We use artificial intelligence more and more, in absolutely everything, right? You check documents with it, code, reports, letters, estimates, even medical findings. But here is the paradox: in some tasks artificial intelligence really does make us stronger, and in others it adds errors, adds false confidence, or in the end can make the result worse. Google is launching a separate AI platform for lawyers, and at the same time, in a real court case, invented facts made it into the filings after Perplexity was used.
So where does artificial intelligence already give a person an advantage, and where is it better not to trust it at all? And how do you find that line before the mistake becomes expensive? Today we are going to work that out. Hello everyone! You are on the ToTheMoon channel — technology news and insights from Silicon Valley and around the world. Ilnar, listen, I want to open up this topic and run through it. And it seems to me you have something to say here too, not least as someone who comes from science originally, right? Right now the part of artificial intelligence that deals with legal documents has sped up enormously.
And there was an announcement literally yesterday that Gemini, Google Gemini, has extended the part that concerns its platform, its corporate legal platform. That is, Gemini Enterprise for Legal. It is a corporate platform for legal work, with various industry integrations, with agents, confidentiality, data control and so on. So they are building a separate system outright, one that will let every legal question be handled. Fairly seriously, I think, right? And it seems to me, by the way, that the legal topic will require a wrapper of some kind, because it has to have, on the one hand, models that can voice their opinions, be wrong and so on, the way lawyers do. On the other hand, to have some base of data, of accuracy, right? And right alongside this news a case from Reuters comes up.
They reported that there was a federal appellate proceeding. In short, a judge's clerk used Perplexity. Since I have decided to mention Perplexity today, I will tell you about Perplexity, all right? A judge's clerk used Perplexity and included invented material in the filing, including the wrong parties to the case. And the court was considering handing the case to a different judge. And what emerged, in essence, was the absence of a human check. And right there is a curious piece of information: today almost nine out of ten fresh biomedical papers show signs of help from artificial intelligence.
It matters that this does not mean these papers, these academic papers, were written purely by artificial intelligence. It means the papers carry signs of artificial intelligence inside them. For instance, the material may have been processed by AI, perhaps some parts of it were written, perhaps some of the research was done with it. And here a very interesting aspect comes up, a question in general. On the one hand, it is entirely normal to use artificial intelligence everywhere.
It is the same as people using books before, the same as using search. Because what always surprised me, for example, is that if an academic paper cites some book, you could cite the book, but that book could also contain errors, right? And that book, in principle, has to cite something itself. It cannot cite every single fact. So there could have been an error there anyway. It is a question of checking that error. I think there is an enormous number of errors, and now checking everything is easy.
On the other hand, I have a question about the clerk in general, or about people working under a licence — these are licensed kinds of activity. How is it even possible that they do not check the cases? All right, that happened two years ago with ChatGPT, remember, there were a great many such cases. But now, if you have been handed specific ones, say, I don't know, twenty instances of such and such a case, you can certainly verify that, ask them to send you screenshots of those cases, run a separate check of the code of law, with Claude Code, with anything you like, right?
And then have a person verify those cases as well. This says something about people's illiteracy in how they use it. Not about the essence of artificial intelligence, because it looks here like a mark against AI, but about the illiteracy of the person who did it. That is how it looks to me. And I would like this story not to turn into some manipulative construction, where courts or authorities or some laws start restricting the use of artificial intelligence because of people as illiterate as this. Or, on the other hand, when ordinary people watching us, ourselves included, do not verify something somewhere that ought to be verified, right? For instance, we have some health problem and we asked artificial intelligence.
It turned out a cheap model answered us, or it simply decided to lie, told us about a case that does not exist, and we made our decision. And because of that I have destroyed myself, so to speak. So what do you think about this? In particular, and together with the legal side of it? There is a very interesting story here, I think.
Well, my view again is that the legal topic needs a platform layer of its own, independent of artificial intelligence, built on serious models such as Gemini, Anthropic or Grok and so on.
You have unlocked memories of my university years. When I was defending my thesis, a friend of mine had defended his thesis in law a couple of years earlier, he is a lawyer, that is. And he and I were discussing exactly that: theses, science in mathematics and in law. He said, yours is interesting, of course. When you write a thesis, he said, you have to obtain certain facts. That is, you derive a proof, and there cannot be opinions, there cannot be opinions. Whether you do it right, whether you think well or badly. It is a fact that can be proved, that can be arrived at.
Whereas with us lawyers, he says, it is different. We defend our opinion. How is a thesis built with us, he says? We find some topic, we form our opinion on that topic, we try to justify it, but it may be right, it may be wrong, but it is an opinion, it is not a fact, right? And that is the big difference between mathematics as a science and law as a science. Because the foundation under mathematics is much harder. There is something else I want to say here. On the applicability of artificial intelligence in law. In data science and everything to do with machine learning, AI systems and the rest, very great importance attaches to data sets, that is, to collected data that make it clear exactly how something should have ended, what went in and what is expected to come out.
On that basis models can be tuned well, so that they answer the next cases correctly. So then, assembling data sets on the legal side is naturally a laborious task, but definitely a solvable one, because there is an enormous history of cases: what went in, what came out, the testimony gathered, the facts, the evidence, everything, everything, everything you need. Or in some economic area, or in any other. And what the court finally concluded. Yes, clearly in that sense the model will be tuned to that data set, right? That is, if you train a model on one set of judges, it will try to behave exactly the same way, but one way or another the data set can be assembled.
And most likely, with careful work, you can get an AI that understands cases very well, understands which legal framework we are in, civil law or some other system of law, and on that basis takes the right decisions and so on. As for the specific case you brought up. First of all, it is a problem not of AI and not even of Perplexity in particular, which by
all appearances hallucinated and led to that result. It is unambiguously a person's poor attitude to their own duties. Does everyone do it that way? No, definitely not. But we do hear about the cases when it happens, right? That is, there are, I don't know, thousands of cases where clerks used an AI and made no mistake, or the mistake was never found. And here the mistake came to light. And, well,
It feels, you know, as if he went into Google search and took the Gemini AI answer, because he clearly used Perplexity that way, by the look of it. Right?
Quite possibly. Quite possibly, yes. I don't know how it went exactly, but it looks very much like it, doesn't it? He had a look, possibly glanced at it with one eye and decided the job was done. Cases like that are simply louder. They get pushed more, we hear about them more, and so they are the ones in plain sight. This will definitely keep happening, because with the adoption of AI, on top of work becoming easier, you end up having to work more, because you do your tasks faster, so, come on, do more work so as not to fall behind everyone else. And when there is more of it, you start cutting corners.
This is your average employee starting to use an AI. More is demanded of you. You start cutting corners. Then you checked five or six times. It seems not to make mistakes. And this time it probably doesn't either, because your deadline is burning and there is something else to be done. And off we go. In IT, which is a bit closer to me, it often happens that you make a pull request, that is, you send in your changes, and other people have to look at them. And what looks at them first is not people: they launch some agent, which gives an opinion of its own.
And if the opinion is positive, the person may already look at it with a bias, may assume everything is in order. They run their eye over it and let it through, right? And there can be mistakes there too, which lead to, I don't know, a service going down, some other problem. So it can definitely happen anywhere. And in general the preconditions are there. Again, I repeat, the way I see it is that working with AI, on the one hand, seems easier, and on the other hand the load starts to grow, the load grows, people cut corners.
And cases like this will come up more and more often. What is the cure? The cure is proper processes. There have to be processes that do not forbid the use of AI but regulate it. Checks appear, gatekeepers appear, maybe even an AI checking an AI. That can happen too. That can generally be set up quite well, so that there are fewer cases like this. But in this particular instance I am sure it is simply down to the person's inattention and carelessness, caused by something or other.
Here, by the way, I want to add that in general the question, by the way, there is detailed data on it.
I was looking it up in parallel: it was actually a judge's assistant. And a fairly serious role, apparently, right? And the story here, this story, is obviously not brand new, but it did happen within the last year or so. And what is the story here? It says that the document contained the wrong parties to the case, statements absent from the record of the proceedings, phrases that were not in the law itself and so on. So the story is like this. And it turns out that these people not only fail to fully understand what they themselves do professionally, they also fail to understand the instruments they are using. Because it is obvious, and I am putting this shade on it.
You know, the sense that all artificial intelligence is like that. Because, obviously, there are models of different levels right now, there are models that are absolutely stupid, there are models that will give you an error with a probability of ninety percent. And that is normal. They may be trained on one specific topic or for one specific case. We are constantly taking apart, in real high-end systems, how models glitch, in Fable or in 5.6 Sol, let alone all sorts of micro things.
And it seems to me that this is an approach, as if an inadequate approach. That is, why do people who deal with such matters pay no attention and not work out what these models are, not pay for them, for instance? I have the feeling that this whole Perplexity story, even though it was the free tier, looks critically, extremely strange. And it seems to me that this is how most people behave today. That is, more than ninety percent of people will behave exactly like that: well, it gave me something, so it gave me something.
I remember it was always like that before. Well, the sort of thing where I found it on Google, or I found it on Yandex. The fact that you found it there does not mean the information is correct, and yet people seem to believe that information without question. They believe it come what may. That is how it is. It is like the weather, right? A person says: I looked at the weather, it was different. Well, it always changes. If you actually look, it changes all the time. There.
It reminds me of how grandmothers always believe what they say on television.
Yes, yes, yes, yes, yes.
Every generation has its own unconditional source that it believes. For some it is the television, for some it is Instagram.
Yes.
Newspapers before, right.
Yes, newspapers too. If it was in the paper, that was that. It was set in stone, as it were.
That is a very important thing you are saying, by the way. And it seems to me this is a problem in society today, right? In the sense that when a person tells you something — and we are bringing up a great many cases right now — the good news is that it can be verified. Right now you can take any case, run it and verify it. And one thing is a different opinion somewhere, that may be fine, or a deadline shifted slightly somewhere. But there are things where a person cannot give out disinformation or make some very strange choice just for the sake of making one.
Here it looks as though the story was that people used the documents for the fun of it and wrote out some ruling for the judge. And I think there is a great deal of that in the writing of laws around the world right now. Can you imagine what is going on everywhere, in every ministry in the world, in different corporations? How much is being written that way now, whatever they may say about how you cannot use artificial intelligence in a company, or cannot use artificial intelligence when creating something. Even the very fact that nine out of ten papers, biomedical or whatever, can you imagine, this is a field where people are attentive to it and even so artificial intelligence is in there.
Yes. And that is exactly the thing I wanted to comment on as well. How is this whole story with science, with citing books and so on, actually built? Well, at least how it was built until now, but so far the story is holding. When I, let me tell it through mathematics, in other sciences it is similar, but mathematics is simply easier for me. When you prove some theorem, you write that paper and send the paper to a journal. Until it is published, it is considered not to exist. Why are journals trusted? They are trusted because journals, peer-reviewed journals, treat what they publish, treat what they publish, with great care. There are reviewers. Those reviewers work blind, that is, they do not know who the author of the paper is, they study the paper, they go through all the derivations. If they have questions or doubts, they ask them. And you, you do not know who your reviewer is either, this depersonalised letter arrives.
And so, through the editors, you begin to correspond. If the reviewers have checked the publication and, roughly speaking, put their signature to it, that they know for certain that everything here is in order, and they are most likely from the same field and understand it. And only after that is the paper published. And once it is published, you, or another person, can lean on it, knowing that this work has already been studied. You can also, if we are talking about mathematics, repeat all those manoeuvres independently.
And we did a lot of that at university, in fact. Our supervisor made us re-prove theorems, simply to be sure that all of it really is correct, and then you use it in your own work. And then there are journal rankings. That is, there are those trusted more and those trusted less. And the ones trusted more usually guard their reputation very carefully and do not let through papers they are not sure about. Things happen. There is a heap of them, if you dig, news that in the journal Nature some paper was published and then retracted, I don't know, five years later, because there were fakes in it or something else of the sort. That happens. In mathematics far more rarely.
But that is what the reputation of the publication is built on. It can be books, journals, well, more often the academic ones.
Listen, but with artificial intelligence, from the point of view of the AI model, it turns out this is an unsolvable task, because we have not seen any weights or guaranteed confirmations and it turns out we cannot see them either, because the models themselves carry a completely different architecture, right, and a completely different meaning.
Yes. But look, if I write a paper with the help of artificial intelligence and send it to a journal, that stays on the journal's side, whether to publish it or not publish it. Then that whole flywheel with the reviewers starts up. And what do we come to here? I do not know how it is on the journals' side, because for the last five years I have not been publishing, but most likely the same problem as in IT. In IT the problem now is not writing code. Any fool can generate code with the help of, I don't know, Claude Code, any other tools.
Yes.
But the problem is getting it through afterwards. Exactly the same story as the one I have just told about reviewers exists in IT. There is the pull request, that is, when you send your changes into the project, and then other employees have to read them, get to know them, be sure that what is there really is not nonsense, that it will work, that it will not break anything. And after that your changes go into the project. And then users get a new little button, some new handy feature, something else like that.
And the problem, the narrow neck, arises at the stage where there are a great many pull requests, roughly speaking papers going to the publisher, and there are not enough reviewers to check all of it carefully. And it is at that stage, if they start handing all of it to agents without some additional control on the human side, then, Sasha, you will definitely get exactly the problems that happened with the clerk who, the judge's assistant there, let some hallucinated material into the case.
All of this can break the story with scientific journals too. But I hope that will not happen. Publications will simply come out more slowly, or the journals will be thicker, or reviewers will have to be paid more so that there are more of them. But in IT it is the same problem, because an enormous amount of code is being generated. Someone has to read it, proofread it, put their signature to it, roughly speaking, that it will not break anything, and then let it all through. The load is moving from writing code, or writing academic papers, over to the side of licensing them.
And at the same time there is an enormous number of journals that do not treat their reputation that way, where you can publish faster for money. People often sin with that when you need, I don't know, to defend a thesis, you need N publications, and you can buy those publications in silly journals. Although in good academic circles that is considered bad form. Yes, if you have publications like that, people already look at you a bit askance, and your own reputation suffers. Well, that is already a bit behind the scenes.
Let me remind you that it matters a great deal if you subscribe to our channel. All of the promotion is built, of course, on your support, on likes, on comments. Do not forget to do it, we are very grateful for it. And we will continue. Listen, but this topic, it turns out, works in any field, not only in IT, it will work if, in the course of some interior design projects of Tanya's, builders start handing over documents and those documents get checked by artificial intelligence or go on further. Or in analytics, when reports are handed in, and people stop reading them or make their presentations, their conclusions, on the basis of artificial intelligence.
Someone wrote: I got a bit tired of watching your episode. I made a summary of that episode with Gemini. I wrote to that person: then why do you need YouTube at all? Go into Gemini and ask Gemini any question. Then you do not need any channels at all and you do not need YouTube itself, do you? Because watching all of it is one element, and making a summary is another. That is, on the one hand we see the problem when people stop understanding the data, in fact stop looking at the full picture and simply ask artificial intelligence to draw some conclusions. And here it is important to understand that artificial intelligence sees the conclusions, or sees the main thing, the way it considers correct.
On the other hand, I personally see an incredibly large plus in the fact that I now have the ability to double-check people, and not only in fields I understand well and even where I do understand well. More and more different reporting data is given to me: finance, marketing, sales, product analytics, infrastructure, IT people, anyone. I immediately run that data through artificial intelligence and check how far the data is consistent with itself. Whether it is the case that one cell says one thing, the second and the third another, and in the third a formula was typed in by hand incorrectly, and so on.
So I, for one, get an enormous plus out of that as help from the point of view of additional analysis. Plus I have the ability, with the help of artificial intelligence, to draw some additional conclusions that could never even have occurred to me. At the same time I can do it in fields I never understood. That is, if someone in construction sends me an estimate today, I will of course send that estimate to artificial intelligence, or he sends me some architectural plan and I ask: are there no errors here? Not in order to—
There is no problem here, Sasha.
Yes, but look, I—
Because you will automatically have more trust in artificial intelligence.
No, no, no, no, no. And that, that is the whole paradox, that it does not work that way for me.
That is with—
I will do it not in order, yes, I will do it not in order to stand against the person who, for example, brought me a medical finding, look, or in order to brought me a construction estimate, or the person who brought me a report, in some details, I don't know, a guide for plants and planting, I don't know, for landscaping, where I do not understand much, or for repairing a car, I will rather do it in order to fill out my understanding of the picture in front of me, fill it out and in some places make sure of something, get an additional opinion somewhere, have some probabilities highlighted for me. But once again, I will still work it out.
I will try to study it, I will try to look. And I, Tanya, will not of course act in the mode of saying: listen, give me your comments, and immediately sending them to my doctor or immediately sending them to my designer. That is the thing you must not do, right? Under no circumstances. Because if—
That is extremely widespread right now.
Very. I think in your profession it is super widespread. At the same time it is a normal human action, isn't it? If he spent his additional time and decided to look into it. And even if he comes to you and says: Tatyana, listen, I have had a look, and I have such and such an additional problem. And that, on the contrary, would be a help to you. It does happen that it is that kind of help: oh, right, great! And you wrote it down. But for that both sides have to be reasonable, so that the side that did it takes it calmly that the other person is expressing an opinion.
And in our society that is not always liked. Our doctors say: we are the professionals, we know. Programmers say: we are the professionals, we know. And the designer says: we are the professionals, we know. So people have to be open to conversation, on the one hand. And the person who does this in artificial intelligence has to be calm about the fact that artificial intelligence can deceive. Artificial intelligence can simply voice a different opinion, and it will not necessarily be true and real, right?
Finding this now, again, is probably about relations between people, because you clearly, Ilnar, I wanted to add that exactly the kind of stories you have just described also come up, probably, in any field, right? There, where they double-check some—
It is astonishing that the whole story Ilnar told about journals, completely unexpectedly, works exactly the same way.
Yes, yes, yes.
We send off our interiors in exactly the same way, reviewers look at them in the same way and give their opinion. They publish, they do not publish. If they published, that is it, you are the god of design. If they did not publish, that is it, forget it. But simply the fact that, you know, Sasha, it seems to me there are professions like, for instance, a designer. People come to them who already have a sense of beauty at bottom, right? Otherwise they would not have come. It is not that they need to eat or they are dying and need to be cured.
They wanted beauty, and that after all is a vague substance, and it feels as though my clients in ninety percent of cases come for confidence in decisions. They have a thousand options they like, and they need to gather all of it into one. On top of that the husband tells them something, their mother tells them, their friends tell them and so on. And then there is AI on top with all of that, you know, with its enormous volume of analysed data. And they are simply in complete terror about what to do next, you understand?
It seems to me that in my case, at this stage, it does more harm than good. Although I allow that in medicine, say, or in mathematics, where things are more, you know, precise, less abstract, it works better.
I think that on the subject of harm, basically, in general, look, if you do not adapt to this, well, if you do not keep up with the times, you are absolutely right, an incredible number of industries and people will suffer, even where they are not to blame at all. Because indeed, we have come into a new time. I am not saying that these people are bad, the ones who do not use artificial intelligence, in design, for instance. But indeed, this is the kind of time it is now, indeed the competition will be different now. If a person does not apply something and does not show something extra somewhere, then that may, well, people may start losing work because of it. Here, right now, part of the renovation is being done around me, they are painting the verandas, wooden verandas, and the people do it, well, clearly in a wooden way, technologically speaking. And if those people had shown me something extra, printed something out, brought me some description that was additionally interesting, interacted with me somehow and talked a bit more broadly, that is, adapted to me as, perhaps, a modern person, it would have been easier for them to get the next orders afterwards, right? Of course I would never in my life order any additional work from these particular people, not
under any circumstances, because it is all crude work, in the mode of: well, if you need it, we will do it for you the way we want to do it. And it seems to me that this problem now will arise in a huge number of areas, with lawyers and with accountants alike. If some lawyer of mine cannot take it calmly that in the course of my businesses I will be giving him comments from artificial intelligence, then I will stop working with him, because in modern reality that is wrong, exactly the same way as I have dismissed an enormous number of analysts, marketers, editors, various kinds of people, stopped working with them, because they go by the old methods. But you, Tanya, are right.
That basically it does damage too, because people, I see how people bring me materials made with the help of artificial intelligence. Or when you tell them something, they say: oh, we will do the same thing, we will bring the same report. We have the same thing built right here. I had a case like that the other day. I asked for some data to be sent to me, from my own analytics team, in one of the businesses, and they sent it, and then I asked them for the source data behind those particular numbers. They say to me: and what do you want to build? That is, they had built me something, it did not satisfy me.
I asked for the source data, decided to do it all myself. I came to them and said: your data is incorrect in the end. And now they say to me: well, we will build it all ourselves. Send us what you want. And I want to say to them: get out, I do not need you any more, because for years before this you did not have the brains to extend and develop the analytical reports. As soon as I developed them, you are saying: now give it to us, we will do it ourselves. No, guys, now I do not need you to do it. So how do you not lose that line, and how do you build yourself in additionally, get that, it seems to me, for people, for people to get additional opportunities with the help of artificial intelligence and not lose their clients.
That is very interesting. But I hear you very clearly, Tanya. I understand you, I understand this problem. There. It seems to me it is in every field—
We said at some point that it will soon become good manners, before sending off your work, to ask Claude, or ChatGPT, or Gemini for feedback, something like is everything all right here, is it done correctly, are there no errors here? And about three months ago you seemed sceptical about it, Sasha, and now I realise that all the accountants, analysts and the rest, builders with their estimates, before sending it to you, should simply have asked in ChatGPT, preferably the Pro version, simply asked, well, a sanity check of sorts: is everything all right, are there any errors? And it would have been much easier for them to work with you.
You know what the problem is? It is wrong too, it hallucinates too. I notice it in different fields, not only in my own. It tells you so confidently that here you have made some incorrect statement and so on. And then you realise that it is as if it really is one more person with an opinion of their own, and the truth lies somewhere at the intersection. That is, I don't know, it seems to me that this endless re-checking with artificial intelligence, first of all, wipes out all individuality, everything gets done uniformly somehow.
Even I, after you praised design to me last time, Claude Design, I think, well, what is this, why am I still not using it, if it is so wonderful? So I, two presentations, I think, I simply had to do them for clients in emergency mode. But I am actually going to agree with our subscribers who said that they all come out very much alike, these presentations. And however much I tried to tweak it, just give me back my Canva, I will do all of it quickly, at the right size, in the right position, in the right colour. Because by the time you have explained to it what you want, it simply takes a heap of time, and that personalised style is lost anyway. And it seems to me it is the same with letters. How many times now, receiving letters, you feel, even if there is no long dash there, that an AI sent it to you.
What do our subscribers think about this? Whose side are you on? Tell us, there is a very interesting side to the discussion here, because it seems to me that Ilnar meant exactly
that, he did not mean that double-checking with artificial intelligence means artificial intelligence is guaranteed to tell you the truth and so on. At the same time, if you are doing an analytical calculation, or a construction estimate, or some kind of proposal, and you have a lot of different data there, a lot of different data, there are things that are hard to double-check. No question there, probably. In terms of style, I might not ask artificial intelligence to double-check style, to double-check style in design, but, for example, in terms of dimensions, details, whether the arithmetic of the estimate matches, formulas, spelling errors, anything at all.
Why would people not double-check that? Why would people not review it? Why would people not get to the bottom of it? The question is: do people want to get to the bottom of things? What you can see now is that, on the one hand, many do not want to double-check, they say artificial intelligence is rubbish. And now you have to look, to see it here, or that it gets things badly wrong. And there are people who do nothing but use artificial intelligence in their work. They come along, they generate a document.
Recently one person, a director, sent me a whole document. I open it. I say: what is this nonsense? Ah, well, yes, I put it together in artificial intelligence. I say: do not send me documents like that, I may not read any documents from you at all.
There you go.
But look, look. This is not a question of artificial intelligence. Today artificial intelligence is basically smarter than any person. If you take it at the base level, it is still smarter in breadth. Clearly there are individual—
It is not about intelligence, it is about the quantity of data.
Look, there are, there are individual particularities, but the further question here is one of approach, how well a person can set the task properly, do certain things properly by hand or not by hand, or apply certain things. Yes, Ilnar, I asked that additionally so that our viewers write in the comments. Yes. Yes, yes.
Tanya, look, one thing is writing a letter with the help of an AI, and quite another is writing it yourself and asking it to fix the spelling and put the commas in properly. That second part. I would still like people, when sending messages, well, I don't know, to fix the typos, to put the commas in properly. But if they generate the letter, that is a very bad story. That is, you are not doing it, you are, as it were, a so-called neuro-words person, you just dump it in.
Ilnar, I have a line here. I completely agree with you that it is unpleasant to get a letter with mistakes, right? Well, especially if, well, for instance, I have, you know, this kind of division, that my first language and the language I know well is still Russian. English lags a long way behind Russian. And when a letter with mistakes arrives in Russian, for instance, I think: my God, what century are we living in, after all. And with English letters, since I have this eternal emigrant complex, I use the Pro version, Sasha, so that you do not say anything now. I give it the right prompt, that please use my style, do not make it very formal, as it were, and so on.
Just fix the grammatical errors. It still gives me some version of the letter which is no longer mine, you understand, which is—
I think, Tanya, that is how it is used, because I process hundreds of texts every day, and there is such a notion as preserving the style, or preserving ninety-nine percent of everything, or fixing only the spelling errors. You know, Apple has, even Apple has a built-in artificial intelligence for checking text. It has two different kinds of text check. The first is called Proofread and the second is called Rewrite. What is Proofread? It is when it fixes specific errors inside your words, inside those words, yes.
Or puts the commas in properly for you. And Rewrite is when it takes your text and rewrites it the way it thinks it would have been right to say it in English. And indeed, in Rewrite it can rewrite for you, first of all, as many emphases, as many different styles, meanings, ways of delivering it, details. It can also drift away from it, although now and then it does that in a fun and cool way. But once again, it is a blunt mechanism. And still, what you are saying now: if you are on a proper Pro version, or even not Pro, Tanya, simply on thinking, on extra thinking or on thinking harder, where there is deep reasoning or simply some reasoning at all, and you ask it simply to fix the spelling errors, it will fix them.
What you say about the Russian language, that is something Claude Code suffers from very badly, for instance. That is, the way Claude Code comments, if you talk to it in Russian, in English, the way it produces comments describing business logic, there is no such Russian language, right? But that is how it is built in these models. Still, the quality of the Russian, if you know how to use the model, you roughly understand what ranges it has. Well, I can tell you that a great deal, we run a great many texts through, precisely re-checking them.
We do not write from scratch. Simply, if you write from scratch, you have to understand who is writing it for you. Although even from scratch, I want to say that even from scratch this story is sometimes very cool. The question here is what you are looking for. And it is the same as with employees I come to, they write from scratch, they write complete, complete heresy and complete nonsense. And sometimes it would be better if those employees used only ChatGPT to get the final result.
The question here is one of some line, a good line. Because Ilnar put it well: if you write a letter one hundred percent with artificial intelligence, then it seems to me that is an incredible loss. But if you understand the meaning, understand the context, understand how the model is built, and artificial intelligence does it for you, then why not? I, for instance, have letters where artificial intelligence writes for me one hundred percent. I want to say, sometimes it writes to doctors, sometimes it writes to various support services, it writes even various requests that concern government bodies of some kind. I read the letter through roughly, I see that it conveys what I wanted to say.
It does not matter to me what the lexical structure is there, what words of mine are used, no. The main thing is that it knows exactly how to request the most from someone. But sometimes it overdoes it. The day before yesterday I had a case where I was sending, I had an incident, and I needed a shopping centre to preserve its video recordings. And I have no idea how to write letters like that, well, I do not understand the legality. It wrote it for me, I say to it: listen, why did you give him these extra details? It says: ah, well, yes, the extra details are not needed, I will send it without them. But look, that is already my sound position of reasoning from the point of view of some legal understanding of everything that is going on.
Few people will be able to work that out.
It generally works badly.
But on the whole it, well, I want to say, Tanya, personally I use a great many legal letters, and I really liked the way it put that one together. And the services of those centres and everything replied to me straight away, said: right, we will definitely produce all this data, we will send absolutely everything. This, this, this, this, this. Whereas I would not have been able to make a request like that, I would have had to read through the laws of California, the federal laws of the United States, read through, various codes, rights, constitutions and a heap of everything else. I would never have been able to.
You would have had to study for ten years, Sasha.
And another thing. And if I had asked some lawyer to do it, I would also have had to pay a heap of money. And in fact it all went very nicely, very well, in terms of certain data requests. And this ability in the modern world, it seems to me, to migrate, when you take the decision: do I still go to a doctor or use artificial intelligence, do I go to a designer or use artificial intelligence, go to an analyst or use artificial intelligence, go to a lawyer or—
In your place I would answer: use artificial intelligence. Under no circumstances go to a designer, if that is the sort of question you have.
Listen, no, wait, Tanya, I do go, I have different kinds of designers, you know that, right? And given my life, I do still work with designers of various kinds. And I want to say that there are still questions where I use artificial intelligence, and there are ones where I will always go to a person. And it will go on like that, because in our life, if a person simply has some one-off question once in a lifetime, in forty years they use a designer once, that is one thing. And since an enormous number of different questions come up in my life, I cannot endlessly ask a designer about every little thing.
And that is why I use artificial intelligence. In some places I take the decision myself, and in some places I will come and ask a person, since, for instance, I do not separately use people who, for instance, help with fashion, choose clothes for me, for instance, for particular parties. But sometimes I might ask people like that, if I need it for a particular place, for a particular event. I think it is exactly the cool thing for people to be able to have a broad, broad vision and to be ready to do it in different ways, right?
And not to put a stain on it by saying that these people are definitely not needed. I remember we had this topic, remember, when somewhere, maybe ten years ago or eight, people helped you with fashion, you went and did some shopping.
Yes, yes, yes.
What are those people called?
Stylists.
And I went to the shop together with my friends. Ilnar says: how much money did you spend? And he says, well, there—
Fifty thousand I paid back then, yes.
Fifty thousand, yes, about eight hundred dollars he spent on the person. And one friend of mine is standing there and says: no, you could have paid me.
Or not mine, he said there, right.
That friend of mine is fun, by the way, he understands styles, all of it. It was simply said as a joke. And in fact that is exactly the ability of a person to be broad. Clearly he was joking then. I just remembered it as a joke.
To value it.
I also, yes, that friend is also an amazing person in terms of humour, yes. It was said in a fun way, as it were, on topic. He said: Ilnar, I would have spent a few days with you for that money. There. And, but the point, yes, but the point, the point is that—
I would pay, Ilnar, to spend a couple of days with you.
Yes. But the point, but the point is precisely that a person should have an open mind and be capable in general of doing that, well, and understanding that this and this and this and this can all be good.
On the subject, on the subject of Claude Design. Tanya, we needed to make a disclaimer here when we were giving that example. You have to understand, I make presentations, well, five times a year. I really do not like doing it. It takes me a very long time. And what I end up calling a presentation is very different from the presentations you prepare for clients with all those expensive projects and so on, right? That is, for me it is simply a side activity that I do not enjoy but that has to be done.
And that level Claude Design covers wonderfully. Most likely, on seeing the slides I end up handing over, you would wince and say: what is this nonsense? Who does it like that? Right? And on the other hand, the people who end up looking at it, they do not much like presentations either, but for them it is beautiful, and it is done fast enough. That is, the AI raises the lower level well, it does not pull you up to the top, but those who are weak in some part, it pulls up a great deal.
Good programmers still understand code better than the AI does, even now.
Mhm.
Weak programmers raise their own floor a little higher. And the same thing is happening in design, yes. So when I say that Claude Design is good, I am saying that it raised the lower level well and took off me that part of the work which I do not enjoy. As for high art, as it were, for good high-level presentations and so on, I can say nothing here. And from you I hear that it is not ready for that yet, and I readily believe it. But some of the lower levels it already covers quite well.
Knocking something together super fast, then probably yes. But since for me this is the most important instrument in my work, one of the main ones, in order to establish contact with a client and speak the same language, Claude Design is of course no help to me here.
And I want to say from my side once again, here again a disclaimer, yes, that, simply, Claude Design is after all quite a serious system.
It has been built into Claude, into Claude Code now, by the way. And, well, clearly it is built in not quite, not quite architecturally yet, but I want to say, Tanya, exactly that I make a great many presentations and all sorts of selling PDFs, sometimes pages, micro, micro things of some kind. Indeed, it is not a system that, as you say, easily lays out footers and headers, and easily, and easily holds the page structure sometimes. But for me Claude Design is precisely the chance to design different things even more widely.
I hate Figma, Sketch, all ready-made templates in PowerPoint, in Word, yes. All my life I have worked, including living through an enormous number of these template stories. I do not like CMS systems with ready-made templates and all the rest. And Claude Design gave me for the first time the chance, on the one hand, to get templating where it is needed, and on the other hand to go off into incredible width. Yesterday I am driving in my car and thinking, why in my interface, I have this long screen, huge, very big, right, well in the car, in an Escalade, a modern one.
And I, I think, why on this big map here do they not use certain free zones to display micro-features that are very useful to the driver. And how cool it would be if in design they did that. Before, look, that was very hard to do. Now a designer has the chance to take this piece of the screen and say: listen, offer me forty variations on how I could display, for instance, tyre pressure in a micro zone on the Google map. And how could I display this? And how could I do that? You now have possibilities that help professionals, and I am quite a professional person in terms of UX/UI for platforms and interfaces, to do something extra.
Before, I would have had to say to a designer: listen, tell me what you can offer. And I would have been limited by that person. I, for instance, see that Claude Design now gives me so many unreal alternatives. But, Tanya, look, in interfaces or in a web presentation, or in websites, for instance, it will not give me that in interior design. I can say that, for instance, where I do not use image generation at all today, or, well, generation of various images in terms of redoing things, it is in terms of interiors. I see that the system, or, for instance, in terms of a person's appearance, their clothes.
It is too unreal. Right now an enormous, by the way, someone write in, whoever uses this, because I am sure it suits a lot of people. For many people, picking the right style, for instance, on themselves or for someone, or paying an extra thirty dollars or fifty dollars, or a hundred. Or, as Ilnar just said, eight hundred dollars he paid to a person who helped him buy things in a shop. He will say: you are ill, if I buy eight hundred dollars' worth of things once a year, for instance.
The question here is that for many this is opportunities and new instruments. But indeed, for me, for instance, so that I light it up for it, well, Tanya, you know, right, you— Tanya, when she comes to visit me, I say all the time: Tanya, listen, can you help me here? How can I place, I don't know, two sofas here? It seems a stupid question. You know, it seems a dumb question, and in fact I, I understand that it is a very hard question, and I need a person who instantly clicks and produces it. And I trust that person.
So far artificial intelligence, for instance, Tanya, solves such questions for me at zero. Indeed, that is a fact. And I, for instance, will not trust it, even when after all I have here a Chinese man, a Chinese man together with a Mexican man, decided to paint for me together. I have no prejudice about nationality. But all of it looks, but all of it looks simply like very, very hard communication. It turned out that way. That was the system of contractors, that is how it turned out. I can still say this: I now regret that I did not ask artificial intelligence about choosing a particular type of paint together with it. But at the same time I understand that artificial intelligence could have got it very badly wrong.
Although, remember, there was this case where my friend was painting my garage inside, and he chose the paint entirely with artificial intelligence. Tanya, I will note, when you arrived, you even said: I did not even expect it to look that good.
The right paint, yes.
And that good. Remember, you even—
The one I always use.
You say: I did not expect at all that it would be that cool. Because it really does look damn cool! Well, as though even cooler than you expected.
He only chose the paint, he did all the rest of the work.
But there is a nuance. Yes, but there is a nuance. This was still done by that person, who, let us say, who, who on the whole has a feel inside, has vision, views, approaches, designs, everything. And it was still that person. And another person, whatever he did, however he drew it, he would not have done it like that. Well, because that is exactly why you said it just now, right: it was still this person who did it, wasn't it? And this is where we will end our episode. Although there is a very interesting topic, we will see when we take it apart further in terms of—
I wonder whether Ilnar wore that clothing he bought for—
Very much so.
Yes. And afterwards everyone said it was very good that I chose it. Yes.
Yes, I liked it very much, by the way.
The moral of this fable is:
Further on there—
To designers and so on.
Further on, the further question is precisely to keep it for yourself, to file it away in your head, that it turns out this works. And it is cool to have a whole spectrum, when you choose yourself, when you choose with the help of a newspaper, a magazine, artificial intelligence, a girlfriend, a friend, your mother, your father, guests or designers, fashion designers, all the rest. That is, to allow yourself the idea that beauty, well, it can be varied.
Of course.
I think that was a big plus. He was exactly the one who wore it, and exactly the one who did it, and was exactly in it, in this—
And in this case I vindicated you.
Yes, one hundred percent. The ToTheMoon channel, technology news from Silicon Valley. See you in exactly a week.