Sibling Founders' Story Behind Claude3: Resignation from Baidu Then OpenAI
Background
- Company: Anthropic
- Founded: February 2021
- Founding Team:
- Founder and CEO: Dario Amodei
- President: Daniela Amodei
- Core Product: Claude series models
- Financing History:
- May 2021: Raised $124 million in financing, led by Skype co-founder Jaan Tallinn (Series A);
- April 2022: Raised $580 million in financing, led by FTX CEO Sam Bankman-Fried (Series B);
- February 2023: Received a $300 million investment from Google, giving Google a 10% stake;
- May 2023: Raised $450 million in financing, led by Spark Capital (Series C);
- August 2023: Received a $100 million investment from South Korea's largest telecommunications company, SK, followed by a $4 billion investment from Amazon and a $2 billion investment from Google;
- February 2024: Raised $750 million in investment from venture capital firm Menlo Ventures.
- Valuation: $18.4 billion (as of December 2023)
While OpenAI and Musk's "feud" continues, competitor Anthropic quietly released the Claude3 series model, claiming it to be the fastest and most outstanding AI model currently on the market. Netizens commented: The world's most powerful model changed hands overnight, marking the end of the GPT-4 era!
So, what kind of people are behind this company that aims to surpass GPT-4?
Halfway Into a Career: Technological Idealists
The cheerful siblings in the photo are Dario Amodei and Daniela Amodei. They are the founders of Anthropic, with Dario serving as CEO and Daniela as President.
However, originally, their life scripts had nothing to do with artificial intelligence.
Born in Italy in 1983 to an American-Italian family, Dario grew up in the United States. He later attended Stanford University, where he studied physics.
"I just wanted to understand the universe. Artificial intelligence was not within my consideration at all; it seemed like science fiction," Dario recalled in an interview.
As he was about to graduate, Dario began to study Moore's Law and read Ray Kurzweil's works, vaguely feeling that AI might indeed develop, but at that time, he did not know how to proceed.
So, he made a bold decision to pursue a graduate degree in neuroscience. "Because that was the closest thing to intelligence that I could really study."
It wasn't until after his graduate studies that Dario began to get involved with AI through AlexNet.
In 2014, as a postdoctoral researcher at Stanford University, Dario's first job outside academia was at Baidu's Silicon Valley AI Lab, working under AI luminary Andrew Ng.
However, Dario left Baidu in less than a year to join Google as a senior researcher.
During his time at Google, Dario co-authored a paper on concrete problems in AI safety, discussing the inherent unpredictability of neural networks. This ignited his unwavering focus on AI safety.
Daniela, his sister, is a typical idealist. Her early career also had nothing to do with artificial intelligence.
She graduated summa cum laude from the University of California, Santa Cruz, with degrees in English Literature, Politics, and Music.
Afterward, she worked in several NGOs in Washington, D.C., focusing on global issues such as poverty, corruption, and the environment.
However, she soon realized that working in these "benevolent institutions" did not have a real impact on these issues. Thus, she ventured into the technology sector.
"I believe the tech sector is one of the few areas where an individual's work can have a very high impact," Daniela said.
Daniela joined a financial technology company called Stripe, where she met Greg Brockman, who served as CTO at Stripe.
As the deep learning boom arrived, Daniela saw AI as the best way to tackle scientific and health issues that humanity could not yet solve.
In 2015, OpenAI was founded, aiming to be an open-source, nonprofit organization focused on building safe AGI for humanity.
The Dario siblings found this highly attractive.
Coincidentally, Brockman, who was CTO at Stripe, was recruited into OpenAI's founding team.
Brockman first brought Dario from Google to OpenAI, where he was in charge of AI safety. Dario quickly rose through the ranks, serving as head of the AI Safety Research Group, Director of Research, and Vice President of Research.
Daniela also joined, taking the position of Vice President of Safety and Policy at OpenAI.
The siblings enjoyed the purest technological pleasure at this nonprofit company.
OpenAI "Defectors"?
As is well known, AI research is costly.
In 2019, facing significant financial pressure, OpenAI announced its reorganization from a nonprofit to a "limited profit" company, securing a $1 billion investment from Microsoft in exchange for licensing its technology to Microsoft.
No longer worried about funding, OpenAI quickly released GPT-2 and GPT-3. However, Dario believed that AI safety, controllability, and interpretability were more important than blindly iterating on models larger than GPT-3.
Disappointed by the gradual commercialization and departure from the founding vision, as OpenAI handed GPT-3 over to Microsoft without resolving AI safety issues, Dario and a group of technological idealists decided to start anew. In February 2021, the siblings, along with five other former OpenAI employees, founded Anthropic.
Anthropic, meaning "related to humans," is committed to "building useful, honest, and harmless systems" and has established a dedicated AI safety team to reduce risks such as misinformation, biosafety abuse, election interference, and privacy issues.
They introduced a technique called "Constitutional AI (CAI)" in Claude's training, where researchers define principles aligned with human decency to guide the system's behavior, such as not generating content that threatens physical safety, violates privacy, or causes harm. These principles include the Universal Declaration of Human Rights, parts of Apple's service terms, and Anthropic's AI research lab principles, ensuring the model is guided by the right values.
Despite Anthropic's recent establishment, Claude was developed earlier than ChatGPT.
Why didn't Anthropic release Claude first, allowing OpenAI to take the lead with ChatGPT? Reports suggest the siblings were concerned about safety issues and believed the model needed to be secure before being released to the public, giving ChatGPT and OpenAI the chance to gain global attention first.
Unlike Altman, who is keen on marketing, the Amodei siblings tend to keep a low profile, rarely giving interviews. Even with the release of the performance-exploding Claude3, they only quietly tweeted a few times on social media.
However, Google and Amazon have heavily invested in this startup. In February last year, Google invested $300 million for a 10% stake; in August, Amazon invested $4 billion, followed by a $2 billion investment from Google.
Even so, Anthropic's relationship with Google is limited to technical support and funding, and the two remain competitors in the large model domain.
No comments