Listening to his podcast on the topic was so disappointing. I think Ezra is a smart guy, but he doesn’t understand the field and the entire premise of the long discussion was that LLMs are going to get us to AGI.
He also casually dropped that he talked to people at firms with "high amount of coding", who told him that by the end of this or next year "most code will not be written by humans".
Yeah, okay. I work each day with Copilot etc and the practical value is there, but there are so many steps missing for this statement to be true, that I highly doubt it.
My case is, wouldn't we already see the tools that are at least getting close to this goal? I can't believe that (or AGI in fact) to be a big bang release. It looks more like baby steps for now.
> but he doesn’t understand the field and the entire premise of the long discussion was that LLMs are going to get us to AGI.
You might disagree with this assessment, but it doesn't show that he doesn't understand the field, since many of the people in the field also think this. At least to the definition of AGI he used - AI that can replace most of the economic work done by humans.
> it doesn't show that he doesn't understand the field, since many of the people in the field also think this
Have you ever interviewed people for software positions?
After being a part of the interview loop myself for a while, I am confident that “many of the people in the field” also don’t understand it. Not even talking about the leetcode gauntlet or complex systems design at all, just pure fundamentals and very basics. Not even talking about the larger scale picture of the business .
With that in mind, the fact that “many of the people in the field” he interviewed at “places with high amount of coding” agreed with him doesn’t say much.
I've been interviewing people for fifteen years, yes.
And while it's true that many people don't understand much, I don't think this applies to many people working at frontier model or AI companies. Especially not the high level people Ezra said he talked with.
Fair. I am not sure if people working at AI companies is a good measurement for this, as they have a personal stake in their claims.
Maybe I am just morally corrupt enough to entertain this idea, but people working at AI companies overexaggerating and overhyping the impact of their work sounds like an obvious move.
I doubt it's that. But he also talked to people outside of these ai companies. The person he interviewed on the podcast was in the Biden administration.
Yeah, okay. I work each day with Copilot etc and the practical value is there, but there are so many steps missing for this statement to be true, that I highly doubt it.
My case is, wouldn't we already see the tools that are at least getting close to this goal? I can't believe that (or AGI in fact) to be a big bang release. It looks more like baby steps for now.