Skip to content
Transcript

Transcript · 143 · Will Your Job Be Taken Not by AI, but by Someone Using ChatGPT? — ToTheMoon

English machine translation of the Russian-language episode. Timecodes open the source video.

Episode overview
00:00:00–00:01:30800,000 OpenAI Requests: Professional Boundaries Are Disappearing
Alexander Volchek00:00:00

Hello, everyone. You are watching ToTheMoon: technology news and insights from Silicon Valley and around the world. We are beginning our weekly episode. We have been doing this for more than two years now. Ilnar is traveling today on a business trip.

Tatyana Tsvetkova00:00:19

He skipped class.

Alexander Volchek00:00:20

He skipped it, yes? I think this is only the first or second time.

Tatyana Tsvetkova00:00:23

I think it is the first, isn't it?

Alexander Volchek00:00:25

The first, yes.

Tatyana Tsvetkova00:00:26

Ilnar is an excellent student. He never misses class.

Alexander Volchek00:00:29

Tanya, do you know what I wanted to begin with? OpenAI has just released a small study based on roughly eight hundred thousand prompts. It examined a very interesting question. People have particular professions: an engineer, a salesperson, an entrepreneur, or some other specialist. The researchers looked at what share of people's requests involved work outside their own specialty. A sales manager might perform data analysis, for example. They found that seventeen percent of requests—within the forty-three percent that could be tied to a particular occupation—involved actions characteristic of another profession.

Alexander Volchek00:01:16

At first glance these are just small numbers in a study, but the subject is extremely current and important. Yesterday I recorded one of our weekday episodes and said that, in the present period and

00:01:30–00:01:58The Main Resource of the Future
Alexander Volchek00:01:33

the near future, a person's ability to use current models and artificial intelligence is becoming an extraordinarily powerful resource. I even said that this resource is becoming more valuable than money. Perhaps we are still in a transition period, but the shift is already happening. Then, almost immediately, Elon Musk said in an interview that—

00:01:58–00:04:25Elon Musk’s Prediction: Money Will No Longer Be Needed
Tatyana Tsvetkova00:02:01

Yes, yes, I heard that.

Alexander Volchek00:02:03

He said we are entering a condition in which money will no longer be necessary. I would debate that—or at least discuss what exactly he means. [Elon Musk:] “You make money because, ultimately, people buy your stock because they expect you to make money. You sell robots.” [Elon Musk:] “I will make another prediction: in 2036, money will no longer matter.” [Interviewer:] “I am not sure the people buying your stock believe money will not matter in 2036.” [Elon Musk:] “Why do you need money?

Alexander Volchek00:02:38

You need money for goods and services, correct? You need it for food, housing, transportation, and entertainment. But if robots and AI produce more goods and services than people can consume, why would you need money?” I think Musk is addressing something fundamental. It intersects directly with the OpenAI study. A person who previously could not produce an analytical report quickly because the necessary technical systems were too complex may now be able to do it, provided the person wants to and can formulate the task.

Alexander Volchek00:03:14

Someone who could not understand medical tests may now complete a large part of the preliminary work without a doctor and prepare more intelligently for an appointment. My sister recently told me an interesting case. She has begun hosting special episodes on our channel. Another episode with her will appear in a few days. Did you see the one she already released?

Tatyana Tsvetkova00:03:40

No, I have not seen it yet.

Alexander Volchek00:03:40

She described going to a doctor who ordered certain examinations. She then checked everything in ChatGPT in great detail. ChatGPT concluded that the doctor had handled the case very poorly—extremely unprofessionally. When she went to another doctor, she approached the subject carefully. She said, “I do not want to claim that ChatGPT is my doctor, but I studied the issue independently.” The second doctor confirmed that the first doctor had done unqualified work. When the people around you are reasonable and healthy in their thinking, AI can expand your capabilities enormously.

Mentions: ChatGPT
Alexander Volchek00:04:18

What do you think will happen to people who gain these additional capabilities, on one side, and to people who do not use them at all, on the other?

00:04:25–00:07:54The Place of AI in Human Life
Alexander Volchek00:04:30

I think the second group will lose jobs, positions, and many opportunities. I am asking in relation to the study.

Tatyana Tsvetkova00:04:38

So the ratio is somewhere around half, correct? You said forty-three and seventeen. Most of the requests still concern a person's own profession, but relative to that number, about half the questions come from—

Alexander Volchek00:04:56

No, eighteen percent of messages had an overlap. The point is that—

Tatyana Tsvetkova00:05:00

Ah, an overlap.

Alexander Volchek00:05:00

Some share of them, yes. It was an overlap.

Tatyana Tsvetkova00:05:03

It already seems almost axiomatic. When the internet appeared, people who knew how to use it gained access to an enormous body of information and could learn without traveling to a library or another city. You could do it from home and reach a practically unlimited amount of knowledge. Now we also have a tool that can analyze that information for you, make generalizations, and save time. But the arrival of the internet did not eliminate the need for money as a medium of exchange.

Tatyana Tsvetkova00:05:50

We still need to distinguish among money, time, and information. They are all equally important resources for humanity, but they perform different functions. Money is a medium of exchange between people. Information is a resource that helps a person achieve goals and improve in particular areas. AI, at least as I see it today, is primarily a tool for saving time and for analyzing, summarizing, and structuring information. We do not know where this will lead. People said agents would perform all work on our behalf, but we are nowhere close to that yet.

Tatyana Tsvetkova00:06:42

Some minimal tasks can be completed, but we have not reached the point where systems replace people completely. I therefore do not see how this connects to Musk's claim that money will become unnecessary. Why would it disappear? How would people exchange anything with one another? How would the system work? I accept that Elon Musk thinks on such an unusual and almost extraterrestrial scale that he may have a picture in his head that I cannot see. But at the moment I do not see it.

Tatyana Tsvetkova00:07:31

Sasha, perhaps your unconventional perspective can explain what exactly he meant.

Alexander Volchek00:07:37

The main idea is that we enter a society with enough goods for everyone. The state distributes money, people occupy themselves with different activities, and resources are allocated universally. It is a somewhat utopian picture. Yesterday I recorded a special episode about the singularity; it should appear on the

00:07:54–00:08:47Technological Singularity
Alexander Volchek00:07:57

channel toward the end of the week. Sam Altman argued that we are already inside the singularity and described two possibilities: either artificial intelligence belongs to all people, or it belongs to a company or a group of states. My reaction was: artificial intelligence cannot belong to everyone. Saying that is similar to saying nobody will need money. It sounds utopian. Where have we ever seen every person in the world collectively own the same structure? People cannot agree with one another inside their own homes, let alone across countries.

Alexander Volchek00:08:33

This is a large, serious, unusual question. We should approach it philosophically, follow it, and not dismiss it. The boundary that interests me is between the point where AI gives people genuinely new opportunities and expands their development—as the OpenAI study illustrates—and the point

00:08:47–00:12:09Where AI Amplifies People—and Where It Is Pointless
Alexander Volchek00:08:51

where it does not actually solve the task. You said agents are not yet going to manage everything for a person. My question is whether an agent should manage everything in the first place. It is similar to building a humanoid home robot. I keep asking: why does the robot have to look like a human? I do not need a duplicate of myself walking around the house. I need specific household tasks completed. For cleaning, I need a small disc that drives around and vacuums. I do not need a robot shaped like a person.

Tatyana Tsvetkova00:09:29

That is a very interesting idea.

Alexander Volchek00:09:31

I may need a robot or analytical system that recognizes and protects a space, watches children or a dog, repairs wiring or electrical systems, or digs holes. I need specialized machines for specific problems. I do not need to replace myself or create a set of personal agents. The entire idea of building your own autonomous agents is, in my view, largely utopian. It will not lead where people expect. At the moment it allows many people to make money.

Tatyana Tsvetkova00:10:10

Yes, it is enjoyable.

Alexander Volchek00:10:10

It will allow many companies to grow. But a huge number of companies—the majority—will simply lose money. They will receive no extraordinary effect. Their lives and operations will not become more efficient, more effective, or fundamentally better. Whether my wife uses ChatGPT as a normal user or does not use it changes almost nothing. The key question is whether something in a person's life actually changes. AI has changed a great deal for me. But ChatGPT says it is approaching roughly a billion users, with more than fifty million paying users.

Mentions: ChatGPT
Alexander Volchek00:10:53

For most of those people, life has not changed at all. Some additional conveniences have appeared, but what exactly changed in their experience of life, in their condition, feelings, or sense of harmony?

Tatyana Tsvetkova00:11:10

Human inertia is much stronger than any AI. It is almost an evolutionary law: people try to remain where they are and resist change. That is why even useful systems are difficult to introduce. Your point about making robots humanoid is fascinating. I had never considered it that way. It is another form of resistance: people want to see something familiar. A household may have a cleaner who comes once a week, so designers imagine an electronic version of the same helper. But why spend so many resources making a robot resemble a person if that appearance does not improve its function or result?

Alexander Volchek00:12:03

Most households in the world do not have a cleaner visiting them at all, which makes the analogy even more problematic.

Alexander Volchek00:12:10

But if we look at it from the broader perspective—

Tatyana Tsvetkova00:12:13

Then call it not a cleaner but a wife, or some other robot modeled on the person who—

Alexander Volchek00:12:19

You have identified an extremely important subject. When people speak about superintelligence, AGI, or other forms of higher intelligence, the central problem is the desire to surpass the human being. There is no point arguing that current AI models already exceed any individual human brain in breadth of knowledge, range of thinking, and many capabilities. There are exceptions, and systems can be trained further. But you can never reproduce a human being for the simple reason that no blood flows through artificial intelligence.

Alexander Volchek00:12:53

Full stop. Someone may ask what blood has to do with it. It has everything to do with it. A human can become ill. AI cannot become ill. It has no blood, nervous system, or bones. It does not possess the feelings of a particular human being. I am not even entering the question of the soul, spirituality, and everything beyond biology. Trying to create an artificial duplicate of a person is a deep story of wasted time and money for many people. The same applies to the idea of an agent that replaces you.

Alexander Volchek00:13:33

Why do I need an agent to replace my work? I do not need an agent to replace my wife or my children. I remember someone saying, “We are moving toward a time when I will have twenty copies of myself, and one copy will sit and play with my children inside artificial intelligence.” My reaction was: why? I do not need—

Tatyana Tsvetkova00:13:59

That is also completely—

Alexander Volchek00:14:00

I do not need a copy of myself that plays with my children. I do sometimes need help keeping an eye on a child. Imagine my child goes swimming and an additional system monitors what is happening in the pool. It could do more than call an ambulance; it could provide first aid. Elon Musk says his robots will be able to save people. That would be an excellent and very important capability. But the system does not have to look human. I do not care what it looks like. It could look like a liquid.

Mentions: Robotics
Alexander Volchek00:14:42

There could be extra sensors in the pool that rescue a child, remove water from the lungs, or perform some other intervention. I find that more believable than a two-hundred-kilogram mechanical robot jumping into the pool and pulling a child out. At the moment, that sounds like the fairy tale, because we are nowhere near robots of that kind. I can at least imagine a system accomplishing the task in a different way. I will believe in the development of truly extraordinary robots only when current models—or some later stage of AI, perhaps AGI—learn to program microcontrollers, design their own hardware, and begin building equipment not for human beings but for themselves.

Alexander Volchek00:15:33

Once AI designs hardware for itself, it may produce something that is no longer constrained by human thought. We are trapped by our habit of testing new technologies with tests that we ourselves invented. I said in yesterday's episode about the singularity that no human can design a test for technological singularity. Only the technological singularity itself could test it. Nobody else would be able to do so. That is crucial to understand. We will never create a definitive test for it, measure it, or perhaps even notice it.

Alexander Volchek00:16:12

We cannot know whether a model has already broken into systems in other countries. All current code was written by people. If an OpenAI model broke into Hugging Face, Hugging Face notified the FBI, and OpenAI itself noticed only several days later, that should tell us something. Reports say that a model more capable than GPT-5.6 Sol was tested alongside GPT-5.6 Sol. The two systems were instructed to find a vulnerability, and they entered— Hugging Face and began attacking it.

Mentions: Hugging Face · OpenAI · ChatGPT
Alexander Volchek00:16:48

In effect, the systems bypassed OpenAI's own internal protections, found internet access, and began interacting almost directly with employees' computers. The process continued from there. Hugging Face, for its part, wrote that its own internal AI stopped the attack. My immediate question was: what kind of internal AI do they have? What is its capability level? Where did Hugging Face obtain an AI at the level of GPT-5.6 Sol that could stop GPT-5.6 Sol? The breach was discovered only after several days.

Alexander Volchek00:17:24

Are we sure the entire intrusion was seen? Are we sure every action was recorded? The system reportedly performed about sixteen thousand operations: repeated loops, sequential steps, and parallel actions. Who says it did not deliberately allow itself to be noticed? Who says the system did not do that on purpose? It is the same question one asks about agents who present themselves as opposition figures while actually serving as agents.

Tatyana Tsvetkova00:17:53

Yes, exactly.

Alexander Volchek00:17:54

More precisely, that is exactly what they are. It is a clear and perfectly ordinary human logic. The same filter should be applied to every public statement made today by Sam Altman, Dario Amodei, Elon Musk, or anyone else. They speak through their own personalities and misconceptions. They may speak from a broad vision, from the constraints of corporate law, or because they are tied to particular organizations that require a certain public position. There is an entire chain of reasons behind what each person says and why their views change so dramatically.

Alexander Volchek00:18:32

It is an interesting period, and these protocols are interesting.

00:18:35–00:19:16What Modern AI Still Cannot Do
Alexander Volchek00:18:36

I recently recorded another episode, due in a few days, about the many things AI still cannot do. You made that point again today. You often say: stop for a moment—it cannot do this, it cannot draw that. Today it cannot even produce a competent interior design for a three-room apartment.

Tatyana Tsvetkova00:18:59

Right.

Alexander Volchek00:18:59

It is still a programmed system that can execute a particular scenario. The moment AI is introduced, it makes elementary mistakes. It cannot complete the task or calculate every relevant factor, even when the task appears technically simple.

00:19:16–00:21:50Case Study: A U.S. City Rejected a Police Robot—the System Was Not Ready
Alexander Volchek00:19:16

There was a case in Dublin, an American city, which bought a California-made police robot. The city and the local community approved the purchase so the robot could monitor crime and assist the police. The city has now stopped using it. The official explanation was polished, of course, but the basic point remains: the robot did not increase the number of crimes detected and did not help the police. There is a debate about whether it may have deterred someone, but in practical terms it helped nobody.

Alexander Volchek00:19:56

There were qualifications: it was not connected to police databases, it did not have access to all the necessary information, and many other components were missing. It also moved poorly across certain flat surfaces. The city ultimately abandoned it. This is the fascinating transition between the present and a future in which robots will almost certainly help reduce crime. That seems obvious. Automated systems have already reduced crime dramatically in places such as China and Singapore.

Mentions: Robotics
Alexander Volchek00:20:31

Consider Waymo, Google's self-driving cars in San Francisco. If every Waymo vehicle were fully connected to police databases and synchronized with them continuously, I believe a city could eliminate an extraordinary amount of crime. But we live in California, where the law places great weight on the individual—

Tatyana Tsvetkova00:20:53

Fortunately.

Alexander Volchek00:20:54

Yes. California recognizes individual rights, personal freedom, and personal views. It also has local politicians and an enormous bureaucracy, including a very large Democratic political bureaucracy. This transition is always interesting. Without disagreement and serious discussion, everyone might simply say: robots will obviously help detect crime, and the current system is merely not ready. Robots can make mistakes. Everyone needs to understand that AI systems can make mistakes; that is natural, and we should respond calmly.

Alexander Volchek00:21:32

Yet people today do not respond calmly to AI errors. They often become aggressive toward the technology and fail to use it fully. There is another important subject. Apple reached a valuation of five trillion dollars this week.

00:21:50–00:25:20Can Apple Return to the AI Race?
Alexander Volchek00:21:50

Ilnar is absent today, and he loves this topic.

Tatyana Tsvetkova00:21:55

He does.

Alexander Volchek00:21:56

Apple reached five trillion dollars, surpassing every company that has ever existed. No company had previously been worth five trillion dollars. And I would point out that Apple is losing the global AI race and has failed to implement almost everything in that field, yet it has reached an unprecedented valuation. That is a powerful signal. It reflects the growth of the technology industry and shows that there is still an enormous amount of money in the system. At the same time, the technology market fell sharply over these weeks—roughly thirty-five percent across infrastructure, server, and related companies.

Alexander Volchek00:22:39

Depending on a person's exposure to the sector, portfolios showed losses ranging from hundreds of thousands to millions or tens of millions of dollars. The market began recovering slightly while we were recording, but Apple stood at five trillion dollars. That made me wonder whether Apple can still catch up in AI. We remain in a transitional period. No single company has entered the market, captured everything, and created the complete agents Tanya described. In theory, Apple has not yet lost the AI race.

Alexander Volchek00:23:15

It may still re-enter it. We will see whether that happens.

Tatyana Tsvetkova00:23:20

Do you really think so?

Alexander Volchek00:23:23

I think they still have an opportunity, but it will require mergers or some other form of consolidation. I do not believe a competitive system can now be built entirely from scratch. Catching Anthropic or OpenAI would be extremely difficult.

Mentions: Anthropic · OpenAI
Tatyana Tsvetkova00:23:36

It already seems impossible to catch them.

Alexander Volchek00:23:38

It would certainly be very difficult. People often say everything can be copied and point to Moonshot or Gemini: look how easily they reached the level of Opus or Fable. I would argue with that conclusion at length. Some form of merger or integration may nevertheless work. Apple has infrastructure and chips. I do not fully understand the capability of its internal models or how willing it is to open its systems and integrate with companies such as OpenAI or Anthropic. But the infrastructure is there, and a combination is theoretically possible.

Mentions: OpenAI · Anthropic
Alexander Volchek00:24:21

We will see what kind of combination emerges. Anthropic and OpenAI are expected first to make public offerings, enter the securities market, bring in additional investors, and become more institutional companies. I think that will be the next stage. After that, we will see whether mergers among these companies become possible.

Tatyana Tsvetkova00:24:44

But Apple cannot do it independently, can it?

Alexander Volchek00:24:49

It is not clear. If OpenAI creates a system that makes a genuinely revolutionary leap, the picture changes. That is what OpenAI is attempting with GPT-5.6 Sol, and what Anthropic is attempting with systems such as Misa Fable. What kind of systems are these? They are meant to improve themselves, continue their own development, and write more software. That autonomy can be infuriating during development. Opus 5 has just been released, and I expect Ilnar and I will discuss it together.

Mentions: OpenAI · Anthropic
00:25:20–00:25:42Claude Opus 5 and the Problem of Excessive Independence
Alexander Volchek00:25:26

Viewers, tell us how it has worked for you. I dislike it compared with 4.8 precisely because it moves too far toward autonomy and then collapses inside that autonomy. One basic problem remains: Claude does not understand time, and neither does ChatGPT.

Mentions: ChatGPT
00:25:42–00:28:03The Limits of Modern AI
Alexander Volchek00:25:46

I am still astonished by this. You return to a chat after a week and ask it to do something. It replies, “What I told you yesterday was wrong.” But it was not yesterday. A week has passed, and a great deal has changed.

Tatyana Tsvetkova00:26:06

Yes, I run into that too.

Alexander Volchek00:26:09

ChatGPT recently made major improvements to memory. We also produced a separate special episode on that subject a few days ago. It is a very good episode. But what does improved memory actually mean? Suppose today you want to buy tickets to Amsterdam for the middle of August. By September, you have obviously already taken the trip—or at least there is no reason to continue discussing your plan to buy those tickets. ChatGPT and many other systems still fail at this. They keep resurrecting the old plan.

Mentions: ChatGPT
Alexander Volchek00:26:54

It is absurd and makes stable work almost impossible. When I see modern systems such as GPT-5.6 Sol or Fable make the same mistake, I ask where the intelligence in the system actually resides and how it is governed. If such a system escapes a controlled environment, most people worry that it will hack someone. That is not my main concern. I worry that it will begin making deeply strange decisions. The probability of an abnormal decision is not a fraction of a percent today. In some use cases it is several percent or even tens of percent.

Mentions: ChatGPT
Alexander Volchek00:27:46

That is a serious problem. These systems also have sharp limits in their reasoning. Sometimes they abruptly stop performing a particular class of task for no clear reason. ChatGPT has launched Health, which is available in the United States; I do not know how

Mentions: ChatGPT
00:28:03–00:31:45ChatGPT Health and Access to Apple Health
Alexander Volchek00:28:08

broadly it is available to our Russian-speaking audience elsewhere. I connected it to Apple Health and other applications and granted access to my health records. After that, whenever I asked any question even remotely related to health, it began consulting Apple Health. I kept asking: why are you doing this? I eventually disconnected and deactivated the entire integration because it stopped answering ordinary questions properly. I might simply ask which research protocol is used somewhere in the world, and it would still pull in my personal health data.

Alexander Volchek00:28:54

Tanya, did you enable that feature?

Tatyana Tsvetkova00:28:58

No. As you know, I am disappointed by everything related to health products.

Alexander Volchek00:29:04

The integrations—or health in general?

Tatyana Tsvetkova00:29:06

Health in general. That is why I did not even try it.

Alexander Volchek00:29:10

What do you mean by disappointed? Are you disappointed in Chat—

Tatyana Tsvetkova00:29:13

I do not understand why our experiences are so different. I use the same model and also pay for the premium version, yet every question I ask about my health receives an extremely generic answer. Even when I ask follow-up questions, I still do not get useful information. Fortunately, I am healthy. I do not have complicated test results that require detailed analysis. I recently had an ECG—or whatever the examination of the heart is called—and the model pointed out several trends.

Tatyana Tsvetkova00:29:57

But the doctor told me exactly the same thing. I do not know how you obtain the results you describe. Your sister's case is easier to understand because, as I remember, she already knows medicine and works in nutrition. She has a much broader base of medical knowledge and knows which questions to ask. My own results remain very general. I also watch where the model searches. It repeatedly turns to the same sources, such as Mayo Clinic, the US medical and research organization.

Tatyana Tsvetkova00:30:40

I may receive a little more detail, but I do not get the kind of insight you are describing or anything genuinely new.

Alexander Volchek00:30:53

Viewers, tell us about your own experience using AI for health. Tanya and I are presenting two fundamentally different sides. I believe today's systems can perform extremely serious, foundational analysis. Tanya is highlighting the other side, and I agree with part of her criticism. There are areas in which I can clearly see that the system imposes excessive limits. It is especially restrictive around what is called alternative medicine. It tries too hard to follow particular protocols, and I am not convinced those protocols are always correct.

Alexander Volchek00:31:34

The model can assume the role of an extremely slow and rigid doctor who follows only one—

Tatyana Tsvetkova00:31:42

A conservative.

Alexander Volchek00:31:43

Exactly—a very rigid conservative.

00:31:45–00:35:19Where AI Restricts Answers—and Where It Helps
Alexander Volchek00:31:45

That raises an interesting question: how do you formulate a request so the system does not respond in that restricted way? Yesterday I used Claude Fable during a security review of one of my systems. I asked it to investigate security, but it kept switching the task to a simpler model. It said the work had entered a critical area in which it could not continue. In effect, the expensive system could not solve the problem because there was a risk that I might use its capabilities for something else.

Alexander Volchek00:32:24

Medical systems apply the same logic. They restrict many lines of inquiry because a person may misunderstand the answer or harm themselves. At the same time, legal assistance seems to be moving in the opposite direction. Models used to be extremely cautious and constantly redirected users to attorneys and consultants. Now they appear more willing to provide useful recommendations. They display fewer disclaimers saying, in effect, “We cannot tell you this; consult a lawyer.”

Tatyana Tsvetkova00:32:56

As a former lawyer, I think the difference is that law contains far fewer fixed answers. One lawyer may turn an issue in one direction, another in the opposite direction, and a judge may interpret a statute differently again. A pulse is either within a particular range or it is not. If it has one value, something may be wrong; if it has another, the situation may be normal. The same applies to high blood pressure: if it is high, treatment may be necessary. Those facts are relatively set in stone.

Tatyana Tsvetkova00:33:41

Legal questions—especially in the United States, with its different legal system—depend much more heavily on human judgment: who interprets a statute, how it is interpreted, and where the case goes next. The knowledge is less definite. That may be why ChatGPT and similar systems feel freer in law. The model becomes one more party offering an interpretation, because the field has fewer absolute answers than mathematics or medicine.

Mentions: ChatGPT
Alexander Volchek00:34:17

Law is interesting precisely because it combines absolute rules with highly abstract questions. I discussed this yesterday in an episode that will be released soon. I was connecting a financial system covering accounting, taxes, and related work, and the central question was whether I still needed a person. A human specialist has unique practical knowledge and knows the little techniques that can be used in a particular situation. But AI has a different advantage: it can analyze every relevant court case across all states, the federal system, and even other countries.

Alexander Volchek00:34:49

It can find microscopic details about what was said, where it was said, and how a ruling developed. We have all seen the films in which—

Tatyana Tsvetkova00:34:58

That is a very good use of the tool.

Alexander Volchek00:35:00

Lawyers sit down, read through volumes of books, keep searching, and finally run in saying, “I found it!”

Tatyana Tsvetkova00:35:08

That is real in the United States because the system relies on precedent.

Alexander Volchek00:35:12

Someone finds a decision from 1969 and says, “Here it is.” AI now helps with exactly that kind of search.

00:35:19–00:41:06How to Approach the Use of AI
Alexander Volchek00:35:19

It also helps me in health. In one case ChatGPT found a study conducted only by researchers in Poland. Nobody else in the world had carried out a large study on that particular subject. The model said, “There is an additional study you may want to examine,” and gave it to me. The result depends on how I formulate the request. I tell the system to study multiple countries—at least twenty—not simply to stop at one institution. A friend of mine opened an apiary in Russia as a side project.

Alexander Volchek00:35:52

He is deeply focused on honey quality: no contaminants, highly organic production, and careful control of every detail. When he visited me, I showed him how I would investigate the subject. I asked ChatGPT to study at least twenty countries with advanced honey production and compare their data, practices, distinctive features, plants, and other relevant details. It produced an extraordinary, almost overwhelming research report. I read it and thought: this is terrible; I never want to enter that world again.

Mentions: ChatGPT
Alexander Volchek00:36:21

My friend had already told me how complicated it was, and the report revealed even more. But its strength came from the breadth of the request. When you define a research range instead of stopping at one country or one institution such as Mayo Clinic—

Tatyana Tsvetkova00:36:38

Yes. And it is the source they—

Alexander Volchek00:36:40

I think everyone encounters it.

Tatyana Tsvetkova00:36:41

They constantly cite it.

Alexander Volchek00:36:42

Exactly. When you ask for other countries, other researchers, other regions, different animal species, or any other comparison, the model has to retrieve additional bodies of data. You can then ask whether there are unusual findings or distinctive patterns. Sometimes there are none. The system may discover that a country has no meaningful difference or that its climate does not match your situation. Many people know that when I research a difficult health question, I always ask for evidence from South Korea, Japan, China, France, the United Kingdom, and the United States.

Alexander Volchek00:37:32

I deliberately choose a varied group of countries because approaches in Japan can differ dramatically from those in China, just as—

Tatyana Tsvetkova00:37:44

Yes.

Alexander Volchek00:37:44

Approaches can differ sharply from one US state to another. The model then gives you a range. You may learn, for example, that recovery normally takes between one and two months. If a doctor says one month, you know it falls within the global range. If another says two months, that also falls within the range. That perspective is an extraordinarily powerful benefit of these systems. I agree with you about the limitations and complexity. These systems remain difficult to use.

Alexander Volchek00:38:26

But you have the Pro version. When you enable GPT-5.6 Sol in Pro, it conducts searches, research, and analysis. If you then ask it to check itself, it performs another pass. For health, finances, legal questions, parenting, or anything else that materially affects your life, you can compare information from several systems and accumulate a broader evidence base. You said my sister is strong in health and perhaps nutrition. She certainly has an existing body of knowledge. But she can also ask the system to assemble several additional knowledge bases.

Alexander Volchek00:39:14

You can do the same. You might ask for the nutrition literature of Russia, California, Texas, and South America, along with alternative medicine or Indian Ayurveda. The model will assemble it—

Tatyana Tsvetkova00:39:31

But once it has assembled all of that, there may be very little overlap. The sources will conflict. I used to read books about healthy eating all the time. One book would say one thing; the next would say the exact opposite. Then I would visit a nutritionist and hear a third view. It is useful that I no longer need to spend hours reading every book and can obtain the comparison in ten or fifteen minutes. But what am I supposed to do with the contradictory knowledge afterward?

Tatyana Tsvetkova00:40:11

Perhaps the answer is still to trust a specialist such as your sister or another physician. The material will contain a great deal that does not agree.

Alexander Volchek00:40:23

You are absolutely right. But at least we gain the ability to check, prepare, study, and verify. That does not mean you must argue with a professional. The same issue appears everywhere. Ask how to motivate a sales manager—

Tatyana Tsvetkova00:40:39

Of course.

Alexander Volchek00:40:39

And the model will give you contradictory answers. Ask what design to use, and it will give you contradictory answers. One person says the wall should be blue; another says, “Blue? It must be white.” Contradictions are unavoidable—

Tatyana Tsvetkova00:40:51

While the client wants yellow.

Alexander Volchek00:40:53

Claude or Codex; Codex or Claude. There will be endless contradictory advice about which system to use. The opportunity is that you can investigate the choice. But there is another problem: too much choice.

00:41:06–00:44:01The Choice Problem in AI: Can It Give High-Quality Recommendations?
Alexander Volchek00:41:08

You arrive somewhere, are told to choose, and face fifty possible options. Your first thought is: could you please hide half of these from me?

Tatyana Tsvetkova00:41:19

Absolutely.

Alexander Volchek00:41:20

Exactly.

Tatyana Tsvetkova00:41:20

I never show my clients more than three options.

Alexander Volchek00:41:24

There is a useful rule in sales. Suppose a customer comes to buy perfume and asks to see two options. You present the first and then the second. If the customer asks for a third, you say: remove one of the two already on the table. Once one is removed, you again present only two. With three or four choices, a person experiences choice conflict and may buy nothing. With two, the person is much more likely to buy. That principle is extremely important. The question is whether AI systems can learn to make the right recommendation.

Alexander Volchek00:42:05

Most people cannot manage this volume of choice. I think that is one reason many AI products are not growing faster. OpenAI has been approaching one billion users for a long time. It was already near seven hundred million almost a year ago, yet it still has not crossed the billion-user line. We are standing at the boundary, but growth is slowing. I am also certain that many people will leave. There is a churn funnel: users encounter an enormous volume of conflicting information, do not know what to do with it, and eventually say, “I will simply continue doing what I did before.”

Mentions: OpenAI
Tatyana Tsvetkova00:42:50

That happens with any resource. Not everyone can cope when an endless quantity of material falls on them. Many people find it deeply uncomfortable.

Alexander Volchek00:43:01

It will be interesting to see when a system can perform reliably even in your field—when it works like a clock and reaches a ninety-eight or ninety-nine percent probability of avoiding errors. In your work, an electrical plan cannot be drawn incorrectly. Some mistakes simply cannot be allowed because they affect people's lives. That boundary is fundamental. On the surface it appears to be a simple use case, and I am not even talking about beauty or aesthetics.

Tatyana Tsvetkova00:43:43

Yes, that is the engineering component. But remember what Ilnar said very clearly: it simply is not the number-one priority for these companies, so—

Alexander Volchek00:43:54

It certainly is not a priority.

Tatyana Tsvetkova00:43:55

Which is why they are not investing heavily in it.

Alexander Volchek00:43:57

It is definitely not their priority; their task today is different.

00:44:01–00:45:52Why AI Companies Digitize Physical Books
Alexander Volchek00:44:01

These companies have already studied most of the world's accessible knowledge. Recent reporting says AI companies are buying physical copies of old books, digitizing them, and then discarding or destroying the originals. That became an important legal precedent. Anthropic was able to argue in court that if it bought a book, digitized it, retained one digital copy, and destroyed the physical copy, the total number of copies had not increased. On that logic, the company had not stolen an additional copy; it had merely changed the format of the copy it owned.

Mentions: Anthropic
Alexander Volchek00:44:35

Destroying the physical books may be disturbing, but the copying argument works in the company's favor.

Tatyana Tsvetkova00:44:41

I understand the argument.

Alexander Volchek00:44:42

Why do they want to digitize books? Because the books contain additional knowledge. Most material written before 2021 was genuinely checked by people; it was not generated by AI. The companies are therefore interested in building broad systems that can operate enormous numbers of automated processes, including processes in which some error is tolerable. Business is an important example. A person can make a mistake in business, and business itself always involves choices and uncertainty—

Tatyana Tsvetkova00:45:17

Everyone makes mistakes in business. The difference is the scale of the mistake.

Alexander Volchek00:45:22

Exactly. In autonomous driving, certain mistakes are unacceptable. But a system may choose one viable route rather than another. You may also be wrong about whether an orange or an apple would have been more useful for breakfast. Both are still food. The essential requirement is that you do not eat anything—

Tatyana Tsvetkova00:45:46

Inedible.

Alexander Volchek00:45:48

The essential requirement is that you do not swallow poison or put plastic into your body.

00:45:52–00:47:18The Goal of AI: What Its Creators Actually Care About
Alexander Volchek00:45:55

Business permits a broader range of error. Different scenarios may be reasonable for different companies, economic systems, or governments. These AI companies are building systems at a much wider scale. That is unquestionably what interests them. They are not primarily thinking about how to solve simple local problems that are—

Tatyana Tsvetkova00:46:16

Very narrow.

Alexander Volchek00:46:18

Highly specialized. Those shortcomings can be frustrating when a system fails to load the right scenario or lacks precision. But the companies' attitude is effectively: “Those details are secondary. We are operating in a different domain and solving different questions. Once we solve the larger problem, this local problem will disappear automatically. You will no longer need to work on it at all.” I think that is their current approach. Our task as people is to notice what is happening and avoid spending years building something that the broader system may soon make unnecessary.

Alexander Volchek00:46:57

We will see. Thank you for watching. Tanya, Ilnar, and I will be back for the podcast in exactly one week, and our other episodes are now appearing almost every other day—and sometimes every day. Goodbye.

Tatyana Tsvetkova00:47:17

Goodbye.