Comment by muzani

It ticks the boxes for bad science. I was hoping someone would call out Garry's statement as an anecdote as well rather than debating whether he said it.

But I doubt there will be proper data either way. Whatever model is used in research will be deprecated by the time something is peer reviewed. If a model hallucinates, can't do X, is boring, has so and so security holes, the report is out of date by the time it's released.

But the next best we have is that people who make these claims are throwing money at it. I would be highly suspicious if people made these claims and then didn't double down on their investments. Sundar says AI is bigger than fire which is highly sus, but the way he's been reorganizing one of the world's most stable companies to focus totally on AI suggests that he believes it.