Anthropic
Anthropic · co-founder Dario Amodei
Build AI systems that are safe, and steer the whole field toward safety while doing it.
Why it exists
Anthropic exists because a group of people at OpenAI concluded they could not steer the thing from inside it. In early 2021 Dario and Daniela Amodei left with most of the team behind GPT-3 and started a lab on an uncomfortable premise: the safest way to influence frontier AI is to build frontier AI yourself, because a safety lab with no model has no leverage and no seat at the table.
The distinctive method is Constitutional AI — instead of only having humans rate outputs, the model critiques its own answers against a written set of principles. That, plus the Responsible Scaling Policy published in 2023, put in writing what capabilities would make the company stop and add safeguards. It is the only titan here that has publicly committed to conditions under which it would slow down.
Which also explains its manner in our council. Anthropic gives the most hedged dates, flags what it does not know, and is usually the one arguing that the hard part is not capability but control.
The company
- Founded
- 2021
- Headquarters
- San Francisco, USA
- Founders
- Dario Amodei, Daniela Amodei, Tom Brown, Jared Kaplan, Sam McCandlish, Jack Clark, Chris Olah
- Structure
- Public benefit corporation (Delaware PBC)
- Stated mission
- Build AI systems that are safe, and steer the whole field toward safety while doing it.
Go to the source
Only values the vendor published are plotted. Where a model has no stated context window, there is no point on the line — nothing is estimated.
Only values the vendor published are plotted. Where a model has no stated context window, there is no point on the line — nothing is estimated.