Skip to content
Transcript

Transcript · 198 · Is AI Reaching a New Level? ChatGPT with a Wallet, New Capabilities from Google and Claude | News of the Week — ToTheMoon

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

Episode overview
00:00:00–00:03:22AI News: Google Ask Maps and ChatGPT Dots for Travel
Alexander Volchek00:00:08

Well, hi everyone! We're on the To The Moon channel: tech news, insights from Silicon Valley and from all over the world. Today I'm a bit farther from Silicon Valley, I went further north to the ocean, to Mendocino, but actually it's very close. Incredible beauty outside the window, just crazy. And we're starting our weekly news. This format came out last week. We'll continue this format, right? Plus this week, on Monday, we also had a new format, by the way: we started recording a short live, and on Thursday a new author will also appear on the channel.

Alexander Volchek00:00:48

Our other episodes, which come out on Friday and on Sunday, wait for them, and the Sunday podcast too. Everything will be there. So, Google Maps puts together a trip plan and looks for open slots at local services. So, what happened? Google is expanding this system it has, Ask Maps. It's a conversational assistant inside the maps. And now it puts together not just a list of interesting places but a day-by-day trip plan with numbered stops, right, and the routes can be shared, exported and so on.

Mentions: Google
Alexander Volchek00:01:22

That is, for example, what scenario works: say you ask, "Put together a three-day trip, for example, to Mendocino along the coast on Highway One, I'm driving from home." So, and, I don't know, short walks, lunch, work there, filming To The Moon videos. And the assistant suggests a sequence of stops that you can change, right there in the chat, and before departure it shows you information on road congestion. Honestly, you know, I'm terribly unhappy with how Gemini works in the car, and I don't really believe they'll manage to do this well now.

Mentions: Gemini
Alexander Volchek00:02:01

But still, Google Maps as a separate app on the phone, if you connect it through CarPlay, works much better, more effectively. And such things, of course, become very much about trips. The problem is, if this system isn't integrated with my ChatGPT, in which I prepared for the trip, what do I do then? If it isn't integrated with something else, what do I do then? For example, my Dots at OpenAI knows where I went, where I am, and it's much easier for it to recommend something to me.

Mentions: Google · ChatGPT
Alexander Volchek00:02:32

For example, we arrived in Mendocino yesterday, and I say to it, "Where are there beaches to walk the dog?" I have the dog with me too, right. And which beaches would you recommend where you can walk the dog without a leash? Show yourself to the camera. There. Which beaches would you recommend, that's it, good girl, run! To walk without a leash. And, by the way, interestingly, this morning Dots sent me: "Under no circumstances go out to the beach with the dog today, because waves of 7–8 meters are expected."

Alexander Volchek00:03:03

Well, indeed, there's a storm here, although of course everything looks cosmically great. Let's move on. This week ChatGPT Wallet came out. Over the last week we got an update, and I've even already talked about it a little in some places. What is it in general? So, OpenAI launched payment with a saved card.

Mentions: ChatGPT
00:03:22–00:07:25ChatGPT Wallet and Google Wallet: Shopping and Managing Finances Through AI
Alexander Volchek00:03:23

And essentially there are separate instructions for stores, so that stores work properly and fully with this system. And I want to say that I tested it in different stores, and it worked, I think, even in stores that definitely didn't know at all that OpenAI and GPT Wallet exist. In some old, obscure stores. I added several cards there and said: "Go to this site for me, fill it in, find me such-and-such goods there." There I bought rice, some items, then I bought cod separately, went in and bought one more item, a New Year's a New Year's gift was needed, a, there's this German toy, this German toy, there's this German smoker toy.

Alexander Volchek00:04:10

My dad first brought it, I think, in '91, in '92, and at fairs, I traveled a lot in Europe, and I've remembered it since that time. And for a friend of mine who keeps bees, in Russia, in Sochi, I decided to buy such a smoker, a beekeeper. And what I liked, look: when you just ask ChatGPT to find out something, how much something costs, it can skip some details. And, for example, yesterday I had a situation when Dots had already started completing a purchase fully through this Wallet.

Mentions: ChatGPT
Alexander Volchek00:04:40

And you, sort of, place purchases, it only works on the Pro plan, it only works through Dots, and in Dots you say: "Make a purchase for me." It completes it fully and at the end sends you a kind of summary, like the final cart, in full, which you pay for. Just like Apple Pay actually does it, in Apple Pay the final cart is visible. And you click one button, "confirm", and you buy it. So yesterday, on one of the purchases, it sent me: "Actually, the system didn't tell you here, there are additional fees." I think it's a very convenient system.

Mentions: Apple
Alexander Volchek00:05:13

I started using it, I liked it a lot, yes. That is, over these few days. Write whether you liked it and whether you expect such things from other systems. We'll talk more, Google has something like this, similar things, a little similar. Just one of the news items: Gemini analyzes personal spending through Google Wallet. They have Google Wallet, it already existed. They didn't manage to do the thing that, I think, OpenAI plans to do. I think OpenAI plans to run billions of dollars, maybe hundreds of billions, billions of dollars through its system, through this direct integration with banks, right, bypassing various websites.

Mentions: Google · Gemini
Alexander Volchek00:05:46

And websites will essentially be certain intermediaries. They have to be made great, designed great, and in general one of the key things now, I'll add to the news, one of the key things, a very important one: if you have websites or some portals. Now you really need to adapt them a lot, of course, so that a robot doesn't just do some search on them, but somehow gets the instruction that you need. So, on November 13 Google began rolling out Google Wallet integration directly in the Gemini app.

Mentions: Google · Gemini
Alexander Volchek00:06:18

And before, Gemini's personalization covered, there's, of course, we all know, Gmail, mail, there are photos. Now the answers can also rely on financial transactions. And essentially, well, they integrate with their Wallet and with some other systems. Essentially, ChatGPT has finance, right, as you know, and Claude also has its own finance module for integrating with various systems. And, for example, what's a clear example? Here, Gemini and Google Wallet. You write, for example, in Gemini: "Google Wallet, which subscriptions am I spending money on now, which have become more expensive over the year?"

Mentions: Gemini · ChatGPT · Claude · Google
Alexander Volchek00:06:52

It, accordingly, checks and looks at them for you. It's a good topic, financial integration with these systems in general is good, I like it. There will be fewer problems with banks, fewer problems with various obscure financial services, consultants, with everything else. So, next. The next news item: companies got their own cards and a consultant in ChatGPT. So, what can you do now? You can go into ChatGPT, add your company through the business module, and you don't necessarily need a business subscription, I was able to do it on a personal subscription too.

Mentions: ChatGPT
Alexander Volchek00:07:24

Then you describe this company, that is, you give some, well, some description here. To be honest, the description looks very strange so far.

00:07:25–00:13:12ChatGPT for Business, Gemini on Pixel and Gmail: AI Inside Apps
Alexander Volchek00:07:29

I don't understand why artificial intelligence doesn't step in here and make some, well, super cool trendy description. And inside you have the option to make, like, your own consultant who will be able to consult users. There's an important nuance: when you click on this consultant, it isn't independent advice from ChatGPT itself, right, but a separately designated conversation with a representative of the company. And a business can sponsor such an agent, of course, and can direct users to it from ads. And, well, of course, ad placements are labeled.

Alexander Volchek00:08:08

So, by the way, which of you has ever run OpenAI ads? We have OpenAI ads running in one project right now, in different ways, well, there's a different effect. And here the question, of course, is how to measure all this, right? That is, either you really measure the number of sign-ups, leads that come to you, I don't know, purchases that happened, or you measure some PR effect, right? Of course, right now OpenAI gives a certain PR effect. Gemini on Pixel. So, what happened? Again, there was a November update now for Pixel and Android, right?

Mentions: Gemini
Alexander Volchek00:08:42

And they expanded the set of features. They have Gemini Intelligence, and artificial intelligence now works with the phone's context, not only answers questions in a separate app. And certain phone models that exist got the new capabilities, for example, the Pixel 10, right? And Motorola will release the Signature 27, and orders in the States will only start now, I think in a few days they'll be there. And, accordingly, this part of the features was on the Pixel 11 before, and now they're expanding it. What are some examples, for example?

Alexander Volchek00:09:19

It turns out they can turn live dictation into a neater message. That is, the user says: "Well, I'll probably get there around seven o'clock, half past seven. So we'll be there for dinner." And the system suggests a specific, nice text without filler words, without all the extra details. And plus, they have a cool thing where you can write: "Show me the flight, the weather and the booking for the trip." Artificial intelligence builds a widget on the home screen. That is, you can make different widgets. In general, of course, the system of widgets built with artificial intelligence will obviously develop a lot.

Alexander Volchek00:09:55

But let me remind you of my view: I'm sure that all these systems will migrate into their own, sort of, operating systems. And essentially these widgets will be inside such systems. So, Gemini in Gmail has opened up to more work and school accounts. Many people have Gmail, and many of you haven't seen various features. Previously, some of the features depended on the specific Workspace edition, if you use it corporately. And now Google is expanding access to essentially two capabilities: a summary of a long email thread and a non-professionalized assistant, it's called Help Me Write, right, creating and reworking a letter according to instructions.

Mentions: Gemini · Google
Alexander Volchek00:10:37

I think Google is lagging behind the market in these respects, and this isn't some super news anymore, just some micro, micro, micro, micro things. And here I also want to talk about an interesting micro thing. There's news that Medium, which is a fairly large mass medium, has introduced an assistant right inside personal notes, right? That is, what happened there? For example, in the editor you can write: "In the fall I want to shoot an episode of To The Moon on Anthropic going public with an IPO, for example.

Mentions: Google
Alexander Volchek00:11:08

And remind me what arguments I've already written down on this, for example." And the assistant looks for suitable notes within Medium itself, or right in the middle of an article you can leave some comment, for example, and write to this artificial internal intelligence: "Suggest a clearer headline or do something else." On the one hand, the topic seems alive, right, and we understand that inside the systems that serve such various infrastructures or are mass media, this will develop.

Alexander Volchek00:11:44

On the other hand, to be honest, I understand that they'll lose the market, right? Because people will write articles more and more. By the way, there's a cool news item: how the job and labor market has changed, right? People will more and more, obviously, people will more and more describe all this directly in artificial intelligence systems. Now, just a second. I'll put some more wood on, the fire seems to have gone out. Claude got shared spreadsheets, boards, ideas, analytical documents.

Alexander Volchek00:12:16

Anthropic opened Claude Sheets and Claude Whiteboards in beta and made it generally available for paid plans. The idea is not to forward results from the chat to a separate program where artificial intelligence no longer sees further changes, but to run the document itself with the assistant and colleagues, right? For example, let's say there's some organizer, I don't know, a podcast organizer says: "Make a plan of sponsor placements for three months with budgets and people responsible."

Alexander Volchek00:12:47

And Claude, accordingly, creates sheets with formulas, drop-down lists and formatting, and a person can correct values in the cells. And of course there's an assistant working inside that can recalculate a whole lot of things. And at the same time, Anthropic openly admits that this Sheets has basic capabilities for now, and complex tasks are better solved directly in Excel, in full-fledged programs of various kinds. I am, of course, deeply convinced that people won't use Excel and Word separately at all, right?

Mentions: Claude
00:13:12–00:17:55Claude Sheets and Meetings: Spreadsheets, Documents and Meetings with AI
Alexander Volchek00:13:14

And for me personally Excel has absolutely, completely, well, 99% disappeared from my daily life. I wonder how things are with you, by the way, with Excel in general? Do you use Excel? Do you have Excel? Write. I've essentially stopped understanding, well, why I need it at all. Claude Meetings — now this is an interesting news item. So, the beta rollout of this has begun. Anthropic showed Claude Meetings as a feature that would appear later. And they announced, they announced this feature just a few days ago. It's not just a promise, because there's been a lot, a lot of discussion about it in general.

Mentions: Claude
Alexander Volchek00:14:00

For example, what will happen in Zoom, what will happen in Google, in Google Meet, Microsoft Teams and so on. So, how does it work? To a video meeting in Google Meet — right now they have Google Meet as a partner — a separate participant is added, Claude, and during the discussion you can ask it to put together some document, draw some diagram even on a shared board inside this Google Meet, or, I don't know, bring up some past financial expenses that you have in the company, and go and draw on the data it has.

Mentions: Google · Claude
Alexander Volchek00:14:36

Of course, at the end it makes various summaries and everything else. Note that, of course, obviously, obviously, personal assistants will be inside all these systems, rather than these systems' personal assistants helping us. In Zoom, for example, there's an AI assistant, I don't use it. Which of you uses it? I don't use it, right? I'm fundamentally not interested in getting additional third-party systems. Yes, of course it can send, send a transcript to email, my other system can take the transcript from the email, but it's much more beneficial for me to immediately launch a full analysis of this event with the help of, for example, ChatGPT or Anthropic Claude. Of course, what's interesting here is the story that it's built inside, right?

Mentions: ChatGPT · Claude
Alexander Volchek00:15:25

I think it's a certain benefit, because it can give suggestions within the meetings that are taking place. And, and I once expressed the thought that the solution will be cool, fundamentally cool, if this system can also close some issues. That is, you immediately say: "Listen, let's launch a new promotion today." And then ChatGPT or Claude or, say, Google Gemini go off in parallel and launch the new promotion, and the meeting isn't even over yet, and they already come back and say: "The new promotion is set up," right?

Mentions: Google · ChatGPT · Claude · Gemini
Alexander Volchek00:16:01

Now that's a really cool thing. Of course, I'd like something like that to work. Scribe v3 came out, it's from ElevenLabs. And I can say right away that it's a system for working with audio, if you don't know, and for me it's the best system for working with the Russian language. Once a friend of mine recommended it to me, because I was heavily using ChatGPT, Transcribe, OpenAI's products. Before that, of course, Google has a huge infrastructure for everything to do with voices. Well, for example, working with Russian, I think ElevenLabs does it best today.

Mentions: ChatGPT · Google
Alexander Volchek00:16:38

It's very affordable. A model that works great. Scribe v3, I think about four days ago we moved all our projects for analyzing conversations well, and even for analyzing podcasts like this, because all the transcriptions are posted on the website, and articles are made from them. We moved everything to Scribe v3. Write who has used it. On November 14 the official Grok Imagine account announced that creating images and videos is now available directly in Grok. And before, this generator of theirs, Imagine, existed as a more separate mode. Now, then, you can do this inside their infrastructure.

Alexander Volchek00:17:25

Well, and logically, obviously, they'll connect all the infrastructure with each other, and maybe some cool feature will appear soon. You're driving a Tesla and you say: "Grok, take a photo of me and what's on the street, and make a cool picture out of it," right? Or the car is driving, saves everything from the cameras, and you have the option to make a small fly-through video. For example, in my Escalade, the Cadillac, the electric one, it has this, in this car, it records the movement around you as you drive, the whole road completely, right?

00:17:55–00:20:17ElevenLabs Scribe 3 and Grok Imagine: New Tools for Audio and Video
Alexander Volchek00:18:01

But there's no connection at all there to make this video for me. Although the videos would be fun, and even fun as something to send within the family if you went somewhere, or, it's fun, if you're a blogger, to post even some small content, say, I don't know, on Instagram or Telegram or somewhere else. In short, this will develop, it'll be great. So, Gemini has updated its module for healthcare and for clinics' administrative processes. And now in American clinics, true, an employee can turn to different systems that aren't connected to each other with the help of Gemini.

Mentions: Gemini · United States
Alexander Volchek00:18:43

That is, Gemini has essentially taken on the function of a connector between different systems. And one of Google's executives, Google Cloud's Thomas Kurian, said that the main problem of healthcare is a lack of information. And, of course, one of the most important stories is the fragmentation of the data that's inside. And artificial intelligence systems will help with this. You and I understand that the healthcare system is very corrupt, constrained, there are people's lives at stake, everything has to be re-checked 800 times.

Alexander Volchek00:19:19

Many of the standards we use are already 20–30 years old. And, of course, artificial intelligence will be adopted there very slowly, with difficulty, and people will more and more get things themselves with the help of artificial intelligence systems and get various information, rather than with the help of clinics. So, what happened in programming models? Well, first of all, Claude Fable 5.2 and Mythos 5.2 came out. And I already covered this in several episodes last Friday.

Alexander Volchek00:19:47

I just want to mention it once again: last Friday there was a special episode, watch it. It's super strong news for this week. And there have been significant changes in code development there. I recommend everyone try this model. By the way, who else has this feeling: for the last three days I've had just terrible negativity toward ChatGPT, specifically the regular chat. That is, I was negative from time to time, well, quite often, about Codex, but not about ChatGPT. And what am I seeing now?

Mentions: ChatGPT
00:20:17–00:23:30Google Gemini in Medicine and Claude Fable 5.2: New AI Capabilities
Alexander Volchek00:20:19

Of course, ChatGPT has started answering very poorly, and, relatively speaking, on the most expensive plan, on fairly simple tasks: it doesn't get to the bottom of things, doesn't finish tasks, answers very abstractly, very quickly, very badly. Not great. But once again: this is, of course, my habit of the quality that already exists. But that's my conclusion for now, right? And more and more I want to use systems like Fable, again, when developing certain solutions. I don't like how code, how Claude answers ordinary questions. I like ChatGPT more.

Mentions: ChatGPT · Claude
Alexander Volchek00:20:51

By the way, who likes Anthropic Claude more for answers in everyday life, or, say, business analysis, or analysis of professional activity, or, say, working with texts? Well, in short, everyday life. Write, it'll be interesting. Gemini Carbon, we know, right, 4, there's Gemini 4 Carbon, and it was originally presented on October 28. So, what's new now? Access has expanded in Google's apps and tools, and, accordingly, more and more of those who haven't seen Carbon will see it.

Mentions: Claude · Google
Alexander Volchek00:21:23

Carbon will start appearing more and more in various integrations and in various systems. I think it's fairly useful news. And here's from Google: Antigravity launched, well, several programmers on one team. So, in Antigravity they got, it turns out, regarding these programmers, a new feature appeared, it's called Swarm, and it's a swarm of agents. And previously, even a very strong assistant ran a task in a single thread. For example, it read files there, wrote code, checked various changes itself.

Alexander Volchek00:21:58

And for example, what has happened now? A coordinating agent can split a large task among several specialized executors who work in parallel. I do have a question, of course: guys, but Claude and ChatGPT showed this very seriously half a year ago, right? Maybe even earlier than half a year ago. Have you really done something fundamentally new? Here, rather, ChatGPT Dots, right, really distributes work, right, it assigns different tasks. It glitches sometimes, true, but it assigns different tasks. What's new with you? That is, the example Google describes: for example, you ask, say, to move an online store to a new payment system, for example, you want to switch, I don't know, from Stripe to PayPal, while keeping the reporting, all the tests, all the details that were there in the payment system.

Mentions: ChatGPT · Claude · Google
Alexander Volchek00:22:49

And this coordinator separates out the payment interface, the server logic, the automated checks, and the executors work in isolated, separate, separate Git worktrees, that is, separate working copies of the project, and so simultaneous editing doesn't make changes collide in the same files. Let's see what happens, right? And on a shared board, by the way, the executors exchange conclusions in messages, and a person sees their work, roles and so on. Let me remind you, about a couple of months ago I covered Anthropic's Claude. They published research on agents working in parallel.

Mentions: Google · Claude
Alexander Volchek00:23:27

A very cool topic, right, when they launched, say, 80 agents or 10 agents at the same time and watched how agents can negotiate with each other, right, and can communicate with each other.

00:23:30–00:29:36Google Antigravity, Qwen 4 Max and Claude Agents: AI Agents for Programming
Alexander Volchek00:23:36

Well, all the systems are gradually catching up to this, gradually moving there. This feature, by the way, is available for Ultra subscribers in Antigravity and only for Gemini 4 Carbon. Well, okay. Qwen, a bit about Chinese systems. So, previously there was, they released Qwen, Qwen Max, Qwen 4 Max. It was in a preview version, that's a limited early version, it appeared last week. And so now a separate production identifier has been specified, Qwen 4 Max has appeared. And for those who use these models, look, this has appeared. The model is available through the API. Of course, with pay-per-use, a token subscription and so on, right?

Alexander Volchek00:24:15

So, for example, you can give the assistant project documents, photos, various diagrams, tables, and ask it to find various contradictions. It's not even clear what updates happened, right? It all needs testing, because when they say "a million context," it means you can add a table. We've been chewing on all this for several years already. Claude's agent management service can now be hosted on your own premises. Now this, by the way, is cool and very interesting news, because corporations have the problem that there are data storage rules, and Anthropic already supported running isolated execution environments for agent interfaces on the client's infrastructure, right?

Mentions: Claude
Alexander Volchek00:24:55

That's when the agent's code could essentially be used in its own sandbox, right? But the management service stayed with the provider. And now there's a new change, when you can keep this management on your own premises. And, for example, an organization can store agents there, all their histories, on its own, in terms of managing their life cycles. For example, a financial company can create artificial intelligence agents that study reports, for example, on the internal network.

Alexander Volchek00:25:34

And so it turns out the events of their work remain under the control of the corporate infrastructure. What do I want to say here in general? But this, by the way, doesn't mean that Claude works without the cloud, right? What I want to say is that the whole world keeps talking about open source, and we keep talking about how we need to have isolated systems, data. Not everyone will want to hand over data. Let's think broader. Does anyone really think that ChatGPT or Anthropic won't have the functionality, really won't have the functionality, where you go in and say: "For me this is a secret task, you must solve it only within my phone."

Mentions: Claude · ChatGPT
Alexander Volchek00:26:09

Or: "This is very important to me," or "for me, these finances that concern my personal accounts must not leave this infrastructure here." And, for example, what can ChatGPT do? First, it can anonymize a lot of data, encode it, encrypt it somewhere on its side, recognize it, do it. And what else can it do? It can launch some small local model for you inside its system. Personally, I'll tell you honestly, I'm sure that's how it will be. What happened next? Smart glasses from Gemini have appeared. And it turns out the frame holds a 12 MP camera, microphones, open-type speakers, various controls.

Mentions: ChatGPT · Gemini
Alexander Volchek00:26:55

And a person can look at a poster, for example, some, I don't know, with some ad or for a concert, or at a café, and check, for example, in Gemini, I don't know, if it's a restaurant, what the prices are inside. Or you saw some concert and can look up the upcoming events. Personally, I don't believe at all in this type of device, as always-on devices. Well, I don't understand what it is, because clearly the device shouldn't basically be built into glasses, or they should make something that will work, that you can attach to glasses separately, or wear, I don't know, on a T-shirt, or put somewhere else.

Mentions: Gemini
Alexander Volchek00:27:42

But it's clearly not a story of just ordinary glasses. And Zuckerberg promoted this story very heavily. And somehow I don't see it getting really wide adoption in the world. So, among the interesting devices that have appeared, the new Tesla Cybercab came out. Look, two days ago I filmed about it, a very short special episode. This is our new format that has appeared, Live. What does "came out" mean? It had already come out before. They started selling it. A $300 subscription, available in Texas, and you can buy it and ride in the car without a license. Watch that episode, it's very interesting.

Alexander Volchek00:28:25

Amazon has the Zoox robotaxi. A very interesting topic. I still haven't ridden in them, I didn't manage to ride. In Las Vegas I had access to them, but I would have had to walk a bit to get into the car. It's this four-seat cabin. Now, then, the robotaxis have appeared in Uber. Again, they're the active launch in Las Vegas. They're expanding this cabin. I really like how this cabin looks, that it's still an unusual car. Because if you take Waymo from Google, of which there are a lot here in San Francisco, you're still riding in a Jaguar, right? And now they have these new minivans.

Mentions: Google
Alexander Volchek00:29:03

I don't know what brand this company is. In our Bay Area there's a very, a car brand-manufacturer, a lot of them drive around, but clearly the future will be and should be something different. Zoox is more like the Cybercab, right? That is, there's no steering wheel, no pedals, no mirrors at all. A very interesting topic, and it'd be cool to know whether you'll be able to buy one like that for yourself. So, NVIDIA and Gemini Robotics, they have an update. NVIDIA, here, GR00T 2.

00:29:36–00:34:15Tesla Cybercab, Zoox and NVIDIA: Robotaxis, Robots and Gemini Smart Glasses
Alexander Volchek00:29:38

The model learns not only from recordings of a specific robot's movements but also from videos of people and other manipulators, to train robots. The topic of robots, of course, is very interesting in terms of how they'll figure things out. Let me remind you that probably the biggest robot in the world that exists is the robot vacuum. I mean the most widespread robot that exists is the robot vacuum. Of course, various things could be pulled in there, you could pull in a camera or pull in a car.

Mentions: Robotics
Alexander Volchek00:30:04

And I'm sure that robots, of course, it's not a question of the robot looking like Optimus, like an ordinary living person. By the way, this company NEO, where I made a prepayment for a robot, that robot just disappeared, they sent nothing. I need to get the prepayment back, by the way. It was small, I think about 300 dollars, but I need to get it back. Let me remind you, so, there's a robot, it's made by the company 1X, the robot is called NEO. The company has its main headquarters in California. Although, by the way, in the media they're described as Norwegian, they accordingly have part of the office in Norway.

Mentions: Robotics
Alexander Volchek00:30:38

The point of this robot, Ilnar laughed a lot about this, was that if the robot can't do something, a personal assistant from the company connects to it, right? Well, and there were all sorts of jokes: "What can't you manage to do?" And at some point a connection- I made a prepayment, I'll say right away, I chose a black robot. Tanya's husband, Steve, who's also my partner, forwarded it to me right away. At first he didn't understand what it was, because this robot is shown white everywhere, and I ordered a black one.

Mentions: Robotics
Alexander Volchek00:31:09

And he says a bit like: "What is this, what's this there? Like a fake?" I say: "No, it's this robot." And then he saw that I had, so to speak, prepaid for it. Well, it's just like putting down a deposit. I thought that, after all, given being in the Valley, discussing various things in technology, well, actually we have a channel, and I think it would be very- NVIDIA is developing this area strongly in terms of its infrastructure and interaction with robots and, of course, the use of its chips inside.

Mentions: Robotics
Alexander Volchek00:31:38

Gemini Robotics 2.5 is Google DeepMind, it's developing the transfer of skills and working with space and sequences of actions, right? And it turns out their materials include a broad set of 200 tasks and trials with partners, including a production scenario with Mercedes-Benz, and stated high percentages, they have fairly, well, seriously high percentages on selected, on selected operations, well, in terms of the result that was shown. The topic of robots, of course, is very interesting. Apple.

Mentions: Apple · Robotics
Alexander Volchek00:32:17

The news is that Apple is recruiting up to 200,000 people for research into medical artificial intelligence. In general I like in some ways how Apple Health is made, but at the same time, I'll say honestly, I terribly dislike what Apple is doing in terms of artificial intelligence. And the fact that they're recruiting 200,000 people for research is a seemingly great thing, because Apple knows how to do good research. But where is Apple now in terms of artificial intelligence? We'll see, we'll see what comes of it, right?

Alexander Volchek00:32:52

And Apple's vice president for health, Sumbul Desai, stated: "At Apple, we believe that responsible innovation begins with scientific rigor and clinical validity." Well, they're flying around somewhere in their own world, then. An interesting study, and I touched on it and talked about it with you from the very beginning. OpenAI and Upwork said that the structure of freelance orders is changing. They did a study. Actually, they studied 63 different categories there. And what did they find?

Mentions: Apple
Alexander Volchek00:33:24

So, previously an entrepreneur could separately order, for example, a draft of ad copy or a simple illustration, or, say, order video trimming independently. Now he can often do the preparation himself in artificial intelligence and give the contractor a more complex assignment, for example, to organize a full ad campaign, or merge materials, or check quality and release final versions. So, in general, demand in this area will of course change. I think two or three studies can't settle it here. It's clear that the number of such orders, in general orders in certain categories, will fall very substantially.

Alexander Volchek00:34:07

The market will restructure incredibly strongly, and, well, you, as a participant in this world, can take something interesting for yourself, right?

00:34:15–00:38:38Apple Health, OpenAI and Upwork: How AI Is Changing the Labor Market
Alexander Volchek00:34:20

You can see which orders are in high demand now and offer them, for example. In America, one of the representatives of the Federal Reserve System, by the way, talked about there being big problems with artificial intelligence, and he voices, of course, this concept of American inequality and the future of artificial intelligence: that there will be huge shocks in the labor market, and we need to prepare for them in advance. But, well, we'll see, we'll see what actually happens.

Alexander Volchek00:34:56

All of this, by the way, the story about wealth, relates first of all to the growing divide between the poor and the rich, or the concentration of money between the poor and the rich. One of the news items says that schoolchildren use artificial intelligence more often than they receive instructions on how to use it. And this is a new piece from OpenAI, and it was devoted precisely to a problem familiar to parents and teachers: that students use generative artificial intelligence to prepare homework, but adults don't always explain how to work with it without fooling themselves.

Alexander Volchek00:35:39

When I read studies like this, I think: "But adults themselves don't know how to work with artificial intelligence without fooling themselves." My sister and I were just discussing this question, in terms of children and understanding where children don't understand what's going on, where parents don't understand what's going on with artificial intelligence systems, that is, what to do about it? What's the situation at home with children for you in general? My own kids don't use artificial intelligence at all, essentially they're isolated from it.

Alexander Volchek00:36:09

But sometimes my son may take a chat and ask and find out something in artificial intelligence. At my daughter's school, in 11th grade, basically it isn't accepted- It isn't accepted to use artificial intelligence. There's an interesting story at Google DeepMind. Google DeepMind has developed Multimodal Knowledge Agent. And what's the logic? That most popular rankings ask artificial intelligence to answer a question or write a program. But in real work you need, for example, to open a drawing, fix mismatched windows in a plan, in some facade, for example, prepare a presentation with clippings of source pages, and not make up various details.

Alexander Volchek00:36:53

So they launched this Multimodal Knowledge Agent — a test of agent systems that read different types of files and create finished work documents, right? And essentially, what's the method there? Specialists from 83 professions in 10 fields compiled 200 complex tasks. Well, I started with construction, there's construction, there's finance, there's education, there's law, science. And a typical task is designed for several hours of an expert's work and contains source files, an expected result and about 70 different verifiable conditions.

Alexander Volchek00:37:28

Artificial intelligence can use various programs, it can view images, run code and take up to 250 sequential steps. And another model checks the results against a rubric. The team compares it with expert assessments. So, at the time of the check, in the Artificial Analysis ranking on the hard part of the criteria, Claude Fable 5.2 leads, Gemini Carbon is in second place and Claude Opus 5.5 is in third place, right? Fable 5.2 has 57%, compared with Opus at 53%. This, by the way, doesn't mean that 57% of projects are fully completed.

Alexander Volchek00:38:05

That is, Google notes that Carbon takes second place in quality in this test and costs on average six times less than the leader, yes. But there, again, I shot an episode last Friday, watch it, on the topic of how a model's cost depends on the type of model. It's a very interesting topic. So, well, if we touch on it a little, we all know that Jensen Huang has left NVIDIA. I want to recall this in the weekly news. They say there will be big problems, with Jensen Huang. And some claim that NVIDIA's strength has always been in its people inside.

Mentions: Google
00:38:38–00:40:45Google DeepMind: Testing AI on Real Tasks and Ranking Models
Alexander Volchek00:38:42

And some say, for example, Elon Musk in particular, that many of the deals that were within NVIDIA and the artificial intelligence market rested on a handshake with Jensen, not on a written contract. And this is a very important topic. By the way, I already thought, when Jensen was being removed a week ago, what a huge number of startups will die, or projects inside will die. That is, what will happen? By the way, the US Treasury Secretary said that they need to look for NVIDIA's chief executive outside NVIDIA, right?

Alexander Volchek00:39:13

And this is the secretary's public position. And, by the way, we see how the government wants to get into running, any kind of running of companies. Anthropic continues to prepare to go public. So far there's nothing interesting yet. If you multiply their stated valuation, they want a bit more than a trillion there, or, say, a trillion, roughly. But look, it doesn't say anything. That is, they could go out and list, and they could have four trillion, or they could have three.

Alexander Volchek00:39:43

Meanwhile the Fed, again, in the US reported in a newly published report, they had one on November 13, that there's more than a trillion in outstanding debt in artificial intelligence. But this actually isn't, isn't directly the debt of artificial intelligence, it's debt related to advanced computing, right? And, accordingly, they say there's a risk of funding cuts for this industry in general. Italy fined Meta 805 million dollars for processing photos. Again, then, Europe fines American companies, and I think it will keep fining them.

Alexander Volchek00:40:22

And Meta will keep doing everything to squeeze the maximum out of its users. Well, roughly like that. So, Instagram is expanding the definition of teenagers who have stated their age. This topic is interesting, by the way, because of what? In general, in all systems, you know, you could kind of lie. You could state that you're not 20 years old but 50, or that you're not 15 but 25.

00:40:45–00:44:10NVIDIA and Anthropic: The Situation Around Jensen Huang, the IPO and the AI Market's Debts
Alexander Volchek00:40:46

You live not in Minsk but in, in Los Altos Hills. You could state a lot, that you live not in Paris but are actually somewhere in South America. You could report some, swap addresses and so on. Obviously, all these systems will start digging deeper. And if, for example, a child has Instagram, they constantly post photos of themselves there, then the system will diagnose, check these photos, check the posts, check which audience follows them, roughly figure out how old the person is, ask them for additional authorization, which this person, of course, won't be able to do.

Mentions: United States
Alexander Volchek00:41:25

Or it will ask: "Do you run this account for yourself or not? If for yourself, then please confirm your face and your age." And that's it. Australia has begun changing its approach to artificial intelligence incidents, and this is a topic that will develop further in the world in general, because OpenAI was used to hack a government system. Well, and it will continue, and it will continue to be used, right? So, nonprofit organizations are demanding through the courts to stop the development of frontier models.

Alexander Volchek00:41:54

So, they've again filed a big lawsuit in a San Francisco court, filed a lawsuit against Anthropic, Google, Meta, OpenAI, SpaceX. And the plaintiffs want to limit the development and deployment of more powerful models, citing the risks of losing control, cyberattacks and everything else. But, guys, wait, these companies will pause, China won't pause. And are you sure they're ready to pause this now, while competing with each other? And how will you figure it out? This is very important, right? How will you figure it out? Is the model being developed super powerful or not super powerful?

Mentions: Google
Alexander Volchek00:42:27

And, by the way, on November 9 the Paris Call for AI Safety and Security declaration appeared. It's an appeal for joint safety of frontier neural networks. There are, then, six countries there: Bahrain, Canada, Finland, France, Norway and the Netherlands. OpenAI, Anthropic, Microsoft, Mistral, Hugging Face and others signed this declaration. And, then, it's also about the need to agree on principles. Everyone signs various agreements. But, as we know, we have an organization, the UN.

Alexander Volchek00:42:53

But people were killed then and are killed now. Countries fought then and fight now. And wars in many countries that haven't stopped for decades haven't stopped and won't stop. And finally, the last news item. Google Ads, that's Google's advertising, shows how much invalid traffic wasn't counted in the bill. Essentially, there's such a concept, previously Google Ads reported that it had automatically filtered out, for example, some invalid ad traffic, spam or something else, for example, artificial clicks.

Mentions: Google
Alexander Volchek00:43:22

And now it shows how much of that billing there was. I generally believe that, of course, the industry will develop very strongly if Google is interested in this, right? Or if Facebook is interested in this. Well, who the big mass advertisers are, right? Or if Yandex is interested in this. But, as you and I know, they're interested in increasing their sales volumes, their volumes of ad sales. So use frontier artificial intelligence models, like ChatGPT or like Claude, to create infrastructure to analyze, actually, your ad campaign, your ad campaigns. And this will give you an incredibly strong breakthrough. Thank you all very much.

Alexander Volchek00:44:00

Our weekly news episode is over. Please don't forget to support our channel. Always support us with a comment if you love us.

00:44:10–00:44:19AI Safety and Regulation: Lawsuits, Meta and Google Ads
Alexander Volchek00:44:10

And see you in the next few days in new episodes. Bye, everyone.