How valuable is the title of "Former OpenAI Employee" in the market?

On February 25th local time, according to Business Insider, the new company Thinking Machines Lab, former chief technology officer of OpenAI, is launching a $1 billion financing with a valuation of $9 billion.

At present, Thinking Machines Lab has not disclosed any timetables or specific details of the product or technology. The company's public information is only a team of former OpenAI employees of more than 20 people, and their vision: to build a future where "everyone has access to knowledge and tools to enable AI to serve people's unique needs and goals."

Mira Murati and Thinking Machines Lab

The capital appeal of OpenAI entrepreneurs has formed a "snowball effect". Before Murati, SSI, founded by Ilya Sutskever, former chief scientist of OpenAI, had already obtained a valuation of US$30 billion based on OpenAI genes and one concept.

Since Musk withdraws from OpenAI in 2018, former OpenAI employees have founded more than 30 new companies with a total financing of more than $9 billion. These companies have formed a complete ecological chain covering AI security (Anthropic), infrastructure (xAI), and vertical applications (Perplexity).

This reminds people of the wave of Silicon Valley entrepreneurship after PayPal was acquired by eBay in 2002, and the founders such as Musk and Peter Thiel left, forming a wave of Silicon Valley entrepreneurship - the "PayPal Gang", which has emerged legendary companies such as Tesla, LinkedIn, and YouTube. OpenAI's runaway employees are also forming their "OpenAI Gang".

But the script of "OpenAI Gang" is more radical: "PayPal Gang" has created two companies with a 10 billion yuan in 10 years, and "OpenAI Gang" has spawned five companies with a 10 billion yuan valuation in just two years after the launch of ChatGPT. Among them, Anthropic is valued at US$61.5 billion, Ilya Sutskever's SSI is valued at US$30 billion, and Musk's xAI valuation is valued at US$24 billion. In the next three years, a hundred billion unicorn is likely to be born in the "OpenAI Gang".

The new round of Silicon Valley "talent fission" initiated by the "OpenAI Gang" has affected the entire Silicon Valley and even reshapes the power map of global AI.

OpenA's 11 co-founders are currently onlySam Altman and Wojciech Zaremba, head of the language and code generation team, are still on the job.

2024 is the peak of OpenAI's resignation. During this year, Ilya Sutskever (resigned in May 2024) and John Schulman (resigned in August 2024) have successively resigned. The OpenAI security team has shrunk from 30 to 16, with a 47% reduction; key figures such as Chief Technology Officer Mira Murati and Chief Research Officer Bob McGrew among the executives have resigned one after another; in the technical team, core technical talents such as Alec Radford, chief designer of GPT series, and Tim Brooks, head of Sora (joining Google), etc.; deep learning expert Ian Goodfellow joined Google, and Andrej Karpathy founded an education company after leaving for the second time.

"Gathering is a ball of fire, dispersing is a star full of sky."

Among the core technical backbones who joined OpenAI before 2018, more than 45% chose to set up new portals. These new "portals" also dismantled OpenAI's technology gene bank and reorganized three major strategic groups.

First of all, there are the "direct troops" that continue the OpenAI gene. They can be said to be a group of OpenAI 2.0 ambitious people.

Mira Murati's Thinking Machines Lab almost completely transplants OpenAI's R&D architecture: John Schulman is responsible for the reinforcement learning framework, Lilian Weng leads AI security systems, and even GPT-4's neural architecture diagrams are directly used as technical blueprints for new projects.

Their "Open Science Manifesto" directly points to the closed trend of OpenAI in recent years, and plans to create a "more transparent AGI R&D path" through the continuous disclosure of technical blogs, papers and code. This has also triggered some chain reactions in the AI ​​industry: three top researchers at Google DeepMind joined with the Transformer-XL architecture.

Ilya Sutskever's Safe Superintelligence Inc. (SSI) took another path. Sutskever co-founded the company with two other researchers, Daniel Gross and Daniel Levy. They gave up all short-term commercialization goals and focused on building "irreversible security super intelligence" - a technology framework that is almost philosophical. The company has just been established, and institutions such as a16z, Sequoia Capital decided to invest $1 billion to "pay the bill" for Sutskever's ideal.

Ilya Sutskever and SSI

The other faction is the "Superor" who has left ChatGPT before.

Anthropic, founded by Dario Amodei, has evolved from "OpenAI opposition" to its most dangerous competitor. Its Claude 3 series model is no match for GPT-4 in multiple tests. In addition, Anthropic has established an exclusive cooperation with Amazon AWS, which means that Anthropic is gradually eroding the foundation of OpenAI in terms of computing power. The chip technology jointly developed by Anthropic and AWS may further weaken OpenAI's bargaining power in Nvidia's GPU procurement.

Another representative of this sect is Musk. Although Musk left OpenAI in 2018, some of the founding members of xAI founded have worked at OpenAI, including Igor Babuschkin and Kyle Kosic, who later returned to OpenAI. Thanks to Musk's powerful resources, xAI poses a threat to OpenAI in many aspects such as talent, data, and computing power. By integrating the real-time social data flow of Musk's X platform, xAI's Grok-3 can instantly capture hot events on the X platform to generate answers, while ChatGPT's training data is as of 2023, with a significant gap in timeliness. This closed-loop of data is difficult for OpenAI to replicate on the Microsoft ecosystem.

However, Musk's positioning of xAI is not a disruptor of OpenAI, but rather a desire to regain the original intention of "OpenAI". xAI adheres to the "maximum open source" strategy, such as the Grok-1 model open sourced with the Apache 2.0 protocol, attracting global developers to participate in ecological construction. This is in sharp contrast to OpenAI's recent tendency to close source (such as GPT-4 only provides API services).

The third faction is some "breakers" who reconstruct industrial logic.

Perplexity, founded by former OpenAI research scientist Aravind Srinivas, is one of the first companies to transform search engines with AI models. Perplexity replaces the search page's link list through AI-powered answer generation. Today, more than 20 million searches per day, and the financing scale is more than $500 million (valued at $9 billion).

Adept founder is David Luan, former vice president of engineering at OpenAI, who has been involved in technical research on language, supercomputing, reinforcement learning, as well as security and policy development for GPT-2, GPT-3, CLIP and DALL-E projects. Adept focuses on developing AI Agents, with the goal of helping users automate complex tasks (such as generating compliance reports, design drawings, etc.) through large models combined with tool calling capabilities. The ACT-1 model it developed can directly operate office software, Photoshop, etc. The company's core founding team currently includes David Luan has switched to Amazon's AGI team.

Covariant is an embodied smart startup with a valuation of $1 billion. Its founding teams are all from the robot team disbanded by OpenAI. The technical genes are derived from the experience of GPT model research and development. We focus on developing basic robot models, with the goal of realizing autonomous robot operations through multimodal AI, especially focusing on warehousing and logistics automation. However, currently, Pieter Abbeel, Peter Chen and Rocky Duan, the three "OpenAI Gang" members of Covariant's core founding team, have joined Amazon.

Some "OpenAI helps "startups

Source: Public information, compiled: Flagship AI technology's leap from "tool attributes" to "productivity factor" has given birth to three types of industrial opportunities: alternative scenarios (such as subverting traditional search engines), incremental scenarios (such as intelligent transformation of manufacturing), and reconfiguration scenarios (such as underlying breakthroughs in life sciences). The common characteristics of these scenarios are: a grayscale space with the potential to build data flywheel (user interaction data feeding back model), deep interaction with the physical world (robot action data/biological experimental data), and ethical supervision.

The technology spillover of OpenAI is providing the underlying impetus for this industrial transformation. Its early open source strategies (such as part of GPT-2 open source) formed the "dandelion effect" of technological diffusion, but when the technological breakthrough entered the deep water zone, closed source commercialization became an inevitable choice.

This contradiction has spawned two phenomena: on the one hand, resigned talents transfer Transformer architecture, reinforcement learning and other technologies to vertical scenarios (such as manufacturing and biotechnology), and build barriers through scenario data; on the other hand, giants achieve technical positioning through talent acquisition, forming a closed loop of "technical harvesting".

The "OpenAI Gang" is making rapid progress, but the old boss OpenAI is "stubborn".

In terms of technology and products, the release date of GPT-5 has been repeatedly postponed, while the mainstream ChatGPT products are generally believed by the market to be unable to keep up with the development of the industry.

In the market, latecomer DeepSeek has begun to gradually catch up with OpenAI. Its model performance is close to ChatGPT but the training cost is only 5% of GPT-4. This low-cost reproduction path is breaking down the technical barriers of OpenAI.

However, a large part of the rapid growth of the "OpenAI Gang" is due to the internal contradictions of OpenAI companies.

The core research team of OpenAI can be said to have fallen apart, with only Sam Al left in the 11 co-foundersTman and Wojciech Zaremba are on the job, and 45% of core researchers have been out of service.

Wojciech Zaremba

Co-founder Ilya Sutskever left to start SSI, chief scientist Andrej Karpathy publicly shared his Transformer optimization experience, and Tim Brooks, head of Sora video generation project, switched to Google DeepMind. In the technical team, more than half of the authors of early GPT versions have left, and most of them have joined the ranks of OpenAI competitors.

At the same time, according to data compiled by Lightcast, which tracks recruitment information, OpenAI's own recruitment focus seems to have changed. In 2021, 23% of the company’s recruitment information is general research positions. In 2024, general research only accounted for 4.4% of its recruitment information, which also indirectly reflects that the position of scientific research talents in OpenAI is changing.

The organizational cultural conflicts brought about by commercial transformation are becoming increasingly obvious. While the employee size has expanded by 225% in three years, the early hacker spirit has gradually been replaced by the KPI system, and some researchers bluntly said that they are "forced to shift from exploratory research to product iteration."

This strategic swing has led OpenAI to a double dilemma: it requires continuous output breakthrough technologies to maintain valuation, and it has to face the competitive pressure of former employees to use their methodology to quickly replicate their results.

The winner of the AI ​​industry is not about the breakthrough of the parameters of the laboratory, but about who can inject technical genes into the industrial capillaries - reconstructing the underlying logic of the business world in the answer flow of search engines, the motion trajectory of robotic arms, and the molecular dynamics of biological cells.

The rapid rise of the "OpenAI Gang" and "PayPal Gang" is largely due to the "blessing" of California law.

California has been a catalyst for innovation in Silicon Valley since its legislation prohibits the competition in 1872. According to Section 16600 of the California Business and Career Code, any provision that restricts professional freedom is invalid, and this institutional design directly promotes the free flow of technical talents.

The average service cycle of programmers in Silicon Valley is only 3-5 years, far lower than that of other technology centers. This high-frequency flow has formed a "knowledge spillover" effect - taking Fairchild Semiconductor as an example, its resigned employees founded 12 semiconductor giants such as Intel and AMD, laying the industrial foundation of Silicon Valley.

The law prohibiting the competition agreement seems to be insufficiently protected by innovative companies, but in fact it promotes innovation even more. The flow of technicians has accelerated the spread of technology and lowered the threshold for innovation.

2024 US Federal Trade CommissionThe Council (FTC) predicts that after the full ban on competition agreement in April 2024, the US's innovation vitality will be further released. 8,500 new companies may be added in the first year of the policy implementation, the number of patents surged by 17,000-29,000, and 3,000-5,000 new patents. In the next 10 years, the annual patent growth rate will be 11-19%.

Capital is also an important driving force for the rise of OpenAI Gang.

Silicon Valley venture capital accounts for more than 30% of the United States. Sequoia Capital, Kleineng Huaying and other institutions have built a complete financing chain from seed turn to IPO. This capital-intensive model has given rise to a dual effect.

First of all, capital is the engine that drives innovation. Angel investors provide not only funds, but also industry resource integration. Uber was founded with only $200,000 from the two founders, and only 3 registered taxis were available. After receiving $1.25 million in angel investment, it started rapid financing and its valuation reached $40 billion by 2015.

The long-term attention of venture capital to the technology industry has also promoted the upgrading of the technology industry. Sequoia Capital invested in Apple in 1978 and invested in Oracle in 1984, laying its influence in the fields of semiconductors and computers; in 2020, it began to deeply deploy artificial intelligence and participate in cutting-edge projects such as OpenAI. The 10 billion dollar investment of international capital (such as Microsoft) in AI has prompted the commercialization cycle of generative AI technology to be shortened from several years to several months.

Capital also provides innovative companies with higher fault tolerance. The speed of screening failed projects by accelerator is as important as successful projects. According to startup analysis agency startuptalky, statistics, the failure rate of startups worldwide is 90%, and the failure rate of startups in Silicon Valley is 83%. Although startups are not easy to succeed, in the venture capital investment grid, failure experience can be quickly converted into nutrients for new projects.

Image source: startuptalky.com

However, capital has also changed the development path of these innovative companies to a certain extent.

The top AI projects received a valuation of over one billion US dollars before they were released, which in disguise has led to the difficulty of obtaining resources of other small and medium-sized innovation teams. This structural imbalance is more prominent in the regional distribution. The survey results of database management company Dealroom show that the venture capital (US$24.7 billion) obtained by the US Bay Area in a single quarter is equivalent to the sum of the 2nd-5th venture capital centers in the world (London, Beijing, Bangalore, Berlin). At the same time, although emerging markets such as India have achieved a 133% financing growth, 97% of funds flow to "unicorn" companies with valuations of over US$1 billion.

In addition, capital has a strong "path dependence", and capital prefers fields with quantifiable returns, which has also made it difficult for many emerging basic science innovations to be strongly supported at the capital level. For example, in the field of quantum computing, Guo Guo, founder of Yuanyuan Quantum, a domestic quantum computing startup.Ping, in the early stage of starting a business, sold a house to start a business due to insufficient funds. Guo Guoping's first financing was actually achieved in 2015. Data released by the Ministry of Science and Technology showed that my country's total investment in scientific research was less than 2.2% of GDP, of which basic research funds accounted for only 4.7% of R&D investment.

Not only lack of support, big capital is also locking in top talents through the temptation of "money", which has basically locked the salary of CTO-level positions of startups in seven figures (US companies are US dollars, while Chinese companies are RMB), forming a cycle of "giants monopoly talents - capital chases giants".

However, there are certain risks in the significant advancement of these "OpenAI Gangs".

Mira Murati and Ilya Sutskever both received billions of dollars in financing with just one idea. This all comes from their trust premium for the technical capabilities of OpenAI's top teams, but this trust also has risks - whether AI technology can be in the exponential growth stage for a long time, followed by vertical scene data that can form a monopoly barrier. When these two risks encounter real challenges (such as slowing breakthroughs in multimodal models and surge in industry data acquisition costs), capital overheating may trigger an industry reshuffle.