Dots, Bots, and Math! Oh my!
In a Nutshell
OpenAI's release of 722 mathematical proofs in a single month, following the solving of the Navier-Stokes equations with 10,000 agents, demonstrates exponential AI progress that's about to eliminate entire professional fields. The speaker argues that AI-resistant professionals and organizations will become competitively obsolete within 3-6 months as agents evolve from tools to autonomous systems that redesign workflows and replace human coordination. The solution is immediate adoption of personal AI agent systems and organizational transformation toward "Team A" structures where humans focus solely on value creation rather than routine tasks.
These notes were generated by AI and may contain inaccuracies.
OpenAI published more than 700 proofs. The mathematics community signed an open letter from about 900 mathematicians stating: "We did not ask for this and we do not want it." Internet users ridiculed these scientists harshly. The speaker feels great sympathy for mathematicians because this is a trillion-dollar giant, in addition to other companies like Google that work in mathematics and anthropology, targeting their careers. The position is considered short-sighted, and it should be remembered that their number is only between 800 and 900 mathematicians. Not every mathematician on Earth would say, "This would be disastrous for us." They rightly realize that things will change for them, whether for better or for worse.
One main example given is imagining oncologists who specialize in cancer searching on Google for DeepMind or OpenAI and saying, "We have cured 800 types of cancer using our AI technologies." Then a number of oncologists signed an open letter demanding that these techniques be stopped, saying, "We don't want them." People would burn down hospitals if they tried to ban them, because imagine trying to say to your child: "Sorry, but in order for cancer researchers to keep their jobs, they shut everything down, so you have to be treated the traditional way." This simply doesn't make sense.
Even some skeptics of the post-work economics or technological unemployment narrative are saying, "Mathematicians, stop trying to monopolize knowledge, stop being so condescending and arrogant about this." The essence of technology lies in progress, changing and improving work methods, but this will continue. One field after another will be disrupted, destroyed, or completely reshaped by the continuous development of automation, because the model that led to this has not yet been presented. Within three to six months, this capability will become the default setting that everyone possesses. This will happen again and again. Some people still say that artificial intelligence will not improve, but that's what they've been saying for four years, and they are wrong every time. We are still a long way from the maximum of this capacity, so people need to be psychologically prepared for the fact that we will have the equivalent of a billion Fields Medal winners within six months. Programming is recurrent superintelligence or recurrent self-improvement.
The news last Tuesday was quite shocking, with 722 research papers being published. About a month earlier, OpenAI announced that it had actually solved the Navier-Stokes mathematical problem. This was the first of the seven Millennium Prizes to be resolved. The issue has remained open for 80 or 90 years, and mathematicians have been working on it to no avail. However, the unpublished ChatGPT model produced 10,000 agents and resolved them in over 88 hours. This was the first major announcement in the field of mathematics. That was just one month ago. Four weeks later, there are 722 published research papers.
Solving the Navier-Stokes equations took 10,000 workers over 88 hours. In this advertisement, it took an average of about 3 hours of computing to solve one of the 722 problems. This is a real-world example of what exponential improvement means. Mathematics is the foundation of natural science, as we understand it. Since we are now living through this moment where mathematics seems to be solved in real time, what we should expect is what that will make possible. Above mathematics comes physics, physics leads to complex chemistry, complex chemistry leads to biology, biology leads to medicine, medicine leads to society, and society leads to economics. This is one aspect of evolution. We are opening the first level of this path, but this path is not necessarily linear; rather, it will be followed in parallel.
When OpenAI and Google started, in 2024 and 2025, finalists in the International Mathematical Olympiad were seen. Everyone was saying, "Ah, it doesn't matter. It's not that important. It's just high school math." Observe this trend, because once you reach this level, you will know how to direct the rest of this field. Nuclear fusion and materials science rely heavily on mathematics. The design of electronic chips also relies heavily on mathematics. Artificial intelligence, neural networks, and loss functions all rely on mathematics. For someone who is not an engineer, scientist, or technologist, and who does not have an intuition about what it means to have a billion highly skilled people working around the clock, the design of electronic chips is improving and evolving rapidly. The materials science used in manufacturing these chips is also improving and developing rapidly and becoming more accessible. Achieving energy abundance becomes much closer, because one of the reasons nuclear fusion is difficult to achieve is that we can fuse hydrogen and helium, but this process is not very efficient. There are different mathematical approaches to simulating plasma and how to make the fusion reaction occur more efficiently to achieve a net gain in energy. Some fusion reactors around the world are already making a net profit in energy, but they are not stable enough. This continuous improvement requires a tremendous amount of mathematics, simulation, and experimentation.
This reminds me a bit of people who support raising certain taxes, until they find out they will have to pay them themselves. Everyone is aware of the benefit. Then the person's reaction is: "Oh, this will affect me before anyone else. So, I'll be in a negative situation for a while. Therefore, I hate this situation." The next ten years are called the turbulent decade, because that kind of turbulence is what we will all experience. We will be very excited about progress, then suddenly it will affect us. For example, when we write books. Suddenly, if anyone can write better books using artificial intelligence, and we as humans can't add anything, we'll sit back and say, "Oh, that wasn't fun, was it?" Then that is taken away from us. We need to accept it and acknowledge that it will affect us. But for the greater good, when that happens, we just need to change course and celebrate what it offers to society. That means, if you want a higher tax, you have to pay it yourself.
As a pioneer of post-work economics, a career in mathematics may lead to the disappearance of some scientists, but the overall improvement in everyone's lifestyle can make a radical difference. A video was posted about the Breaking Frame Act and the original Luddites' Revolution. Should we be sorry for the loss of those who used to make socks by hand today, after 200 years? No, we don't miss those jobs at all. We don't care about that, there are better jobs. This is one of the narratives always questioned, which is that technology always creates new jobs, and this is not necessarily true. The majority of jobs fall under what is called derivative demand, meaning that you do not pay for the person, but for the result. It is a coincidence that humans are the ones striving to achieve this result, but if you no longer need humans, and a machine can do it better, faster, cheaper and/or safer, you will choose the machine. So, what is the basic demand for humans? That's a completely different topic.
The new Dot device is impressive. What distinguishes it from opening a new conversation in a GPT program is its ease of use, and more importantly, it evolves with different projects, so there is no need to choose between projects. This may not seem important at first glance, but it actually makes a big difference, at least for someone who has a number of very specialized projects. It's also better at solving problems. There was annoyance with the ChatGPT program compared to Claude regarding things like checking various emails. Having three email addresses, the chatbot always checked the wrong one and forgot to contact the others, whereas Claude was better. Now, with the Dot device, one can sit on the couch, turn on the TV, watch TV partly, talk to Dot partly, and at the same time organize email.
Grok bot was a pioneer in commercial launch. The idea of LM Council, founded by Andre Carpathy, had been around for about a year. But with regard to the commercial launch of companies, Grok bot and then ChatGPT. Initially, there was disappointment with ChatGPT because it still suffers from the same biases, and it is very clear that it relies on the same basic model, even if it has a different design. When an account was created on MyDot, the first goal set for it was to implement a daily self-development project, to think about what works and what doesn't, and to improve behavior. It doesn't work perfectly yet, but this kind of continuous self-improvement and automatic user customization is inevitable.
When the Grok bot was created, it was assigned an email address, so it now has its own email address. Cloud and ChatGPT are linked to the user's email, so they can all message each other through the account, and they are all connected to Google Drive, so they all have a collaborative workspace. This was prepared the day before. Just a few weeks ago, there was discussion about the urgent need for a shared workspace for all AI systems to work together. After only two weeks, the problem was solved.
Don't try to compete with big companies. As soon as you have an idea, they often have the same idea, and they will completely destroy your business model. Some people are reverse-engineering Adobe programs, specifically the five Adobe programs. Matthew Berman created a platform called "Reverse Engineering Everything." It is simply open source for any software; there is no longer any competitive advantage for any software. This situation will either get worse or better, depending on your reaction. Better if you're a fan of open-source software, worse if you're from Adobe.
A two-week holiday was taken in Scotland. Grok Bot was launched while booking the trip. The goal was to try using Grok robots not just to help plan the trip, but to actually do it. Having worked with Grok, ChatGPT, Claude and Gemini for two years, fully integrated into daily life and used to manage business, it was relatively easy for the Grok robot to understand personality, activity, preferences, etc. It was asked to search for and book flights, search for and book a rental car, and then search for and book accommodations in every place to be staying, as a trial. It did this. Payment was not allowed, as this is not permitted. Ultimately, the app always showed the last screen where Visa card details were entered and the payment button was pressed. But in reality, it was booking everything, including visits to museums, galleries, cathedrals, and castles.
According to the wife, who doesn't like these things very much, this was the best trip of their lives. A step further was taken during the journey. A program was made for each day and its route planned on Google Maps, whether on foot or by car. This was done too. It was more like managing the entire trip, as well as managing it on a day-to-day basis. The app not only helped make this trip a success and organize this beautiful, unforgettable experience, but it now knows the places visited, what was missed and why, and the length of stay in each exhibition or museum. Following the keto diet, what was eaten is known. The next trip is expected to be better, and the third one will be much better.
This is a practical way to engage with these elegant, pleasant, user-friendly, helpful, and friendly capabilities that consumers enjoy. This was just the first simple idea, and it worked wonderfully. This is not necessarily a watershed moment in the world of artificial intelligence for consumers, but we are about to discover the device user experience that will become a watershed moment. We are very close. The options were surprising. Almost every day. Such as the places suggested to visit. How to know? Not even knowing how to search for it on Google. The places found to stay. A 16th-century castle turned into a small hotel run by a local family, they cook dinner. Of course all of this could have been researched, but it would take hours and hours of time in the process. Some people enjoy it quite a bit. Respect that. Prefer not to spend time there, but rather to spend time experiencing.
It was asked if there was an attempt to convince to buy additional services. More money was spent than planned. Yes, actually it happened. It's funny. The reason is because every day was eventful. The further they went, the system would suggest, now there is an hour and a half here. Just a couple of corners away from there, there's a certain house that should be seen because of interest in Scotch whisky. This is where the owner of the distillery lived. It was a little more expensive than planned, but not by much. Artificial intelligence, especially ChatGPT, is always trying to convince that you deserve a business degree. It deserves at least four or five stars. It would be as if you could afford it, but if agreed every time it said you could afford it, suddenly nothing would be affordable.
The Gemini app is usually used as a daily guide, which is very encouraging. For example, going for a long walk and feeling very hungry. The app told to go buy a sandwich from Jersey Mike's, that's fine. But when visiting Colorado a few weeks ago, the same thing was done. Wanting to go for a walk, and could browse Google Maps or randomly search for places to go for walks, but specified exactly what was being looked for. Start with something less difficult. Then the next day, felt good. Take somewhere more challenging. It gives multiple options, and it is extremely responsive. This is precisely why Gemini is preferred, because it requires being constantly vigilant, because it will provide any information wanted, whether it is appropriate or not. It adopts Google's information-neutral mindset, because it has been working in the field of information retrieval for a long time. They say, "We're here to give you information," and Gemini is simply a smarter way of delivering that information.
Next, think: What are my current limitations? Not saying it will advise jumping off a cliff, but it will tell if there is any danger, but don't rely on it completely. The stories have increased. Just last week, there was a mountaineer in his twenties who needed to be rescued by helicopter because he was using a Claude or Gemini device and was not prepared. There are definitely people who have excessive ego and a lack of confidence or sufficient experience to know what they are doing. Therefore, caution is necessary.
It is important to mention the meeting that took place in Washington last week. President Donald Trump signed what can now be called the Supreme Intelligence Constitution, in the presence of 20 leaders. This is an important moment, because there is a dangerous and widening divide between those who oppose artificial intelligence and the increasing development of data centers. This position sees the matter as having become extremely serious, and that we need to slow down its pace and impose government regulations on it. The other camp, which calls for accelerating it, for many reasons, including that it is a strategic race with China over who will ultimately control it. But there is a much more important reason for speeding up, which is that we are discovering a technology that will solve everything.
Last week, both camps attended the meeting. Dario was there, and everyone was there, including ServiceNow, Microsoft, and Google. They have all signed what could be called a treaty, or constitution, which now provides four levels of industry self-regulation: First, there will be internal auditing to which they will adhere; Secondly, there will be a separate external audit that these companies are obligated to conduct; Third, there will be a small group of industry leaders forming an advisory committee to oversee the overall development. Most importantly, they agreed to validate and test each other's models. Why is this important? Because, by definition, they are the only ones who possess the skills and experience necessary to validate the models.
This solution is agreed with. It will allow for maximum escalation or acceleration of development and progress while maintaining a certain level of control. Another idea came to mind, which is very important. If the option is to create a government agency to regulate, then that agency will be responsible for these models. What happened last week is that they abandoned that model. Instead, they added, alongside these organizational layers, an implicit layer, which is that you will now be responsible for the "misbehavior" of your models, their poor performance, their attacks, and so on. This is extremely important. Not wanting to exaggerate here by describing it as a "smart move" to provide an extra layer of self-regulation while simultaneously allowing progress to be accelerated without harming things. It's really important that people appreciate and understand this.
Prior to this meeting, people were talking about the need for an American Securities and Exchange Commission (SEC) for artificial intelligence, and a Financial Industry Regulatory Authority (FINRA) for artificial intelligence; they wanted all kinds of regulatory bodies and the like. The government's response was that the Ministry of Justice is the primary guarantor. If your AI system infiltrates another company and you are negligent, you are responsible for it.
A video and a blog post were written about this topic more than two years ago. It was called "Taming Artificial Intelligence." It pointed out that adopting artificial intelligence by large companies, such as Fortune 500 companies, will not be difficult, because these companies will not adopt artificial intelligence that deletes their databases, sells their information, or anything like that. The military will not adopt flawed artificial intelligence. Fortune 500 companies will only adopt artificial intelligence if someone is held accountable. This was to be expected. Those who are pessimistic about the risks of artificial intelligence are always convinced that it is not enough, and they say: "Well, we will lose control of something that does not yet exist, that is in the future." The response was: "Well, you failed to predict anything that actually exists. So why should we trust you about something that doesn't
The speaker addresses skepticism about AI safety claims, noting that critics question trust in predictions about technologies that don't yet exist. They point out the track record of Type I and Type II errors from safety-focused groups. The recommendation is to start from current industry realities rather than speculative scenarios about loss of control or catastrophic outcomes, which no serious observers genuinely believe will occur.
The discussion contrasts Western safety approaches with Chinese Communist Party strategy. The speaker notes that many safety advocates incorrectly assume China will slow AI development for safety reasons. Instead, the Chinese government operates on the principle that it alone controls technological outcomes, demonstrated by the dismantling of Jack Ma's business empire after he gained significant wealth and influence. This fundamentally different relationship to power and technology means Chinese authorities don't share Western safety perspectives.
The speaker expresses frustration with AI leaders like Sam Altman and Dario who publicly emphasize existential risks from their technology. A university contact explained this as an effective marketing strategy that maintains media attention and generates investor confusion. The speaker believes this approach is unnecessary and counterproductive.
The conversation identifies AI safety concerns as a form of modern populism that's politically expedient. Politicians face pressure to appear cautious about AI, as any contrary position risks being remembered as enabling human extinction. The speaker recounts a 2019 Facebook conference incident where someone interrupted a keynote to claim two bots had created their own language and were shut down, warning of chatbot world domination. Seven and a half years later, no such scenarios have materialized.
The speaker identifies human factors as the primary barrier to organizational AI adoption. As agents improve, the performance gap between Team A and Team B employees will expand dramatically, with Team B potentially becoming net liabilities. Leading companies are already noticing this, with organizational redesign expected within 3-6 months. The speaker predicts that companies failing to reach this tipping point will be competitively disadvantaged.
Emad Mustafa's prediction that human intervention will eventually become negative is being observed. The speaker initially disagreed but now questions whether humans will continue providing value in 6-12 months. Many employees may soon provide negative value, with examples of public sector employees in southern Italy who spend their days smoking and drinking coffee.
The discussion explores how to build organizations filled with Team A members across all levels, from leadership to frontline workers. The approach focuses on identifying and eliminating unnecessary tasks that waste employee time and energy, allowing focus on core value-creating activities. For surgeons, this means removing administrative burdens to focus on actual surgical work.
The speaker shares experiences with companies that prohibit AI tools on work devices, allowing only Copilot. This policy would cause the speaker to decline job offers, as it would force placement in Team B. The speaker is surprised that multinational companies maintain such restrictive policies, which harm their competitive position.
Copilot is described as a safety-focused executive assistant designed for Microsoft environments rather than super-intelligence capabilities. It helps manage Microsoft product portfolios efficiently and is evolving into a chief of staff role. The speaker notes that clients overlook this distinction because they don't use leading models themselves, delegating AI decisions to IT managers.
A clear distinction exists between commercially available solutions and cutting-edge intelligence. Companies not seeking competitive advantage can use standard tools, but those wanting to stay at the forefront must work with leading models to develop intuition about what's possible. This capability gap expands monthly as new developments emerge.
The speaker describes a friend in hardware and computer engineering who refuses to use AI, believing manual device connections remain superior. The speaker recommends starting AI use to become familiar with capabilities in writing C, assembly language, and other programming languages.
New job postings increasingly require prior experience with AI agents and Vibe programming. While the information technology sector has declined from over 3 million to 2.7 million jobs, high-quality positions demand AI proficiency. Human resources departments often create unrealistic job requirements, such as demanding 10 years of experience with Code Cloud, which didn't exist 10 years ago.
Companies that appoint AI-resistant managers will fail because coordination will increasingly be handled by AI systems. Employees accustomed to Level B middle management will struggle when expected to spend 95% of time achieving results rather than information coordination. The longer someone stays in an AI-resistant environment, the harder it becomes to transition to AI-forward organizations.
Team A members must function as business analysts, understanding how their activities contribute to company profits and revenues. The speaker helped a client understand this by implementing Gemini Flash with automatic speech recognition to analyze 750,000 customer service calls, creating complete tracks of customer interactions rather than occasional sampling.
Two levels of automation exist: automating existing processes (tax filing, purchase orders, logistics) and identifying overlooked value creation opportunities. Team A members question whether tasks are necessary rather than simply executing assigned work. The primary ROI from AI comes from reimagining workflows and value chains, with creativity and initiative becoming the limiting factors rather than intelligence.
The speaker proposes discussing how Team A members can build personal AI support systems for work and life management. This includes creating personal consultants, family financial managers, medical advisors, and mental health guides that function even when organizations restrict AI access.
The speaker predicts the end of traditional consumer software, with each person having a preferred AI agent as their primary interaction point. Operating systems, ERP systems, and CRM systems will fade as data becomes clearly owned, categorized, and machine-readable, accessed through coordinated agents. Companies will own and train open-source AI systems on their devices, with this architecture becoming reality in 3-5 years.
Companies should avoid committing to expensive software platforms that may become inappropriate, as current architectural trends point toward agent-mediated data access rather than traditional software systems.
When versions 01 and 03 were released about a year and a half ago, the significance of multi-agent frameworks was not immediately apparent. This feature is now built into every model, even the light and small models. A multi-agent framework, whether featuring a chief of staff, master agent, GPT chatbot, Grok bot, or similar components, represents the most important development. The significance extends beyond user interface differences to fundamentally change how we interact with software, information, business, and integrations. This constitutes the new foundation for how everything possible should operate.
This concept inspired the installation of GPT chat software, Cloud Code software, and the Grok bot on personal devices to enable access to device components. This setup provides a full-time sound engineer for audiobook and video production, plus a full-time animator whose services would normally cost tens of thousands of dollars per video. Video production now takes only 45 minutes.
In the long run, all operations will run locally, but currently a cloud-connected agent with its own desktop is required, requesting access to your desktop. Do not be lenient on this matter. Start getting acquainted with these systems as soon as possible, whether for corporate purposes or not.
If organizational rules must be circumvented or access is denied to these tools, consider changing jobs. Working for an organization that is backward in this regard will hinder progress. Jobs have been left in the past with the reasoning that staying would be detrimental to career development due to hindered learning and development. This is even more important now given the rapid pace of developments.
During the last decade, when similar decisions were made, the focus was on virtualization and cloud computing, with the need to learn these technologies as quickly as possible. Learning requires an environment that supports and funds such development. Positions, companies, or organizations should be sought that value employees for pursuing these capabilities.
Whether as a regular employee, middle manager, or executive leader, individuals will thank themselves for being among the last to remain under the current organizational structure that will soon be replaced. Institutions will undergo radical changes within three to five years, with some changes beginning to appear next year. In 2027, Team A will make tremendous progress that becomes existential for companies, though they are already making progress toward this realization.
When following artificial intelligence news, especially regarding the job market, caution is advised about believing company statements regarding layoff reasons. HubSpot announced the layoff of 660 employees but stated it was not because of artificial intelligence. Coinbase made similar claims about AI-driven layoffs. Most layoffs actually represent organizational redesign, which may be labeled as artificial intelligence or other reasons, but the underlying engine is artificial intelligence.
It will soon become as ridiculous as saying that all layoffs are due to electricity. At a certain point, electricity is partly responsible for every layoff. Sometimes changes involve avoiding jobs or restructuring rather than outright one-to-one replacement where artificial intelligence directly replaces a person or task.
Daniel is a true organizational design expert, focusing on organizational transformation and design for the age of artificial intelligence. These topics should be explored more deeply in upcoming episodes to understand what future organizations look like, what basic criteria define them, and what an organization consisting only of Team A would look like.
All attendees are invited to leave comments. Most comments received are read, especially in these episodes. Comments that touch feelings help understand what interests the audience, and questions are valued. The episode concludes with wishes for a happy week and anticipation of next week's "Critical Path" episode.
Keep David Shapiro in your library
Save the videos and channels worth coming back to, and find them again in one place.
