You may not have heard of Anthropic CEO Dario Amodei, but he’s one of a handful of people responsible for the current AI boom. As VP of Research at OpenAI, Amodei helped discover the scaling laws that project how much smarter an AI gets when given increasing amounts of computing power. He holds PhDs in physics and neuroscience and is probably a genius. Amodei, who left OpenAI in 2021 to found Anthropic, which claims to be a more safety-conscious AI company and is now valued at $61 billion after closing its latest funding round, spoke at Anthropic’s developer day in San Francisco Thursday. He’s quirky, refreshingly frank, and often funny. Here are some of his spicier quips.
On the science and art of training state-of-the-art AI models:
“There is this somewhat superstitious process of model development; it somehow comes together all at the last minute . . . There’s that alchemy that happens at the very last. It doesn’t make any sense. The training process is uniform. You would think that it doesn’t work that way, that it’s all a rational process, but it’s absolutely not.”
On the speed at which AI models are moving toward equaling or exceeding human intelligence in a broad range of tasks.
“It’s only going to be a year or two until these things are basically going to be peers to us.”
On how it feels to be immersed in an accelerating process of creating progressively smarter models.
“It’s like being on a spaceship headed away from earth and experiencing larger and larger time dilation constants.”
On whether Amodei, who is a neuroscientist, has been inspired by the human brain to create better AI models.
“It’s not like I’ve found this thing from the hypothalamus that we can use for making models. It’s all been pretty much from scratch.”
On when he thinks the world will see the first billion-dollar company with one employee.
“2026”
On whether people haven;t yet reckoned with the changing economics of software development brought about by new AI coding agents. He says new agents, like Claude Code, will shorten development time and lower the cost of creating new software.
“I think the world is going to be very different when these things can be made on an ad hoc, one-off basis in a few seconds for less than a dollar.”
On what Anthropic is doing to stop AI models from hallucinating facts or making mistakes.
“It’s really amazing that somehow human beings have managed to never have been mistaken. No one ever makes anything up. I’ve never seen anybody on TV, never seen politicians say things that aren’t true.”
On the predicament where people create new companies based on ideas that aren’t yet supported by current AI models. Amoedi says it’s good to struggle with the creation of the company until the AI becomes available.
“Hitting your head against the wall can actually sometimes be useful.”
On what he’s most excited about in the next five years.
“I hope that in five years we will have vanquished a lot of the diseases that now exist.”
No comments