The rebel AI team has set a record after learning to automatically learn

Get a fugitive Artificial intelligence researchers have secured the first funding record for a new start-up involved in general-purpose AI, marking the latest attempt to create an organization that will ensure the security of the most powerful technology of the time.

The team has split with OpenAI, an organization founded in 2015 with the support of Elon Musk, to ensure that super-intelligent AI systems don’t one day work and harm their makers. Schism continued the differences in the direction of the group in 2019 when Microsoft took a $ 1 billion investment, according to two people who knew about the split.

The new Anthropic company is Dario Amodei, one of the founders of OpenAI and former head of AI security at the organization. It raised $ 124 million in the first round of funding. That’s according to research firm PitchBook, for a group of AI trying to build generally applicable AI technology, rather than setting it up to apply the technology to a particular industry. Based on data presented by the company, the round has a value of $ 845 million in Anthropics.

The investment is led by Jaan Tallinn, an Estonian computer scientist behind Skype. Eric Schmidt, former CEO of Google, and Dustin Moskovitz, one of the founders of Facebook, have also been in favor of the project.

The break with OpenAI began when Amodei launched in December, and has since gathered about 30 researchers, according to an estimate. Among them are Amodei’s sister, Daniela Amodei, president of Anthropic, and a team of researchers who worked on GPA-3 at OpenAI. progress an automatic language system, including Jared Kaplan, Amanda Askell, Tom Henighan, Jack Clark and Sam McCandlish.

OpenAI changed course two years ago, when it sought the support of Microsoft to fuel the growing hunger for computer resources to power deep learning systems. In return, he promised the software company the first rights to market its findings.

“They started with a non-profit, with the goal of democratizing AI,” said Oren Etzioni, head of the AI ​​institute created by Microsoft founder Paul Allen. “Obviously you have to create a return when you get $ 1 billion. I think their career has become more corporate.”

OpenAI has sought to isolate its AI security research from new commercial operations by limiting Microsoft’s presence on its board. However, this still caused internal tensions about the direction and priorities of the organization, according to a person who knew the escape group.

OpenAI did not comment on whether the disagreement over the direction of the investigation led to the split, but said it made internal changes to integrate the work on research and security more closely when Amodei left.

Microsoft won exclusive rights to touch on the findings of the OpenAI research Committing $ 1 billion to protect the team, largely as a technology, to support in-depth computer-intensive learning systems, including GPT-3. Earlier this week, Microsoft said it had incorporated the language system into some of its software creation tools to help people with no coding skills create their own applications.

The rush to market quickly GPT-3 It comes in comparison to the OpenAI management of a previous version of the technology developed in 2019. The group initially said it would not provide any technical details on the progress due to concerns about the potential misuse of the powerful language system, although it later reversed the trajectory.

To isolate itself from commercial interference, Anthropic has registered it as a public benefit corporation or “B corp” with special governance arrangements to support the role of “advanced AI in the responsible development and maintenance of humanity”. These include the creation of a long-term benefit committee made up of people unrelated to the company or sponsors, who will give the final say on issues such as the composition of its committee.

Anthropic said his work will focus on “large-scale AI models,” including making systems easier to interpret and “building ways to better integrate human opinion into the development and deployment of those systems.”

Daily newsletter

# tFTFT provides you with news, comments and analysis on the major companies, technologies and issues that make up the fastest moving sectors in the hands of specialists from around the world. Click here #techFT to access your inbox.