it has no understanding, excuse me. I at least don’t manage to see such an experiment. But we’ll attack this from a few more angles in a moment. So there’s—this is point number one. Now I want to show you another nice source, really amusing. One second. Strange.

[Speaker F] They told me that if I’m working and I have a chronic illness, I may be eligible for a tax exemption. But what is a chronic illness?

[Speaker E] A chronic illness is usually an illness that requires regular medication—wait, one second. I have something here we can learn from. Okay.

[Rabbi Michael Abraham] So there’s a certain man here named Rabbi Aharon Levy. Again, we’re talking about years ago—you have to understand, with today’s models I don’t think everything he describes would happen. But he describes something very amusing here, and I think we can learn various things from it. So let’s try—I’ll share this video for a moment, five minutes, I want you to watch this video for a second. So I scheduled a meeting with artificial intelligence. We sat together for a quarter of an

[Speaker G] hour and I came away with insights. Do you want me to tell you what I spoke with it—or with him—about? So, I asked him: why did Baal HaSulam have opponents? And he answers me: Baal HaSulam encountered opposition for several central reasons. One, teaching Kabbalah to the masses. He supported spreading Kabbalah to the entire nation. Many believed that Kabbalah should remain the domain of select individuals, and they also opposed translating the Zohar and making it accessible to the wider public. Two, his approach to Zionism. He supported the Zionist idea; he saw Zionism as a stage in the process of redemption, and this opinion was controversial in Haredi circles. Therefore they opposed him. Three, innovations in the method of study. He developed a new way of understanding Kabbalah; he incorporated modern ideas in his explanations; he was perceived as deviating from the traditional path. Four, social ideas. He wrote about social justice and socialism; he connected Kabbalah with social repair, and these ideas were viewed as too revolutionary. Five, his method of interpretation. He interpreted the Zohar in a new way, added philosophical explanations, and people disputed his interpretive approach. Therefore they opposed him. They really did oppose him because he advocated teaching Kabbalah to the masses, also because he was Zionist, also because he introduced a new method of study, also because he incorporated social ideas, and also because of his interpretive method. A reasoned answer, right? Amazing. I said to him: tell me, what is the source for the claim that Baal HaSulam saw Zionism as part of the process of redemption? I’ve read a bit—not much, but I’ve studied a little in Baal HaSulam’s books—and I also know people who know all of Baal HaSulam’s teaching. We’ve dealt with this question more than once and didn’t come away with anything clear. Tell me, how do you know he saw Zionism as part of the process of redemption—do you have a source? Sorry, I need to be more careful with that statement. When I examine the issue again: one, I do not have an explicit source that Baal HaSulam saw Zionism as a stage in redemption. So he says: I’m sorry, I need to be more careful with that statement. When I examine the issue again, I do not have an explicit source that Baal HaSulam saw Zionism as a stage in redemption. Two, what is known is that he wrote about national themes in his essays, that he discussed the question of Jewish nationhood, and that he referred to building the land in his writings. Three, the most famous is his essay “Arvut,” in which he discusses the desired social structure, but not in a Zionist-messianic context. Four, perhaps I confused this with his references to national and social processes more generally. I apologize for the emphatic statement in the previous answer, and I would be happy if you could enlighten me with precise sources on this matter. Fine—one who admits and abandons wrongdoing will be shown mercy, even if he is artificial. Tell me—I ask him—what made you say that Baal HaSulam interpreted the Zohar in a new way? I have to re-examine the matter. This is word for word; that’s what happened in the meeting. When I think about it, I have no clear source for comparing Baal HaSulam’s interpretation of the Zohar to earlier interpretations that would allow me to determine that his interpretation was new. What is certainly known is that Baal HaSulam wrote a comprehensive commentary on the Zohar, and he wrote a commentary on the writings of the Ari. But to say that this was a new way is a statement I cannot substantiate without precise comparative sources. I apologize for the unsupported statement, and I would be happy to learn more about the subject. Tell me—I said to him—so why did you state it in the first place if now you say you were wrong? In other words, where does that audacity come from? You are absolutely right. I erred in writing things without checking and substantiating them. This happened because: one, I relied on general impressions instead of precise sources; two, I repeated things I had heard or read without checking their truth; three, I did not exercise sufficient critical judgment. This is a serious methodological mistake, especially when dealing with words of Torah and with describing the views of the great sages of Israel. I acknowledge that you pointed this out to me. It teaches me to be more careful, to be exacting, and to base things on precise sources. Tell me—are you aware that if I hadn’t asked you, I would have left the conversation with completely false information? Yes, you are absolutely right. This is a very serious point. Many readers might have accepted the incorrect information as truth, passed it on to others, based mistaken understandings and outlooks on it, and perhaps even used it in writing or teaching. This reminds me of the severity of the prohibition, “Do not bear a false report,” and of the enormous responsibility involved in conveying information. Such an error is especially serious when it concerns words of Torah and the description of the outlooks of the great sages of Israel. The damage that can be caused is significant. I thank you for exposing the mistake and helping me understand the gravity of the matter. Tell me—I asked him—what connection is there between “Do not bear a false report” and your mistake? You are right again. There is no direct connection between “Do not bear a false report” and my mistake. This is another case in which I tried to add a quotation or a Torah concept without an accurate understanding of its meaning and context. What I did was simply the mistake of stating unsupported facts and presenting them as truth. I need to admit the mistake as it is, without trying to dress it up with irrelevant quotations. And people rely on this? But what did I learn from it? That artificial intelligence does not find it hard to admit mistakes; for us, it does. Just like that, in one second, to say, “I was wrong”? Like King David. They tell him the parable of the poor man’s lamb, they say to him, “You are the man,” and he says, “I have sinned against the Lord.” Not—immediately! That shows that for King David, his emotions were not a factor, his honor—none of that. “I have sinned against the Lord.” So there is something to learn from artificial intelligence.

[Speaker I] This is old—I remember I watched this quite a while ago, so I think…

[Rabbi Michael Abraham] Today’s models, in my view…

[Speaker I] Yes, today’s models are already less… yes.

[Rabbi Michael Abraham] No, but never mind—the idea, I think, can still teach us a few things, and I want to dwell a little on points that came up here. I wrote about this in one of my columns, which is why I took it from there. First of all, maybe at the first stage, there’s a phenomenon here that used to be very common—today less so—of errors in the answers of an artificial intelligence model. That is, an artificial intelligence model can answer you very emphatically, and it can still all be fantasy, yes? A hallucination. And this phenomenon can stem from an error in inference, but it can also stem from the fact—and we talked about this—that the model’s answers are not answers that are deterministically derived from the question. It has different options that depend on the temperature; we talked about that last time. And in practice it samples among several possible answers, and it doesn’t always even choose the highest-probability one. I mentioned this last time and said that this still improves its performance, and therefore normal models are not at temperature zero; they are at a higher temperature, which means that the answers they give are basically statistical answers. Meaning, you have a certain possibility of receiving one answer, another possibility of receiving another answer; there is a certain probability distribution among the various answers, and therefore sometimes you can get incorrect answers. And more than that, sometimes it simply invents sources—it cites sources that do not exist. Now this was, again, much more common a year ago, much less so now; there has been a dramatic improvement in this area. Again, I’m not an expert, but I know from my own use: a year ago, when I asked it questions, you could get outrageous fabrications there. I mean, it invented sources, it cited Maimonides passages that never existed; some of the things simply did not exist at all. Today that almost never happens to me. I mean, it happens very rarely. By the way, a large part of this is not just improvement in the model—I said there has of course also been improvement in the models—but beyond that, the model also uses tools, and very often before it cites something, it is told: wait, wait, are you about to cite Maimonides? Go to the source, go to the place where Maimonides appears, check whether this is an accurate quotation before you print it—yes, before you send it. And then that can prevent a great many hallucinations, as long as it has that source, as long as it has Maimonides and the source it is citing. So with today’s technology this already happens much, much, much less. But in principle it can happen. That is, there are hallucinations even today, even if they are fewer.

[Speaker D] There are things that aren’t on the web at all, so it has nowhere to get them from. Right. There are a lot of mistakes; in Torah material it has lots of mistakes.

[Rabbi Michael Abraham] I don’t know if lots—far fewer than there used to be. I haven’t found lots of mistakes.

[Speaker D] There are; it depends what you look at. I understand. It depends, depends what topic you ask about. A topic that’s open on the web, for example—then there are lots and lots. A few days ago I was looking for some book; there was some dispute regarding priesthood and corpse impurity and so on. I asked ChatGPT and so on, and it didn’t find the source. I had to go into the books and find it myself there, meaning it didn’t…

[Rabbi Michael Abraham] Obviously, things—say, if it’s connected to the internet—then the internet is a very significant source for it. That is, many of its answers are not internal knowledge that it was trained on or that exists inside it in some way, but rather some sort of access to the internet. But even access—first of all, there are errors on the internet too, and second, as you say, there’s a lot of material it may not find on the internet. There are questions for which it won’t find answers on the internet, and therefore all kinds of hallucinations can certainly occur. But again, when it cites a source, usually that source does exist somewhere, and therefore I think that if you give it the right tools to make a comparison, you actually can overcome hallucinations in citation. Maybe another point worth noting here: when we talk about internal information inside it, that too is a somewhat delicate issue. You need to understand: in an ordinary computer, when we talk about information that exists inside it, what we mean is that the information is stored in memory—obviously in magnetic form or, whatever, electrical or technological; it’s not as though there are information items sitting there, but there are marks there whose meaning is this specific information, so it is stored in memory. In a language model, the information is not stored in memory. You have no way of seeing where in this huge neural network inside there this item of information is located. It is not located anywhere. It’s not that it stores information; it does not store information. Rather, its weights became organized in such a way that if you ask it this question, it will answer you according to this information. But it’s not that somewhere it really has this specific information stored—at least not in the simple sense. It may be that in some very, very indirect way it is there in some form; that is even plausible. But it is not memory in the same sense in which we speak about the memory of a classical computer. The memory of a classical computer simply contains the information. The information that you put into it is inside it. In a language model, it’s not that the information you put into it is inside it. This information organizes the weights—excuse me—and now you have weights; you don’t have information. But the assumption is that if you ask a question, the weights are organized so as to give you a correct answer or a relevant answer. But there is absolutely no necessity that this will happen. Even information that was explicitly included in the model’s training—even with respect to that, an error in the model’s answer is possible. Not only information it never encountered; even information it did encounter. Because the information we put into it is not like taking information in a computer and putting it in memory, where the information is then stored there. If you ask a classical computer, “What does Maimonides say in such-and-such a law?”—assuming that information is inside it—there will be no mistake; it will quote Maimonides as it is inside it. In artificial intelligence, that information is not inside it in the form of information; it is somehow stored through organization. It’s very important to understand this: it is stored through the organization of all those weights there, and that organization, hopefully, does include the information in some sense. Meaning that if you ask a question touching on that information, it will somehow use it correctly. But there is no necessity for that to happen. It’s not information storage in the same sense we’re talking about in a classical computer. Besides that, as I said, there is also the temperature. Now the point—in a classical computer, can I just say for a moment?

[Speaker J] In a classical computer, both in the computer and in the network, the information is stored in the same way. It’s just that in a classical computer we know how to extract the information; in a network we just don’t know, because the way it organizes it is simply very complex.

[Rabbi Michael Abraham] The information

[Speaker J] certainly exists inside in some form.

[Rabbi Michael Abraham] No, no. The information does not exist. The weights are organized during training on the basis of that information, but it is definitely possible that the next information on which you trained the weights also disrupted the previous weights somewhat, and then you won’t even be able to extract that earlier information.

[Speaker J] Fine, okay, but if there is something such that if it answers me some answer and it came only from the network, not from the internet and not from anywhere else, then it certainly means that it is hidden somewhere inside the weights. So I—I just can’t read it in a simple way that I…

[Rabbi Michael Abraham] No, it’s not just that you can’t retrieve it. No, that description is too weak. The information is not there. What is there is something that can generate the information in response to a question, but the information is not there at all—not that you can’t extract it. On the contrary, in a certain sense it is really the opposite. Extract it—I can. After all, if I ask, I receive that information. So actually I can extract it. In an ordinary computer too, in an ordinary computer too it’s the same. The point is that the information is not there, not that I can’t extract it.

[Speaker J] No, but in an ordinary computer I claim it’s the same thing. In a classical computer too the information is not really there; there’s nothing there. There’s, I don’t know, electricity or something like that.

[Rabbi Michael Abraham] Never mind—through magnetic bits, whatever, it doesn’t matter to me. But there is a place there, a very specific place where this specific information is encoded. For me, that is called saying the information is there. And I already said that the computer has no consciousness; it knows nothing. But the information is there. In artificial intelligence the information is not there. It’s just a manner of speaking to talk about it using internal information, information inside it. That is not correct. There is no information inside it. A vast amount of information went in, but it is not stored. Rather, the information caused a change or an organization of the weight system such that, one very much hopes, if you ask the questions that require that kind of information, you’ll get the correct answer. But you won’t always get it.

[Speaker J] And therefore I don’t understand—I don’t understand what the difference is between storing numbers as network weights and storing bits inside a computer. Both are types of information that I organize in some order. In one case I know the encoding method, so I can retrieve it, and in the other I don’t.

[Rabbi Michael Abraham] The difference is, as I said before, that it has nothing at all to do with the ability to retrieve. I’ll say it again: in artificial intelligence we do know how to retrieve. We know how to retrieve the information inside it—on the contrary, that is all we know how to do. My claim is that despite the fact that we know how to retrieve it, the information is not there. The information is not there because there is no guarantee—what does it mean to say the information is there? Say in an ordinary computer, if you put into it a passage from Maimonides and you ask, “What does Maimonides say?”, you will get the answer, a quotation of that passage, okay? You know what you will get, because the information is there. You can retrieve it because the information is there. The ability to retrieve is because the information is there, not that the information is there because you can retrieve it. Now, in artificial intelligence, if you ask that question, you are not guaranteed to get the correct Maimonides, even if it was trained on that Maimonides. You are not guaranteed to get the correct Maimonides.

[Speaker J] And if you do get the correct Maimonides? And if you do get the correct Maimonides and it isn’t connected to the internet—

[Rabbi Michael Abraham] —not at all—then that means the correct Maimonides is somewhere in the network’s weights. I’m saying that if you get the correct Maimonides, even if you get the correct Maimonides, even that still doesn’t mean it, because in the next question you might not get that Maimonides that way. There is sampling there; there is statistics there. Meaning, in the end there is some sort of representation of that Maimonides there in some fashion, in a very, very indirect way. You cannot say the information is stored there. It isn’t.

[Speaker J] No, I agree in principle, I agree in principle that—so yes, in the end it is statistical, but that statistics is only at the end; it’s only in which next word I sample.

[Rabbi Michael Abraham] In principle, the distribution over words—but the model is the model. The model is this whole story taken together; that’s the model. Including the statistics, including the—everything.

[Speaker J] No, but that isn’t considered information. The way I—this is considered to me that you’re detached, that this—

[Speaker L] Rabbi, the Rabbi needs to mute—it doesn’t matter that you—one moment, one moment, mute. Constant noise. We can’t see—even me, not even—

[Rabbi Michael Abraham] One moment, one moment, what’s happening here?

[Speaker D] Shmuel said something.

[Rabbi Michael Abraham] Let’s mute everyone. Okay. Now, whoever wants to ask should ask. But in short, what I want to say is that within this whole large system, including everything, including the output, including everything, information is not really stored. That is, I think it is a mistake to see this information as stored information. And not only because of the temperature and the sampling at the end of the process. Even within the system itself, even at temperature zero, you will not always get that information. You won’t get that information because sometimes other information overwrote it, changed the weights, and you won’t get it. If by chance other information did not overwrite it, and at temperature zero every time you ask you get that answer, then you’re right—then de facto that item of information is there. But when one asks in general, when one asks in general, “Is the information that I put into the system indeed stored within it?”—the answer is unequivocally no. Unequivocally no. Ah, agreed.

[Speaker J] Okay.

[Speaker M] Rabbi, Rabbi. Yes. Rabbi, if that professor from Berkeley is looking for some model or some way to try to prove that an LLM has no consciousness, why can’t we do it simply: if we take a fresh LLM like that, which doesn’t know anything, and train it on lies that we know are lies—even logical lies, emotional lies, lies, everything the opposite of what we know and feel is true. Well? And train it a lot, and I think it will go along with us, even though we understand that this proves to us that it feels nothing and understands nothing.

[Rabbi Michael Abraham] No, why? It will go along with us because it has a different emotion from ours. That is what the people who believe in this would say.

[Speaker M] No, but we ourselves know that there is nothing behind it—clear logical contradictions.

[Rabbi Michael Abraham] Why is there nothing behind it? Why? It will tell you about all sorts of things—feelings that have nothing to do with us—that it feels tremendous love for someone who murdered it. Fine, just suppose. That doesn’t fit how we feel. But maybe it really does feel tremendous love for someone who murders it—why not?

[Speaker M] I’ll teach it that 1 plus 1

[Rabbi Michael Abraham] is 3. It’s obvious that 1 plus 1… If you teach it that 1 plus 1 is 3, that’s a different story. Then this is something else. Obviously, if you don’t train it, it doesn’t know. But a small child too, if you teach him that 1 plus 1 is 3, he will tell you that 1 plus 1 is 3. But that doesn’t mean there is no understanding or consciousness behind him. It means you didn’t teach him correctly. But that’s true of a child too, not just a computer. It doesn’t seem to me

[Speaker C] so. It seems to me a child would simply think that 3 is 2 in such a case, but he would understand that… no.

[Rabbi Michael Abraham] A child will think simply that 3 is 2. There is no principled problem with confusing children even with errors that are outright mathematical errors.

[Speaker M] Right, right.

[Speaker C] That’s true at the beginning, but

[Rabbi Michael Abraham] In the end… it’s always true; it just depends how complex the matter is.

[Speaker M] Not only children, also aduConverts, look at Flatland.

[Rabbi Michael Abraham] Truthfully, you do not even need to confuse them. From most people you will get wrong answers even without confusion—because there are thinking errors. Who is more…

[Speaker C] No, it is necessarily built into human rationality that 1 plus 1 is 2. We are talking about a scalar over the real numbers, as in, that is how we grasp it. It is not a matter of my knowing it because they taught me; it is a matter of my being able to reach that conclusion just through the concepts.

[Rabbi Michael Abraham] You can reach it, but usually people do not reach it unless someone teaches them.

[Speaker C] Okay, so that is true, so okay…

[Rabbi Michael Abraham] The learning here is not constitutive learning; it is guiding learning, or revealing learning. Okay? There is still room for learning. And if you teach someone nonsense, then at least some people will learn that nonsense and that is what they will think.

[Speaker C] What I meant was that with AI, if you train it that 1 plus 1 is 3, it will not necessarily reach the conclusion—if you give it time to think about it—that 3 is 2. Meaning, it does not really have the understanding… no, then the question is not well defined.

[Rabbi Michael Abraham] Because if you trained it that 1 plus 1 is 3, then what will it answer to the question what is 2 plus 1? Right? I mean—

[Speaker C] There is…

[Rabbi Michael Abraham] What, you have to describe to me the entire arithmetic you fed it, not just 1 plus 1. And then either you arrive at a system that has internal consistency, only it is different from our system, or you really arrive at something that is just nonsense, like someone who never learned mathematics. Like—we talked in the introductory lessons, we talked about the one-two-many system, right? The people who have 1, 2, and many. Okay? So something of that sort too. Fine? These questions are not fully defined. You need to define clearly what we are talking about, what you train it on, what you get, and then ask what the significance of the matter is.

[Speaker D] Okay, but Rabbi, what comes out of all this, from everything you just said about intelligence compared to a standard, ordinary, classical computer, is that in the answers of artificial intelligence you can often get mistakes, which does not happen at all with a computer. Right.

[Rabbi Michael Abraham] Meaning,

[Speaker D] Meaning, every question you ask artificial intelligence, if it requires thinking…

[Rabbi Michael Abraham] Always—well, not always, but usually—if something knows how to do more things, it will do them a bit less well, right? A classical computer knows how to do one thing, or every software program knows how to do one thing, okay? That one thing it does precisely. The whole leap from a computer to artificial intelligence is that with artificial intelligence, by means of one device, you do many, many tasks, solve many, many problems. The price is that you need—that is, you get a quality that is not perfect.

[Speaker D] How does it happen that artificial intelligence—because it uses the web, right? And the biggest aid tool is the internet, of course, right? So…

Depends how you open it.

Yes, so what happens is that for some reason the artificial intelligence takes some side, a side in an argument, right? Which may not be logical at all, or it brings sources that do not even… exist. No, I did not understand, what is the question?

[Rabbi Michael Abraham] I will tell you—

[Speaker D] —what happened to me personally, right? We were sitting here on the balcony with the whole family, I took a book that came out not long ago, a book on Kabbalah, and we ran an experiment, right? We used ChatGPT—no, we used Gemini—to ask it about sources. So we took some source—I will not open up the whole issue of that source here—and basically I asked it where this is written, whether it is really written. So Gemini answered me that there is no such thing, it is nonsense and superstition, and it is false. Now I open the book, I look at what is written. Meaning, artificial intelligence gives me an answer that is completely contrary to reality.

[Rabbi Michael Abraham] What do you mean contrary to reality? It did not know that book. What is the problem?

[Speaker D] No, no, that book exists online, that is the whole point.

[Rabbi Michael Abraham] Ah, ah, okay.

[Speaker D] That book exists online, that book exists on Sefaria, it exists…

[Rabbi Michael Abraham] Fine, so the question whether it is written and the question whether it is nonsense are not the same question; they are not opposites, right? It could also be…

[Speaker D] No, no, absolutely. I asked where it is written, whether it is written on a certain topic, and then it answers me—Gemini answers me—“I found where it is written on various Breslov sites and the…” Okay, fine. In short, hallucinations.

[Rabbi Michael Abraham] Okay, so that is fine.

[Speaker D] Yes, so that is it, so it writes to me: “That is not correct, it is like this, people say that, and it is not written at all, and there is no connection,” and I look at the book and it is right there before my eyes. Yes. Meaning, the artificial intelligence for some reason is not telling me the truth and not answering correctly.

[Rabbi Michael Abraham] Okay, fine, so that is an example—another example of what we talked about earlier, right?

[Speaker D] Yes, but something else follows from this. So how can I trust it in a case where I do not have a source?

[Rabbi Michael Abraham] You cannot. You can only estimate, you can only assume, but right, there is no certainty. Obviously.

[Speaker M] No, I did not understand. If, suppose, that book exists, and nine billion people apart from… what the speaker said—said there is no such book, lied, said, “There is no such book,” and there is one place where he is sitting on the balcony and seeing this book, then what is the model’s training? What will it learn? That there is no such book, even though there is, and it can freely see it. But nine billion people told it there is not.

[Rabbi Michael Abraham] No, but if it can access the internet and check that book, and the book itself exists—rather than the claim that there is such a book or the claim that there is not—but the book itself is there.

[Speaker M] Yes, but nine billion said there is not.

[Rabbi Michael Abraham] They lied. But nine billion lied.

[Speaker M] But how does it get through training?

[Rabbi Michael Abraham] But it does not…

[Speaker M] But how does it get through training? How do you train a model?

[Rabbi Michael Abraham] You examine the range of opinions.

[Speaker M] If the range of opinions overwhelmingly says…

[Rabbi Michael Abraham] Okay, so I am saying, this is what we talked about earlier. This is what we talked about earlier: that sometimes one piece of information in training can override another piece of information that was included, that entered an earlier stage of training. Right? And therefore it is really incorrect to think that when I inserted information into this network, the information is there. No, the information is not there. So this is a very important point; you have to understand it. Because when we use figurative language, we often say: “It uses internal information,” or “it uses the internet,” or “it uses external databases,” or something like that. There is no such thing as internal information. Internal information is just a manner of speaking, because we are used to the age of ordinary computers, where it supposedly means it uses the data of its network, so that is called internal information—but not that the information is actually there. The information is not there. Anyway, perhaps just one point that in recent times appears a bit less but you still hear it from time to time. You know, there were various reports also from the United States, but also here in Israel, of lawyers who submitted a statement of defense or a claim—I do not know—to the court, and cited precedents and all kinds of things there that never existed. Meaning, they simply asked the model to draft the thing for them, and they were not aware that this fellow sometimes has hallucinations, and they arrive before a judge. Now the judge himself can also swallow that line, because maybe he also has no patience to check, and in fact I am sure that happened too. But there were also judges who did decide to check the matter and did not find it, or who knew the field and knew there was no such thing, I do not know. In any case, they came to the conclusion that it was fabricated, and there were lawyers who lost their license because of such a thing. Because one really has to be very careful in using models, and the stronger they are, the more inclined we are to think we can trust them—which is also true, statistically—and still there are cases where not. Meaning, if the information is critical, you need to check it. Do not rely on the model when the information is critical. By the way, that is one of the reasons they do not let artificial intelligence—or it is difficult to let artificial intelligence—handle sensitive systems and make medical decisions or military decisions. There are rules that if artificial intelligence makes military decisions, there has to be some human in the loop, and there has to be a person in that loop who makes the decision. Why? After all, artificial intelligence is wonderful, and human beings also sometimes make mistakes, but still there is some feeling that artificial intelligence can just fantasize, and therefore we need a person there to supervise, even though a person can also make mistakes. With autonomous cars, for example, this really is a strange phenomenon—I wrote about it too—because autonomous cars apparently, already years ago, were much safer than human drivers. It has been that way for years, at least from what I have read. And even so they did not let the thing run. By now there are some places where they do, but in principle for many years they did not let the car operate because every so often it still made some mistake, had some accident. But human beings also have accidents every so often, and therefore when you check statistically what the chance of an accident is when a human drives and when artificial intelligence drives, the chance that artificial intelligence will have an accident is much lower than the chance that a human being will have an accident. But every time artificial intelligence had an accident, it made headlines all over the world, because people are frightened to place their fate in the hands of a machine—and a statistical machine that can make decisions, yes, fanciful ones. But on the other hand, from pure safety considerations, an autonomous car is much safer than a car with a human driver, statistically. But somehow we have this kind of phobia about handing our fate over to machines, and therefore this gets delayed. And therefore in these contexts too—legal and others—there is some requirement that there always be a human in the loop, always a person in the loop, meaning that the entire process should not all be done in some automatic way.

[Speaker M] But that is really an interesting question: if it really will certainly reach much, much higher safety, and we still insist on not allowing it, then it is an interesting question why. Because we need to blame someone. When a malfunction happens, we have to blame someone. We invent for him freedom of choice, free will, we do all kinds of things so that we can punish him and take revenge on him, and so on.

[Rabbi Michael Abraham] One of the explanations for this aversion really is that we want there always to be some factor that bears responsibility. But I do not see any logic in that explanation. Because the fact that I want someone to bear responsibility—this relates to my latest post—the fact that I see that if I want someone to bear responsibility, it is usually in order to improve the numbers. We assume that if he bears responsibility, he will be more careful to function better. But if the model’s numbers are better even without its bearing responsibility, then why should I care that there is no one bearing responsibility? So I want the model to do it, and no one will bear responsibility. Some want to claim that its programmer, its trainer, should bear responsibility, but in my view that is not logical. If its trainer trained a machine that truly achieves better results, but statistically sometimes it causes harm, that is not the trainer’s fault. The trainer only improved the situation. That particular error is not because of the quality of the training. If he was negligent, that is something else, but we are not talking about negligence; rather, the statistics are such that sometimes there are mistakes. I see no logic in placing the responsibility on the trainer, and I also see no logic in avoiding the use of artificial intelligence in this context. It is a phobia. In my opinion it has no justification whatsoever.

[Speaker D] But we have a lot of autopilot—in planes, in…

[Rabbi Michael Abraham] No, that is what I am saying. Today it has already entered a bit more, still in a very, very measured way. I said there is a very great reluctance regarding this matter, but yes, little by little it is entering. And again, people also do not disclose—many times, you know, in the army for example, the term drone is not acceptable; you are not allowed to say drone. You have to say—what is it called?—an aircraft with a remote pilot, something like that, what is it called? A remotely piloted aircraft. Not an unmanned aircraft. Right? Not an unmanned aerial vehicle, but a remotely piloted aircraft. Because there always has to be some human in the loop, meaning that… Now I do not think they always do that, I do not know, it seems to me not, but declaratively somehow there is some requirement that there always be a person in the loop, in the process. Okay, so that is the first point. A second point: notice the wording—and now I am going back to the video, yes?—the wording of the model, yes? In particular one has to be careful when it comes to words of Torah, because—and Torah scholars, this is very, very important. Only its fear of Heaven really fills me with trembling, yes? Just look at how much reverence it has for Torah scholars. It is really amazing. And now I also think about the fact that after all it was trained in Silicon Valley, not in Ponevezh, apparently. Meaning, I do not think there are many models that were trained in Ponevezh, so the question is then—so it also did not draw this from its training, so where did it draw it from? Meaning, where is this coming from? We talked about this—from all the Torah books that it…

[Speaker I] What? From all the Torah books… after all, it was trained on very, very many biblical books and biblical sources and…

[Rabbi Michael Abraham] Yes, so either the Torah books, or Torah literature, or the Torah context trained it that there one has to relate with sacred awe to the authors, unlike—say—physicists, journalists, or all kinds of other people; or it receives instructions; or it identifies who is speaking to it. It may be that it knows who is speaking to it, and it relates to the person speaking to it, because the context includes—the model also learns who is speaking to it, and clearly it uses… Now I, for example, use the model in contexts I am working on, yes? In structuring knowledge and so on. I have some project I am involved in. Whatever I ask it, it immediately says, “Yes, you know, and this is also relevant to structuring Torah knowledge.” Leave it alone—I asked you what was written in Yedioth Ahronoth this morning, what… leave me alone with… Meaning, it already knows me, and everything it manages to explain to me in terms of how it connects to my basic, primary interest. So it knows the person speaking to it, and therefore it adapts the speech also to the person, or the kind of answers, or the style of speaking, and so on. So there can be many influencing factors here, but that too is a point that is very important to recognize. Of course, the apologies, and the politeness, and all these things—but there it is probably not really familiarity with the speaker, but some kind of instructions it gets from the company that built it, that trained it. They give it instructions to relate politely, to apologize when it made a mistake, and so on and so on, and therefore it really… because here, if it were training on human texts, that is the last thing it would do. Meaning, people really do not admit mistakes and do not apologize for any mistake they usually made, and if it were merely imitating us, then it seems to me that is really not what you would see. So here I think this is a good example of the phenomenon I mentioned earlier too: besides the naked model, yes, with the weights and the calculations it performs, there are tools around it, there is some wrapper that it is packaged inside, which also tells it various directives and steers it in various directions, where that steering can already be deterministic, not statistical. And it gets an instruction: “Know that…” Meaning, in the end, after it calculated its calculations, it got some result, “Ah, I was mistaken.” Now there is an instruction that is deterministic, unrelated to its own calculation—or it can be deterministic—and that instruction says: be… if you made a mistake in a Torah context, apologize three times, and lavish praise on the Torah author, and how important a person he is, and that one must be very careful with every word he says as with burning coals, okay? It is simply receiving an instruction that this is what it has to do. Meaning, the models we encounter today are definitely not just a neural network. There is a lot, a lot, a lot of surrounding framework: agents, instructions, prompts, and various such environments that affect the answer we get, and that too is important to understand. Meaning, the result is not only the result of the computation of a neural network, but far beyond that. Now, one additional point—I will give you, or perhaps I will preface the second point with one more remark in the context of what I said before: these feelings of trembling and the deep apology that we hear from it—would you infer from this that there really is behind it some kind of emotion, meaning mental acts of one kind or another? I assume not, right? Meaning, although when you meet a human being who says such things, then you assume that behind the apology and the trembling there really is some kind of feeling. I, at least, am not even inclined to think that behind these very, very sensitive expressions of the model there are feelings sitting there. Therefore, the fact that it imitates human beings does not in any way mean that behind it there are also the mental dimensions that there are behind a human being. It imitates human syntax, but that does not mean that behind it there are the meanings, the semantics, and about that I have already spoken a lot in the past, and I am repeating it here. The claim of the previous source that I brought, that English post by Efron, yes? The claim of that source there was not a claim of impression resulting from imitation of human beings, but the opposite. Meaning, there is empirical data here: we trained it on something else, and it still did the step by step. So there he wanted to argue—not justifiably, but he wanted to argue—that the phenomena themselves show it. Not that the imitation of human beings causes us to think that just as behind a human being there are feelings, apparently behind the model there are feelings too. That inference, in my view, is utterly baseless. Anything is possible. It may be that it has feelings behind this. I see no basis whatsoever for assuming that. And therefore the fact that it imitates us well—we need to be careful not to let that mislead us.

[Speaker M] Rabbi, why does the Rabbi, so to speak, in quotation marks, belittle its apology so much, when the Rabbi previously accepted and even liked the apology at Sde Boker, when you asked the students?

[Rabbi Michael Abraham] Ah, I am getting to that, getting to that. Good question. So—you know what—I will already answer it here. I will explain the question perhaps. I once spoke, actually more than once, about the talk I gave at the environmental school in Sde Boker. They asked me there to speak about repentance, atonement, and the like, but not in religious contexts. So I said there that if someone who harmed me comes and feels no pangs of conscience whatsoever, but he understands that he was in the wrong, and he approaches and apologizes—

[Speaker D] —to me,

[Rabbi Michael Abraham] —feels nothing, goes on his way happy and cheerful, whistling a merry tune, and apologizes to me and says, “Listen, I hurt you, I was in the wrong, I apologize, forgive me.” Inside himself he is cheerful, happy, in high spirits, he is at a dance party inside, he feels no pangs of conscience, no sorrow, nothing. And I asked them—but he understood it—

[Speaker B] —intellectually, I asked them—

[Rabbi Michael Abraham] —there, whether they would accept such an apology. They said no, and I said that in my view that is the loftiest apology I can imagine, and that this person in fact apologized because he truly understood that he was in the wrong, and not because his stomach hurts and he wants to satisfy his stomach pains. Maybe we even spoke about this in this series, I do not remember.

[Speaker D] Anyway, yes, yes, we already spoke about it several times.

[Rabbi Michael Abraham] Okay, anyway, so Shmuel asks: then really, what is the problem? What is bad, in my eyes, about the model’s apologies? It too, admittedly, feels nothing, but on the contrary: an apology that does not come from pangs of conscience and emotion—that is the loftiest apology. So what do you say? Good question.

[Speaker I] But the model does not understand intellectually. What? The model does not understand intellectually.

[Rabbi Michael Abraham] Correct. Obviously merely moving one’s lips in apology is worthless; that is not what I was talking about. I was talking about someone who understood that he was in the wrong, made a decision to apologize—he could also have chosen not to apologize—he made a decision to apologize. There is no emotional dimension to it, that is all. But there is an intellectual dimension to it, there is judgment behind it, and therefore that is what gives it value. In the model, in my estimation—again, one can argue—but in my estimation, with the model all of that is absent. Not only does it have no feelings, it also has no decisions, no judgment, no understanding; none of that is there. It only emits words, that is all.

[Speaker M] But in the absence of feelings, then obviously that same person who asked the Rabbi for forgiveness also activated a neural network, suppose identical to what exists in the model. So what is the difference? What is consciousness? This is mysticism; you are talking complete mysticism.

[Rabbi Michael Abraham] So I am a mystic, fine, well, we are returning to the same issue—so I am a mystic. Behind it there is consciousness and judgment and a decision that acts through the neural network—so what? Fine, we already talked about this, it is not… fine, we will return to it, this is another point.

[Speaker C] Now, look, Rabbi, may I ask a question about that? Yes. I am interested—I do not know how related it is—but I am interested in the opposite question: would you accept an apology driven solely by emotion and with zero understanding behind it? If—

[Rabbi Michael Abraham] If I knew that was the case, then no. Agreed. We cannot know, but… Now, a very interesting point that ostensibly is not connected to artificial intelligence, but it seems to me that it can be connected. Let us look at Zionism, yes? One of the accusations against Baal HaSulam was that he was close to the Zionist movement. Then the fellow, Rabbi Aharon Levi, asks it: tell me, how do you know that he was close to the Zionist movement? Did it search in its sources, yes? In its internal information, or on the inter… I do not know where, and say, “Ah, sorry, indeed I was mistaken, I was too categorical, I did not find clear information on the matter of whether he was indeed close to the Zionist movement.” Now look, let me show you that this information really is there, and not anywhere else but in Wikipedia. So here, take a look.

[Speaker I] That is the AI’s preferred source.

[Rabbi Michael Abraham] Okay, soSo, that’s why I say: none other than Wikipedia. Now it’s even more astonishing. Look. His attitude toward… this is about Baal HaSulam. His attitude toward the Land of Israel and the State of Israel—so he encouraged immigration to the Land of Israel and so on. Once this became known, here, once this became known, opponents immediately arose against him from within the Haredi public, attributing Zionist motives to him. As a result, the group was disbanded. Citation needed. Meaning, there’s no source here, okay? But apparently they took it from some source. In other words, there was criticism of Baal HaSulam that he was apparently close to the Zionist movement; they accused him of Zionism, okay? Now, I assume that when the model answered him—after all, what did he ask it? He didn’t ask it: what criticism do you have of Baal HaSulam? He asked it: what was Baal HaSulam criticized for, right? Human beings who argued with him, his opponents. A completely correct answer. So what was it apologizing for? Whether Baal HaSulam was not a Zionist or was a Zionist—that’s a completely different question. It could be that he really wasn’t a Zionist—so what? But when the question was what was he criticized for, and the answer is that he was criticized for being a Zionist, that’s a completely correct answer. There were people who criticized him for being a Zionist. So what is the AI apologizing for? Now this is very strange, because Rabbi Aharon Levy asks it: tell me, how do you know he was a Zionist? Now when you ask it how it knows he was a Zionist, that’s a different question—it goes to look for sources. Yes? Now, there are no sources saying that Baal HaSulam was a Zionist, because even Wikipedia didn’t find any, right? So it answers him: no, I didn’t find such sources. And now after it didn’t find such sources, it says: well then, I apologize for the firm assertion I made earlier. Here it’s making a mistake! The earlier assertion was correct. You didn’t find sources, meaning you can’t say he was a Zionist. Fine. If you had said on your own that he was a Zionist, you would need to apologize. But if you say that those who criticized him criticized him because of his being a Zionist, then you have nothing to apologize for—you gave a good answer.

[Speaker D] Here, I just asked Gemini, and it writes the same thing, without sources.

[Rabbi Michael Abraham] What, there?

[Speaker D] That he was close to Zionism, that he supported, yes, Zionism.

[Rabbi Michael Abraham] That model also wrote that. But now ask it the question Rabbi Aharon Levy asked: wait, but how do you know he was a Zionist? I assume today’s models would already do this more intelligently, I don’t know. Again, this was years ago. But, but, but, but—no, let’s not qualify it away; this is a very important point for later, you’ll see why it matters. This is an important point. In other words, basically it gave a correct answer. It has nothing to apologize for. It gave a completely correct answer. Not because Baal HaSulam was a Zionist—he wasn’t asked whether Baal HaSulam was a Zionist. He was asked whether he was accused of Zionism, and the answer is: yes, he was accused of Zionism. A completely correct answer.

[Speaker D] Do you know what it answered me? Where? In David Ben-Gurion’s diaries. Well. It writes that Ben-Gurion documented in his personal diary his meetings with Rabbi Ashlag.

[Rabbi Michael Abraham] No, that doesn’t mean he was a Zionist. I know, it’s known that he had many meetings. Ben-Gurion wrote that he met many times with Rabbi Ashlag, and he told there that Rabbi Ashlag asked him about socialism and he asked him about Kabbalah. Meaning, yes, that’s the… But no, no, no, I haven’t seen any mention of Zionism. I don’t know exactly. But never mind, it’s not important.

[Speaker D] No, the very fact of meeting with Ben-Gurion…

[Rabbi Michael Abraham] Good—meaning that at the principled level, at the principled level, this is an important point. In other words, his answer was indeed a correct answer. The question of whether Ashlag was a Zionist—on that he found no sources, correct, because there aren’t any, let’s say for the sake of argument, or he didn’t find any, fine—but that has nothing to do with the question he answered earlier. Therefore his apology in this case is actually the mistake, not the answer. The apology was the mistake. He apologized by mistake.

[Speaker I] Can’t he say that Wikipedia is his source?

[Rabbi Michael Abraham] What? He—

[Speaker I] Can’t he say that Wikipedia is his source?

[Rabbi Michael Abraham] But Wikipedia doesn’t say he was a Zionist. Wikipedia says he was accused of Zionism. If you ask: how do you know he was a Zionist? He can’t bring Wikipedia, because Wikipedia doesn’t say he was a Zionist.

[Speaker C] It could also be that it’s trained, so to speak, to understand the pragmatics behind the sentence. That when you ask someone—this isn’t a logical necessity, that’s true—but when you ask someone what he was usually accused of, you’re talking about valid accusations, accusations with some basis, and not just random baseless accusations.

[Rabbi Michael Abraham] I don’t understand. But, but, but who said these accusations weren’t baseless?

[Speaker C] I’m not saying they were, and I’m also not saying they weren’t. I’m saying that in the pragmatic sense, at the level of language pragmatics, do you place trust in the accusers? No, I’m not saying that. I’m saying that in the pragmatics of language, when I ask someone: what were you accused of? I don’t mean just nonsense people said about him, I don’t know, when he was six years old. I mean things… What does that mean? They accused him—so what? What difference does it make? I agree, I agree that logically, I agree that logically you’re right. No, no, even pragmatically, I don’t see… How? Why, how pragmatically? People use the concept of accusation in the sense of… Yes, so what?

[Rabbi Michael Abraham] So what does it mean? It…

[Speaker C] An accusation doesn’t mean the accusation is justified. I don’t think people use it that way. But it’s like… maybe that’s how it understood it, that’s my guess.

[Rabbi Michael Abraham] Never mind. In any case, the claim is that basically there wasn’t any error here at all; the apology itself was the error. Now, if… Now, here one could discuss the error in the apology, yes? Where did the error in the apology come from? So one could say, again, that this is the result of some instruction: if someone tells you, listen, you made a mistake, you confused me, or something like that, then apologize. Okay? Don’t insist and say no, that’s not true, I’m actually right. I don’t know—that depends on the instructions it receives. So I’m just having this discussion; later I’ll use it again, but for now I’m having this discussion so that you start to get a feel—or not start, really—that… the way I did it: how do you check questions like these? What kinds of mistakes are there? Where do these mistakes come from? To what extent does this indicate its being human, and so on. Okay? So that’s basically the purpose of this little exercise, but later I’ll make more substantive use of it. Good, we’ll stop here, though, because the…

[Speaker M] Rabbi, Rabbi, Rabbi, can I ask something basic? I didn’t understand the issue of statistics and temperature. Suppose I take two completely identical models, with the same weights, they underwent exactly the same training, and at the same second I ask them the same question, exactly the same question. Won’t they give the same answer?

[Rabbi Michael Abraham] Not necessarily.

[Speaker M] How can that be?

[Rabbi Michael Abraham] The same model itself, not two models.

[Speaker M] No, at the same… no, I can’t ask the exact same model at the same second, but if I ask two identical models that underwent the same training…

[Rabbi Michael Abraham] No, ask it a second later—but, but, but it’s completely untouched, nobody has done anything with it, and after another minute ask it again. Let’s say you shut it down completely, ask it a question, close it off from any influence, and then ask it again the same question. The same model! You still won’t necessarily get the same answer. So how—what’s the explanation for that?

[Speaker M] How can that be, basically?

[Rabbi Michael Abraham] Statistics! It’s doing a random draw. I explained this in the previous lesson or the one before that. It has several possible answers, each with a certain probability. It doesn’t necessarily choose the answer with the highest probability; rather, it samples according to the probabilities. In other words, if one probability is higher, then more often it will sample that answer, or there’s a greater chance it will sample that answer, but that doesn’t mean it will always sample it.

[Speaker M] Only if the temperature is zero. And what is the temp…

[Rabbi Michael Abraham] What is temperature? Just briefly, what’s meant by it? No, it’s borrowed from physics; it’s a term that doesn’t really mean there’s actual temperature there, but in physics that’s the role it plays. At zero temperature, basically what happens is that the option with the highest probability is always what will happen. When you raise the temperature, then there’s also some chance of getting events with lower probability.

[Speaker D] AI never gives exactly the same answer to the same question. If a few minutes later you repeat it…

[Rabbi Michael Abraham] No, sometimes it’s just a different wording but it’s the same answer, but sometimes the answer itself will also be different.

[Speaker D] No, for example, yes? If you change it a little—I asked a question, I was looking up hypotheses just now, yes? Following the hypotheses that it had already found answers to, yes? I looked in biology, in the genome, what it found, yes? I asked it a question. The question: were the answers of Genome—yes—tested in laboratories and found to be correct? And the second question: were the artificial intelligence answers about Genome really all correct? Meaning, it’s the same question, I just changed the wording a bit, and it answered completely differently. Meaning, it really formulates things in relation to the question.

[Rabbi Michael Abraham] And that, that—

[Speaker D] It’s the same thing, like, it’s not… And really it gives… a few minutes later it already… it already changes…

[Rabbi Michael Abraham] When you phrase the same question in different words, that increases the chance that you’ll get a different answer. But I’m saying: even with the exact same words, you won’t necessarily get the same answer.

[Speaker D] No, this… you can try it on all of them, like, you can take all the…

[Rabbi Michael Abraham] No, these are known facts, it’s not… there’s nothing to try here, people have tried it. These are known facts. Okay.

[Speaker C] This reminds me that some time ago my lecturer told us a story that during that whole affair with lots of cheating, where people were submitting their assignments using AI, a guy submitted his paper about Leibniz as an empiricist. So they caught him immediately and failed him. And that’s kind of like the fabrications from earlier, that it… ah, okay.

[Rabbi Michael Abraham] That exists, yes.

[Speaker C] But it’s really rare.

[Rabbi Michael Abraham] But for it to say Leibniz was an empiricist—that’s rare. That’s a basic mistake, it’s not… It’s on the internet, it’s… usually there won’t be such a mistake, I think. Good. No.

[Speaker C] Maybe it was also just a very stupid student, but I don’t know to what extent, I don’t know.

[Speaker D] By the way, for some reason AI really, really loves Leibniz. In all the questions, it’s like he’s every—

[Rabbi Michael Abraham] To each his own taste.

[Speaker D] No, but there are references to Leibniz left and right. In Gemini too, in ChatGPT too. Huh?

[Speaker C] It has good taste, that’s why.

[Speaker D] Not sure that’s it. Okay, fine. Fine, any other questions?

[Rabbi Michael Abraham] Rabbi, can I ask about politics?

[Speaker C] I wanted to ask. Yes.

[Rabbi Michael Abraham] Why do I need politics now, as the song says?

[Speaker C] Fine, so, uh, fine, regarding your values—regarding which values would you say determine the… You also didn’t say whom you’re voting for in the end. I did. You said Otzma Yehudit, Otzma LeYisrael? Yes, I said.

[Rabbi Michael Abraham] Yes, there’s a column, there’s a column on the website.

[Speaker C] Ah, I didn’t see it.

[Rabbi Michael Abraham] I read the column, I didn’t see that…

[Speaker C] I said that maybe—

[Rabbi Michael Abraham] Eisenkot, but I’m considering maybe Bennett, I don’t, I don’t know.

[Speaker C] So you didn’t say, yes, okay. Like, you didn’t say… Fine, okay, that’s what—

[Rabbi Michael Abraham] I described the situation as it really is.

[Speaker C] I understand. Yes, I did read that someone almost convinced you to vote for Bennett.

[Rabbi Michael Abraham] Never mind. In the end, what matters is the bloc, not the party.

[Speaker C] Yes, but which values are pushing you to vote? Like, for me, I think security comes before the economy, and then it—

[Rabbi Michael Abraham] It’s not just a question of values. That’s a terribly simplistic description of the considerations. Because even if security, say, stood at the center of my concerns—and it doesn’t—but even if it did stand at the center of my concerns, you still have to check whether there’s a difference between the parties and who is better. In other words, it could be that all the parties would basically do the same thing, so even if security is the most important thing, it still isn’t a criterion for voting. Voting is a more complex process than deciding which values are most important.

[Speaker C] Like, it still comes out… it still comes out that it is a criterion, they’re just equal on it. So…

[Rabbi Michael Abraham] As if, if so… so that’s what I’m saying, then it’s a criterion—

[Speaker C] Not relevant to voting. So for you security is not… And that was also the first thing that jumped out at me when you said you considered voting for Ra’am. Security for you is not…

[Rabbi Michael Abraham] Correct, but I don’t think Ra’am would harm security. So there you have an example of what I told you earlier. Even if security were important in my eyes, it doesn’t seem to me that this would harm security.

[Speaker C] How do you not think that…

[Speaker M] Maybe even the opposite.

[Speaker C] Right. I’m interested in understanding why. Because to me it seems obvious that politics like that would harm security.

[Rabbi Michael Abraham] No, it seems that way to you with absolutely no basis. Why? Why do you think that the presence of Ra’am… If Ra’am are terrorists, then they’ll do terrorism whether they’re in the opposition or the coalition; it really makes no difference, they’ll send suicide bombers. They don’t need the coalition for that. What would they do in the coalition? So first of all, we saw in the change government that nothing happened. The best example someone brought me there was some Tu Bishvat tree-planting or something in the Negev, I don’t even remember the story anymore—come on. Nothing. Abbas came back and explained that he recognizes that he wants to be an Arab citizen within a Jewish state; he recognizes this state; he understands that this is the reality, although I assume he’s not happy about it. Everything is fine. And on the other hand, it would greatly, greatly improve security if Ra’am were inside, because then we would improve our treatment of the Arab population, our handling of the Arab population, which would definitely improve our security.

[Speaker I] Do you believe Ben Gvir is already no longer Kahane? What? Do you believe Ben Gvir is no longer Kahane? No, Ben Gvir is Kahane, what do you mean? Obviously he’s Kahane, what—

[Rabbi Michael Abraham] What does that mean?

[Speaker I] Okay, and Abbas—everything he said in the past, that’s no longer the same Abbas?

[Rabbi Michael Abraham] It has nothing to do with what he said.

[Speaker I] No—

[Rabbi Michael Abraham] I’m looking at what he does.

[Speaker L] But security…

[Rabbi Michael Abraham] And Ben Gvir acts like Kahane? What? Even if he’s an explicit Hamas supporter, that changes nothing for me.

[Speaker I] Ben Gvir, Ben Gvir acts like Kahane? What? Ben Gvir acts like Kahane?

[Rabbi Michael Abraham] No, he doesn’t act like Kahane.

[Speaker I] So why do you say that he is Kahane?

[Rabbi Michael Abraham] You asked me whether he’s a Kahanist; the answer is yes, he’s a Kahanist. The question is what he…

[Speaker I] Okay, so why don’t you think Abbas… so why do you think all of Abbas’s historical statements no longer reflect what he thinks today?

[Rabbi Michael Abraham] Meaning, there are people who say… No, it could be that he thinks that. I told you, you’re not listening to me. It could be that he fully supports Hamas, let’s say for the sake of argument—I don’t think so, but suppose so, okay? So what? What difference does it make? It changes nothing. How does that… The point is, what I wrote there—one of the things I wrote there—is that Ben Gvir is a thug. I simply can’t understand how anyone is capable of voting for a wild man of that sort. Really—an idiot, devoid of any achievements, and I can’t understand how people vote for him. And the only point I can see among the people who vote for him is that he gives you catharsis for feelings of frustration and revenge. And now again, I too have those feelings of frustration and revenge, and I too would like to vent them, but I don’t think that’s policy. And the same thing in reverse with Ra’am: maybe he supports Hamas and therefore I’ll hate him, for the sake of argument, okay, but when I examine the damage he’ll cause, I’m not interested in whether I hate him or not. The question is what will happen. And what will happen, in my opinion, is nothing—only good. Or not only, but mainly good. So I don’t see any… I’m not buying all the apocalyptic scenarios people are hanging on this, absolutely not.

[Speaker D] Rabbi, what… but practically speaking, practically speaking, in our country what happens is that talk is one thing and action is another. Ben Gvir, for example, talks a lot, and in practice he didn’t do anything, and on that, as an agreement…

[Speaker C] Why are you saying he didn’t do anything?

[Speaker D] The weapons, the quantity… He didn’t do anything. Quite the opposite, he did damage because—

[Rabbi Michael Abraham] I also wrote there about his three achievements in—

[Speaker C] I read, I read that part, and I also didn’t agree with you there… but, but…

[Speaker M] He was at a conference… a conference in Netanya that he held a month ago. When I got there, I registered and said I want to see the man. So I had to apologize to the guard that I didn’t have a weapon. He said to me: you’re from Netanya, but we made sure you’d be… Why don’t you have one? I said…

[Rabbi Michael Abraham] The guards there actually let in the people who don’t have weapons.

[Speaker M] Yes, no, but why… Now, when I went in Ben Gvir arrived: Well done to the police! I love you, Netanya, do you know how much I love you? Do you know what I did for you? I made sure you’d be an area eligible for weapons. Is that what Netanya was missing? A handgun for every…

[Speaker D] I didn’t understand—is a weapon a danger?

[Rabbi Michael Abraham] That’s the great achievement?

[Speaker D] No, I—

[Speaker M] I’m saying, the distribution of—

[Rabbi Michael Abraham] On weapon distribution, actually, I agree in principle. I think he did it in a somewhat uncontrolled way, and he certainly didn’t do any systematic preparatory work beforehand to check how to distribute, to whom to distribute, and oversight of the distribution. But that… Why? That’s not true. In principle I’m in favor of distributing weapons.

[Speaker C] Rabbi, there are… yes, there are criteria…

[Speaker I] There are criteria. You can say maybe they’re not strict enough, but there are criteria.

[Speaker C] They count every bullet. It’s not just that they hand over the weapon—they count how many bullets you have. You can’t even…

[Rabbi Michael Abraham] No, no, you’re completely missing the point. The question is what the oversight mechanisms are. Why should I care that they count the bullets—so what? Did you do any staff work on whom to give them to and whom not to give them to, what the consequences of giving out weapons are, where it causes harm? There are all kinds of statistics—again, I also don’t know whether to believe them—but there are all kinds of statistics on damage caused by these added weapons. Fine, there are all kinds of things. You need to do proper, orderly work. That’s the only thing Ben Gvir doesn’t do.

[Speaker D] Thank God. Wait, Lior, Lior, stop for a second. Why are you even getting to the point where you’re talking about weapons and security? Here, in the State of Israel, there is such a thing as state security. There’s an army and there’s police. The police in the State of Israel don’t operate anyway, not at all. Now what is happening today—one second…

[Rabbi Michael Abraham] Actually, that claim I don’t fully accept. There’s a common claim that this is basically privatization of the state’s duties. As if the state is supposed to provide protection, and instead of functioning—you’re in charge of the police—rather than functioning, you privatize it and let the citizens do the job instead. With that I don’t agree. Because you can’t keep enough police officers so that there will be armed officers on every corner.

[Speaker D] I’m only—not that is what I’m talking about. Let me—how shall I put it—friends, listen to a story from me personally. I’m sending—someone asked me to send tefillin, 1,200 pairs of tefillin, abroad, to Moscow for the Chabad community.

[Rabbi Michael Abraham] Okay, in another year or two Deri will already be distributing a prayer shawl and tefillin to every 13-year-old boy; he already announced it.

[Speaker D] No, wait, let me. I organize it, bring it to the airport, ship it. One box out of 10 boxes, 12 boxes, I don’t remember anymore—one box disappeared. Nineteen pairs of tefillin. Nineteen pairs of tefillin, thank God, 70,000 shekels. Gone. I go to the police there at the airport and file a complaint that it disappeared—it was simply stolen at the airport, simply stolen. It didn’t get onto the plane. Someone stole it. I go to the airport police, I file a complaint, they take the complaint from me, yes, I leave the airport, get home—I live in Gush Etzion—I get home, log into the computer, and my complaint is closed. I contact them—it was closed. Just zero…

[Rabbi Michael Abraham] You don’t need to convince us that the police don’t function, but that was true even before Ben Gvir, so—

[Speaker D] Meaning, it’s not…

[Rabbi Michael Abraham] The police don’t function, that’s well known.

[Speaker D] No, I contact Ben Gvir. I contact Ben Gvir. I send him all this, yes? After all, what am I? I’m a citizen of the State of Israel. I send him the complaint and ask him to deal with it, because what is this? What’s going on? What… It’s simply a slap in the face, yes? And… nothing. They simply don’t answer… What does Ben Gvir have to do with it?

[Speaker C] That’s not in his practical area. Like, is this a made-up story? Is this a real story or is this a story…

[Speaker D] What I’m telling you? Yes. It’s a real story. It reached court. I had a court case in Lod against the airport, against the police, and so on. Fine, it’s not… how shall I put it, one more story and one more story, and so on.

[Speaker I] But why do you expect Ben Gvir to handle every individual case? This is not an individual case!

[Speaker D] I only wanted to tell you, yes, that this dysfunction is everywhere. Here, near us, Arabs entered buildings here, stole, broke into apartments here on Rosh Hashanah, people contacted the police, the police opened a case, found the thieves, and the next day closed the case. They didn’t return—although they took all the

The things, the money, they took everything; they didn’t return it to the people.

[Speaker C] But Abraham, you’re shooting yourself in the foot, literally and figuratively. If your argument is that the government is incapable of managing security properly, and then you say Ben Gvir is an idiot because he transferred some of the management of security to the people, then that is exactly the conclusion of what you’re saying. If the state can’t establish security properly, then let’s transfer some of the responsibility to the public.

[Rabbi Michael Abraham] No, he’s not dealing with it. Meaning, if the state isn’t capable, then it should become capable; that’s its job. But when I say you can’t put a police officer on every corner, that’s not because the government is incapable, but because in reality you simply can’t. You can’t maintain that kind of manpower.

[Speaker C] That’s what I mean when I say incapable. Do you really think it’s because of corruption? It’s just that you never had a situation where…

[Rabbi Michael Abraham] No, the police don’t function even where they could function.

[Speaker C] It’s not that the police are worthless, they just don’t function. Are there countries where they function in the way you’re talking about now?

[Rabbi Michael Abraham] I don’t… no, of course not. I don’t know other countries. What I do know is that here, if something happens to you, you don’t even file a complaint with the police because it’s obvious nothing will be done about it. So that’s not… if something gets stolen from you, you have no… you’re just wasting your time. You file a police complaint only because the insurance company requires it, if your car is stolen.

[Speaker C] What interests me, Rabbi, is this: with values like yours, because you called yourself right-wing, and for me the essence of the right is mainly occupation and everything that follows from it, you don’t… you don’t agree with the principles now, say, that the overwhelming majority of right-wing parties are running on occupation, expulsion, and Judaization.

[Rabbi Michael Abraham] Okay, so what? I didn’t understand.

[Speaker C] Do you agree with… do you agree with those values?

[Rabbi Michael Abraham] I agree with… I agree partially. I think it should be done in some way that allows division or compromise with the Palestinians. What do you mean?

[Speaker C] Now Rabbi, Rabbi, Rabbi…

[Rabbi Michael Abraham] I’m against uncontrolled takeover, Smotrich-style. Rabbi, can I ask something?

[Speaker C] I said occupation, expulsion, Judaization. What does compromise have to do with it? Rabbi, can I… fine, no.

[Speaker M] Rabbi, can I ask something? Yes. What is the Rabbi’s plan? After all, in another X years, this will happen for sure: one morning we’ll wake up, and Ynet and Channel 7 and Channel 12 and Channel 14 will surprise us with the news that the number of Palestinians without civil rights under Israeli rule has reached 10 million, a Democratic Party president has risen in the United States, and what is our plan for that morning?

[Rabbi Michael Abraham] I said I’m in favor of compromise, I didn’t understand. What exactly is the question? By the way, I’m not troubled by the Palestinians’ civil rights. That is the last thing troubling me.

[Speaker M] No, to rule, Rabbi, it doesn’t bother him to rule over ten million Palestinians forever without civil rights, and that’s an apartheid state?

[Rabbi Michael Abraham] Absolutely not, absolutely not. Meaning what? I don’t see this as apartheid or anything of the sort. They are my enemies, and I have no choice. I have to be there. And if I’m there, then I rule over them. What can I do? Is there another way?

[Speaker M] No, but is that a permanent solution? That’s it?

[Rabbi Michael Abraham] And the world will allow this? It’s not a permanent solution; if you have a better solution, I’ll be the first to agree. Right now I don’t see a better solution. I said: I’m in favor of compromise. Whenever there is someone with whom it can be made.

[Speaker C] Now you sound

[Speaker M] like someone who votes for Ben Gvir. What? After all, it’s obvious that if we don’t do two states for two peoples, there will be one multinational state. That’s obvious.

[Rabbi Michael Abraham] If such a figure appears, I’ll update you.

[Speaker M] Completely obvious. No, but if

[Speaker C] Wait, but if you’re saying what you’re saying now, why not vote for Ben Gvir and Smotrich and so on?

[Rabbi Michael Abraham] Why should I vote for Ben Gvir? Good Lord, there is nothing he does, from any angle whatsoever, not right-wing and not left-wing and nothing, nothing that justifies voting for Ben Gvir. Because those are their values! The guy does nothing worth anything. Nothing! From any direction, not to the right and not to the left and nothing, simply a nobody!

[Speaker C] Rabbi, with all due respect, that sounds like “true and certain,” not true and not certain. You’re just repeating it like a dogmatic statement, as if—why do you think that…

[Rabbi Michael Abraham] It’s not a dogmatic statement! What has he done? The man just ruins everything he touches. What has he done? He’s done three things he boasts about, and even those aren’t achievements. So what? That’s it! Nothing besides that. Populism, fishing for likes at intersections, that’s the only thing he does.

[Speaker C] Fishing for likes at inters… first of all, I’m not sure to what extent they aren’t achievements, but even that, it seems to me that he does strengthen the… the…

[Rabbi Michael Abraham] He strengthens nothing!

[Rabbi Michael Abraham] Personal security just keeps declining all the time. What has he done in national security? He strengthens hatred toward us around the world!

[Speaker C] Give me one thing like that, that will show

[Rabbi Michael Abraham] me one achievement!

[Speaker M] On that scale… okay, because all the hatred in the world…

[Speaker I] On that scale, really on that scale.

[Speaker M] On that scale. Rabbi, at the beginning of this lecture he gave us, in the talk he gave us, he said openly—after all he knows he’s being filmed—he said that today Palestinian prisoners go in fat and come out thin; that is, in plain language, he starves them. To the point that he said, “My wife told me: come,” he said to me, “My wife told me, come go into the prison a little and see it.”

[Rabbi Michael Abraham] I don’t pity them at all. I’m not coming from that point at all. As far as I’m concerned…

[Speaker C] “That’s a great idea, go to prison instead of taking Ozempic.” I can, I even know stories, I even know stories—I won’t get into it—but that he kicked some of those terrorists who tried to kill him.

[Rabbi Michael Abraham] I’m saying again, that doesn’t interest me at all. I don’t pity them; let him kick them, let him shoot them… in the head, that’s not the point. The point is, as far as I’m concerned they should shoot them all in the head. Why hold them at all? Anyone who is a terrorist should be shot in the head, that’s all. I… but that’s not the issue. I’m talking about the question of what you will achieve, not what I want to happen. What will you achieve? The problem that troubles me is not the moral problems, unlike the left.

[Speaker C] Why did the percentage of… of terrorism go down? Among other things, I think…

[Rabbi Michael Abraham] Certainly not because of that. Just some… why not? Not because of that. It went down because of determined IDF activity following October 7. Those guys became very frightened, and that has absolutely nothing to do with Ben Gvir. He should not put that crown on his head.

[Speaker C] The IDF started acting only from October 7? I don’t understand what exactly you’re…

[Rabbi Michael Abraham] Of course! The determination the IDF is applying both in Judea and Samaria and in Gaza is completely different after October 7, obviously. What kind of question is that?

[Speaker M] They killed thousands.

[Speaker C] And also, with that same value you were talking about, that you’d want to put a bullet in their heads, I very much agree. Do you think that…

[Speaker M] Where—where do you have no standards?

[Rabbi Michael Abraham] You’re inventing a new morality! You’re inventing a new morality. It’s not a value, it’s not a value to put a bullet in someone’s head. I’m saying that a terrorist deserves a bullet in the head, but I’ll weigh whether it is useful. If it isn’t useful, then it isn’t a matter of values. It’s “If someone comes to kill you, rise early to kill him first.” What, it’s not that? No, it’s not “If someone comes to kill you, rise early to kill him first.” That principle is purposive. You kill him so that he won’t kill you; you don’t kill him because he deserves a bullet in the head, and that is exactly the point. You kill him so that others like him won’t come to kill you. No, that’s not true. But I don’t understand how a person over the age of sixteen can fall captive to this stuff. A person should weigh what benefit these steps bring—that’s what you need to weigh. Future deterrence, future deterrence, Rabbi, that reminds me of…

[Speaker C] It’s like… it’s such basic behavior that even an elementary school child understands that if a bully bullies him and he just stands there and doesn’t hit back, then they’ll bully him more in the future.

[Rabbi Michael Abraham] That’s true; I think only a sixteen-year-old understands that. A twenty-year-old understands it’s nonsense. Why? What do you mean nonsense? Because when you have an entire public here full of anger and frustration toward you, a strong hand will not necessarily bring better results; that is a huge mistake. I’m not saying it won’t; I’m only saying you have to weigh it, examine consequences, and not do what your gut tells you because you feel like you have to smash them and smash their mothers too. I weigh things. So you don’t think? You… I also want to smash their mothers, so what? I feel it too. You need judgment, you need examination. The death penalty for terrorists, for example, right? The flagship law, Ben Gvir’s flagship law—total nonsense. Nothing will come of it except likes. Except likes and, of course, international criticism, nothing comes of it, nothing, absolutely nothing. There is already today a possibility of imposing the death penalty and they don’t do it. After the death penalty law passes, it will go through years of legal proceedings, like in the United States, where they’re already more experienced than we are with the death penalty; years of process, and in the end it won’t be carried out. And this whole business will only cause us endless trouble, cost us a lot of money, arouse a lot of criticism, and not one more person will be killed. And even if he is killed, it won’t help in any way. But he likes celebrating with a hangman’s noose on his birthday cake, so in short, he’s an idiot.

[Speaker C] That’s… that’s populism, but I don’t think there’s nothing there in…

[Rabbi Michael Abraham] There is nothing there except populism.

[Speaker C] No, there’s more. I told you what else there is; I think you addressed it. Deterrence. Why, in your view… there’s no deterrence? The average Palestinian who wants to carry out a terror attack is, in your opinion, such an emotional creature that…

[Rabbi Michael Abraham] The average Palestinian does not want to carry out a terror attack. The Palestinian who does want to carry out a terror attack—whether he eats pita or a roll will not stop him from carrying out a terror attack. That’s not how it works. Look at the United States; there are studies on this. Does the death penalty reduce the number of murders? It does not. There are many studies on that. It does not reduce it. The United States is almost the only Western country that has the death penalty, maybe literally the only one, I don’t know. It has the most murderers in the world.

[Speaker C] But we’re not talking only about the death penalty. The death penalty is not the only deterrent.

[Rabbi Michael Abraham] It’s an example of this mode of thinking. So if the death penalty doesn’t deter, then pita will deter?

[Speaker C] No, so destroy the village, for example, or destroy the house.

[Rabbi Michael Abraham] Destroy the village, no problem. Do whatever you want; I have no problem with it. As long as you’ve done orderly, systematic work, reached conclusions, you have a policy, you know what you want to achieve—do it. Shoot them all in the head, no problem at all.

[Speaker C] So you think the left-bloc coalition with Yair Golan and so on would destroy the village?

[Rabbi Michael Abraham] They won’t destroy the village; they’ll build another

[Speaker C] new one.

[Rabbi Michael Abraham] They won’t build another new one, okay, fine, do me a favor, come on. Ah, apparently they would. I’m telling you that even Yair Golan, who is really, really not my cup of tea, there would be no security damage from his sitting in the government, even if he were prime minister.

[Speaker C] Honestly, in my view, how is that even possible? Of course there would be security damage. He would promote…

[Rabbi Michael Abraham] He would promote two states

[Speaker C] for two peoples, something like that.

[Rabbi Michael Abraham] Nothing would happen. I think… I think there’s a fairly good chance he would be tougher than Ben Gvir. What? Yes. Mr….

[Speaker C] Mr. Lior, you…

[Rabbi Michael Abraham] Mr. Lior, would you please look at Yair Golan’s service in the IDF? He was a very militant general in the IDF. Everyone who knew him knows that. No, really. Very militant in the IDF. He pushed the “neighbor procedure.” Yes.

[Speaker C] He pushed what? The neighbor procedure. The neighbor procedure. Ah, okay. Just, Rabbi…

[Rabbi Michael Abraham] I have a question. Everything I’m saying is: it’s all images. “Oh, he’s a leftist, so therefore they’ll kill us and therefore they’ll conquer us.” It’s such nonsense to go with these labels and draw conclusions from them. “Oh, Ra’am, they support Hamas.” You need to think a little, I don’t know, think with judgment, not go after these slogans that inflame people. Again, I feel the same emotions too. The same emotions. These Ben Gvir slogans bring out in me what they bring out in all his admirers. But then I stop for a moment and think: wait, one second—but what did he say and what will it bring us? I’m not talking at all about moral criticism. That is not the point at all. I have no moral problem with shooting all the terrorists in the head. On the contrary, I have a moral problem with not shooting them in the head.

[Speaker D] Actually, in government, the people who caused the most damage were actually the right and the Haredim. Oslo and Gush Katif. Rabbi, their idea…

[Speaker E] Rabbi,

[Speaker O] I have a question: what do you think about claims from the right that say that Ra’am, once inside the government, would basically not allow various military actions that could be… I’ll tell you, these hypotheses are very

[Rabbi Michael Abraham] nice, but they were already tested in reality and proved false.

[Speaker O] What actions were there in the previous government that Ra’am allowed?

[Rabbi Michael Abraham] They were partners in the change government.

[Speaker O] I understand.

[Speaker M] Which was an excellent government.

[Rabbi Michael Abraham] Excellent. A ray of light in the darkness, really.

[Speaker M] Then there wasn’t…

[Speaker O] The question is whether the security situation was the same security situation.

[Rabbi Michael Abraham] In the security situation there too, they carried out actions in Gaza, there too they arrested terrorists and things of that sort. They did everything they had to do, and no Mansour Abbas stopped

[Speaker O] them.

[Speaker D] Wait, do you think Bennett was a good prime minister?

[Rabbi Michael Abraham] My impression is yes, reasonably so. I don’t know; I’m not expert in the details. But beyond whether he was a good prime minister, I’m not talking right now about the details of what he achieved and what he didn’t achieve. The very cooperation and action among all the parties that were partners there in the coalition—in my eyes, that was simply a ray of light in the darkness.

[Speaker D] Yes, but for example, you…

[Rabbi Michael Abraham] Besides that, I also think his performance was probably quite good.

[Speaker D] Do you hear, Rabbi?

[Speaker P] But why is it good that he forms a party with them, meaning with Mansour Abbas? Why is it good that he joins up with them?

[Rabbi Michael Abraham] Because… I have no problem if he doesn’t join with them, but I also have no problem if he does. If he needs them for the coalition, or alternatively if he needs Mansour Abbas in order to deal with Arab society, because without that it will be harder to do it, then let him bring him in. What’s the problem? As I said, in the previous elections I considered voting for Mansour Abbas.

[Speaker I] Who did you end up voting for?

[Rabbi Michael Abraham] I absolutely do not identify with Hamas, and I know he does not condemn Hamas. Or no—so why would you vote for—he does condemn Hamas, but he doesn’t declare it a terrorist organization or something like that.

[Speaker I] So who did you vote for in the end?

[Speaker Q] What kind of compromise do you mean when you say, “I want to reach a compromise”? What do you mean?

[Rabbi Michael Abraham] A territorial compromise.

[Speaker Q] That we’ll live together with the Arabs?

[Rabbi Michael Abraham] No, a territorial compromise. To separate from the Arabs.

[Speaker Q] Ah, territorial. I understand.

[Rabbi Michael Abraham] Again, I

[Speaker C] don’t

[Rabbi Michael Abraham] see, right now, I don’t see any chance for such a thing, so therefore it’s hypothetical.

[Speaker C] If you… if you sound to me like on security you’re right-wing and relatively something that would be called far-right, why would you vote for Ra’am?

[Rabbi Michael Abraham] Right and left have become completely scrambled lately, so those definitions really do not interest me at all. They just don’t.

[Speaker C] I think it’s obvious what is meant here by extreme security right-wing: as you said, defense, occupation, and so on. It’s really not obvious. What, what isn’t obvious? Why would you vote, why would you vote with your set of values for Ra’am? I still don’t understand how you reached that conclusion.

[Rabbi Michael Abraham] What do you mean? I have a set of values that includes, among other things, wanting the Arab population in Israel to integrate. I want there not to be crime there, and Ra’am is an excellent tool for doing that.

[Speaker C] You think the Arab population in Israel

[Rabbi Michael Abraham] isn’t interested? I think, I think Mansour Abbas is the number one figure in the political world in Israel today.

[Speaker C] You think the Arab population

[Rabbi Michael Abraham] in Israel is a leader of historic stature. Despite all the criticism I have of the tail he drags behind him.

[Speaker C] You think the Arab population in Israel doesn’t want, doesn’t support terror, and so on?

[Speaker M] A racist statement. Twenty percent of the population, and a person stands there—forget it, I don’t care whether you think it’s racist or not—of Israeli society, and says, without knowing them, without speaking to them: nurses and doctors and department heads work with me, and you so casually say they all support the destruction of the state? That is blatant racism. When you say something like that, you are guilty of it. Almost all of us are. Almost our whole country is guilty of it.

[Rabbi Michael Abraham] No, I didn’t say all of them. Listen! I understand, I understand that you… simple racism! You would never accept that about any Jew in any other country. Statistics were done: 85% of Israeli Arabs condemn the events of October 7. Oppose and condemn. Statistics were done on this. Now, it was anonymous, not by name, so they weren’t afraid to say if they didn’t condemn it. That’s an interesting figure, I think. Again, the 15% who don’t do that is not a small number. It’s a worrying number. Fine. But 85% do. So I want those 85% to have a representative like Mansour Abbas who will participate with us, and not representatives like that other gang over there, who pull them in an anti direction. Now, please.

[Speaker C] Where did you read that statistic?

[Rabbi Michael Abraham] You can check online. It’s available to anyone who wants it; it’s not obscure. It’s a statistic you can even find through GPT. Speaking of hallucinations, yes? Find it on Google.

[Speaker C] Why, then, do you think there is identification as Palestinians within the Arab population?

[Rabbi Michael Abraham] What do you mean? Because they are Palestinians. What—what—what do you mean, why? Why do I identify with religious Jewry, even though all its leadership today is a collection of desecrators of God’s name, wicked people, and fools?

[Speaker C] “Desecrators of God’s name” is interesting.

[Rabbi Michael Abraham] Because I’m a believing Jew, what can I do?

[Speaker C] No, but that’s not… not exactly the same thing regarding a population… well, I suppose there’s no agreement here on that issue. Fine, so what’s the point here?

[Speaker K] Tell me, have you ever spoken—have you ever spoken with an Arab, Mr. Lior? Where did you get this all-encompassing racism and hatred from, where?

[Speaker C] I… I happened to

[Speaker K] I’ve happened to speak with lots of Arabs.

[Speaker C] Where are these sources from?

[Speaker K] And Rabbi, can I ask a prior question?

[Speaker C] Someone here said he wanted to comment on something. I’ve been many times at the University of Haifa. Have you spoken with a pharmacist?

[Speaker K] Have you spoken with a doctor? Have you spoken with a lawyer?

[Speaker C] I live in Israel. Where do you think I—do you think I travel to America to

[Speaker K] receive medical treatment?

[Speaker C] Brother, what’s your problem?

[Speaker K] You’re not living, you’re just talking. From the way you speak, you are not living in Israel; you live off sources of hatred and racism, and the things you’re saying are simply horrifying. And to go and give—not to present, uh…

[Speaker B] I think we’re losing the discussion. A small question? Yes.

[Speaker K] Fine. A prior question. To go and give—to go and give Ben Gvir, when there is no doubt that he was a man who committed crimes, stood trial for many, many offenses, in addition to that well-known incident with the emblem, when he said, “We’ll get to you too, Rabin,” when he was 17 or 18—so such a person becomes your minister of internal security? There was a question—and he kept Goldstein’s picture—was there a question? Someone wanted to ask? On the… on the wall of the…

[Speaker C] On the wall of his house. I think we’re done with questions for today. Did you hear me? Ah, there’s someone who asked; he lives

[Rabbi Michael Abraham] in Kiryat Arba, what do you want? So what?

[Speaker C] There’s someone who said, “Can you hear me?” I think there’s someone here who…

[Rabbi Michael Abraham] We talked over him. Yes. Michael? Who is it? Doesn’t someone want to ask something else?

[Speaker N] Yes, yes, Michael. Yes. Yes, hello and blessings. Regarding the previous class. Yes. You mentioned the Bible code. So I just wanted to comment: I have at home a book, The Bible Code, by some American journalist, Drosnin, named Michael Drosnin, yes. And there he brings two cases where it supposedly predicted the future through this: one was Rabin’s assassination and the second was some earthquake. Now, regarding Rabin’s assassination, he devotes a whole chapter to how he spoke with the prime minister’s secretary, how he tried to warn him. Later it describes that he also saw something about Netanyahu, that he would be assassinated if he went to Jordan or something like that, in the years ’96, ’98, something like that, and various other predictions that didn’t happen. So he asked Rips: wait, then why didn’t it happen? If this encodes the future, it has to happen. So Rips told him: it’s as if there are several kinds of futures. And I think I also read in the book itself, in that very book, that he wanted to connect this to the issue of free choice and divine foreknowledge, meaning that there are several futures, as the Rabbi once explained. But if so, it’s worthless. That is, if a certain future is written there, you expect it to happen.

[Rabbi Michael Abraham] If it’s written there, then I too can write several futures and one of them will probably happen from time to time. Okay. Yes.

[Speaker N] But he really believed in it; it’s really described there how he speaks with him, and Shimon Peres says to him: wait, if it’s destined to happen, then it will happen, so what do we have—how—how—how…

[Rabbi Michael Abraham] Exactly, why warn him? If it predicts the future, then why warn him? It’s going to happen anyway. Yes, that’s it.

[Speaker N] So that’s good, interesting, that’s—

[Speaker D] I think Shimon Peres used exactly that.

[Speaker N] No, but it’s nice that, so to speak, he warns him and all that. But no, no, no, I don’t make much of one case or two, and in all the other cases he says: fine, it didn’t work out. But the question is what you learn from that.

[Rabbi Michael Abraham] Like the Chabad people say, that the Rebbe really was the Messiah, only we weren’t worthy.

[Speaker N] Yes.

[Speaker D] Why was? For them he still is. Still, okay.

[Rabbi Michael Abraham] Still. Okay. Well, friends, we’ll wrap this up here. Good luck to us. Wait, wait, next week we’ll already be after—

[Speaker M] Thank you very much, thank you very much.

[Speaker D] Wait, how many lectures are there in this series? I don’t know. How many—

[Speaker M] Don’t limit the rabbi, don’t limit the rabbi.

[Speaker D] No, no, exactly, the opposite, I want there to be more.

[Rabbi Michael Abraham] No, no, I don’t have a plan in advance. Uh, okay.

[Speaker C] Next lecture we’ll ask you who you ended up voting for.

[Rabbi Michael Abraham] Yes, I don’t—there are no secrets. I, I—ask and I’ll say, it’s not—

[Speaker D] Next lecture we’ll—

[Rabbi Michael Abraham] We’ll already know. As if—Eizenkot or Bennett, as if there’s some vast, profound difference here. Fine. Yes. Well, friends, have a peaceful Sabbath, a peaceful Sabbath.

[Speaker D] Thank you very much. Peaceful Sabbath. And good results. Bye, bye.

[Rabbi Michael Abraham] Thank you very much.

[Speaker D] All the best.

Artificial Intelligence – Lesson 18 – Rabbi Michael Abraham

Table of Contents

Summary

General Overview

Full Transcript

Leave a ReplyCancel reply

Table of Contents

Summary

General Overview

Full Transcript

שתף

Leave a ReplyCancel reply