VN — VnExpress
Anthropic: From AI Lab to a Trillion-Dollar Empire
AI2 min read
This story tells how Anthropic, a leading AI company, was started in 2021 by a group of former OpenAI staff, led by the Amodei siblings. It also looks at how the company's safety mission now faces hard tests after its IPO (the first sale of its shares to the public).
Two siblings and a startup idea
In 2016, Dario and Daniela Amodei were choosing their next steps in San Francisco. Dario had a PhD in biophysics and was working at Google. Daniela had just finished school and quickly became a director at the payments company Stripe. Greg Brockman, a co-founder of OpenAI, was a friend of Daniela. He first tried to recruit Dario in 2015 and failed, but succeeded in 2016. Daniela joined the AI startup two years later.
The split from OpenAI
Dario worked on AI safety and on scaling laws, which describe how an AI model's performance changes as its size, training data and computing power grow. Daniela became a vice president for safety and policy. In 2017, Brockman wanted OpenAI to sell superintelligent AI, called AGI, to governments. Dario objected and thought about quitting. Tensions grew in 2018 and 2019. In late 2020, Dario asked to leave and start his own company. In December 2020, he and five colleagues left OpenAI. In January 2021, Anthropic was founded as an AI research and development company that puts safety at the center.
Key facts
- Anthropic was founded with 7 former OpenAI staff.
- The company was founded in January 2021, one month after the group left OpenAI in December 2020.
- Dario and Daniela Amodei left with five other colleagues: Jared Kaplan, Chris Olah, Sam McCandlish, Tom Brown and Jack Clark.
- The article describes the company as a trillion-dollar empire, a value it is said to have reached after its IPO.
A safety-first goal
Anthropic says it wants to build AI that is reliable, interpretable and steerable. Reliable means giving stable results. Interpretable means understanding how a model reaches its decisions. Steerable means guiding a model toward the results people want. To reach these goals, the company studies mechanistic interpretability, which looks at what happens inside a model as it handles information and writes answers.
Why it matters
Anthropic was built on the promise of careful, safe AI. Now that it is a huge company after its IPO, it faces pressure to grow and earn money. That makes its safety promise a real test. The story also shows how disagreements over direction can split an AI company and create a new rival.
Glossary
- Anthropic
- An AI company that builds AI systems and focuses on safety.
- OpenAI
- An AI company that makes the GPT models, including ChatGPT.
- IPO
- Initial public offering, the first time a company sells its shares to the public.
- AGI
- Artificial general intelligence, AI that could do most human thinking tasks well.