- Dario Amodei left OpenAI due to differences in AI vision and approach
- He co-founded Anthropic to pursue his own AI development principles
- Amodei believed scaling AI models greatly enhances their performance
Differences over the vision and approach to artificial intelligence: those were the primary reasons Dario Amodei, co-founder of AI firm Anthropic, left OpenAI.
On investor and entrepreneur Nikhil Kamath's podcast ‘WTF Is', Amodei said he joined OpenAI just a few months after it was founded and led the company's research for several years.
Over time, he and a few colleagues developed their own ideas about how AI should be built and what principles the company should follow.
"Eventually, a few other employees and I just kind of had our own revision for how we wanted to make AI and what we wanted the company to stand for. And so we went off and founded Anthropic," he said.
When Nikhil asked if Anthropic was like a "fork" from OpenAI, Amodei said that there were two main reasons. First, he and his co-founders had strong beliefs about how AI should be developed. Second, while they were starting to convince some people at OpenAI, he didn't feel they fully accepted their vision.
One of the key reasons for his departure was his belief in the potential of scaling AI models. Amodei said that by increasing the size of models and the data they receive, their performance could improve dramatically.
He first saw this with GPT-2 in 2019. "You find when you do that, you find incredible increases in performance. And I was finding that in 2019 with GPT-2, when we just first saw the first glimmers of the scaling laws," he said.
If you make models bigger and give them more data and computing power, their performance improves a lot is what he thought. There were some small adjustments like reinforcement learning (RL), but mostly it was about just making the models bigger, he said.
"There were a lot of folks, inside and outside, who didn't believe it at all. And we really made the case to leadership like, 'This is important, this is gonna be a big deal.' And I think they were kind of starting to believe us and ultimately went in that direction," he added.
Another major reason was his concern over responsible AI development. Amodei said his views on OpenAI's approach were doubtful.
He believed that if AI models were going to become like general-purpose thinking tools, almost as capable as the human brain, they could have huge effects on the economy, politics, and society. He felt it was extremely important to build them the right way.
"If these models are gonna be kind of general cognitive agents, like general cognitive tools that match the capability of the human brain, we'd better get this right," he said.
"The economic implications are gonna be enormous. The geopolitical implications are gonna be enormous. The safety implications are gonna be enormous. It's gonna transform how the world works. And so we needed to do it in the right way," he added.
Amodei said that even though OpenAI talked a lot about "doing AI the right way," he wasn't convinced that the company truly had a serious commitment to it. "And so my view is always, 'Don't argue with someone else's vision. Don't try to get someone to do things the way you want to,'" he said.
He advised that if you have a strong idea and some people who agree with you, it's better to go your own way. Then, you're responsible only for your own successes and mistakes.
"You don't have to answer for anyone else's. And you know, maybe your vision works out, maybe it doesn't, but at least it's yours," Amodei added.
