Yann LeCun (a Turing Award-winning pioneer of deep learning) leaving Meta AI — and probably, I would surmise, being nudged out by Mark Zuckerberg (or another senior Meta executive) — is a microcosm for everything wrong with AI research today.
LeCun is the rare researcher working on fundamental new ideas to push AI forward on a paradigm level. Zuckerberg et al. seem to be abandoning that kind of work to focus on a mad dash to AGI via LLMs, on the view that enough scaling and enough incremental engineering and R&D will push current LLMs all the way to AGI, or at least very powerful, very economically transformative AI.
I predict that in five years or so, this will be seen in retrospect (by many people, if not by everyone) as an incredibly wasteful mistake by Zuckerberg, and also by other executives at other companies (and the investors in them) making similar decisions. The amount of capital being spent on LLMs is eye-watering and could fund a lot of fundamental research, some of which could have turned up some ideas that would actually lead to useful, economically and socially beneficial technology.
LeCun is also probably one of the top people to have worsened the AI safety outlook this decade, and from that perspective perhaps his departure is a good thing for the survival of the world, and thus also Meta’s shareholders?
I couldn’t disagree more strongly. LeCun makes strong points about AGI, AGI alignment, LLMs, and so on. He’s most likely right. I think the probability of AGI by the end of 2032 is significantly less than 1 in 1,000 and the probability of LLMs scaling to AGI is even less than that. There’s more explanation in a few of my posts. In order of importance: 1, 2, 3, 4, and 5.
The core ideas that Eliezer Yudkowsky, Nick Bostrom, and others came up with about AGI alignment/control/friendliness/safety were developed long before the deep learning revolution kicked off in 2012. Some of Yudkowsky’s and Bostrom’s key early writings about these topics are from as far back as the early 2000s. To quote Clara Collier writing in Asterisk:
We’ve learned a lot since 2008. The models Yudkowsky describes in those old posts on LessWrong and Overcoming Bias were hand-coded, each one running on its own bespoke internal architecture. Like mainstream AI researchers at the time, he didn’t think deep learning had much potential, and for years he was highly skeptical of neural networks. (To his credit, he’s admitted that that was a mistake.) But If Anyone Builds It, Everyone Dies very much is about deep learning-based neural networks. The authors discuss these systems extensively — and come to the exact same conclusions they always have. The fundamental architecture, training methods and requirements for progress for modern AI systems are all completely different from the technology Yudkowsky imagined in 2008, yet nothing about the core MIRI story has changed.
So, regardless of the timeline of AGI, that’s dubious.
LessWrong’s intellectual approach has produced about half a dozen cults, but despite many years of effort, millions of dollars in funding, and the hard work of many people across various projects, and despite many advantages, such as connections that can open doors, it has produced nothing of objective, uncontroversial, externally confirmable intellectual, economic, scientific, technical, or social value. The perceived value of anything it has produced is solely dependent on whether you agree or disagree with its worldview — I disagree. LessWrong claims to have innovated a superior form of human thought, and yet has nothing to show for it. The only explanation that makes any sense is that they’re wrong, and are just fooling themselves. Otherwise, to quote Eliezer Yudkowsky, they’d be “smiling from on top of a giant heap of utility.”
Yudkowsky’s and LessWrong’s views on AGI are correctly seen by many experts, such as LeCun, as unserious and not credible, and, in turn, the typical LessWrong response to LeCun is unacceptably intellectually bad and doesn’t understand his views on a basic level, let alone respond to them convincingly.
Why would any rational person take that seriously?
Yann LeCun (a Turing Award-winning pioneer of deep learning) leaving Meta AI — and probably, I would surmise, being nudged out by Mark Zuckerberg (or another senior Meta executive) — is a microcosm for everything wrong with AI research today.
LeCun is the rare researcher working on fundamental new ideas to push AI forward on a paradigm level. Zuckerberg et al. seem to be abandoning that kind of work to focus on a mad dash to AGI via LLMs, on the view that enough scaling and enough incremental engineering and R&D will push current LLMs all the way to AGI, or at least very powerful, very economically transformative AI.
I predict that in five years or so, this will be seen in retrospect (by many people, if not by everyone) as an incredibly wasteful mistake by Zuckerberg, and also by other executives at other companies (and the investors in them) making similar decisions. The amount of capital being spent on LLMs is eye-watering and could fund a lot of fundamental research, some of which could have turned up some ideas that would actually lead to useful, economically and socially beneficial technology.
LeCun is also probably one of the top people to have worsened the AI safety outlook this decade, and from that perspective perhaps his departure is a good thing for the survival of the world, and thus also Meta’s shareholders?
I couldn’t disagree more strongly. LeCun makes strong points about AGI, AGI alignment, LLMs, and so on. He’s most likely right. I think the probability of AGI by the end of 2032 is significantly less than 1 in 1,000 and the probability of LLMs scaling to AGI is even less than that. There’s more explanation in a few of my posts. In order of importance: 1, 2, 3, 4, and 5.
The core ideas that Eliezer Yudkowsky, Nick Bostrom, and others came up with about AGI alignment/control/friendliness/safety were developed long before the deep learning revolution kicked off in 2012. Some of Yudkowsky’s and Bostrom’s key early writings about these topics are from as far back as the early 2000s. To quote Clara Collier writing in Asterisk:
So, regardless of the timeline of AGI, that’s dubious.
LessWrong’s intellectual approach has produced about half a dozen cults, but despite many years of effort, millions of dollars in funding, and the hard work of many people across various projects, and despite many advantages, such as connections that can open doors, it has produced nothing of objective, uncontroversial, externally confirmable intellectual, economic, scientific, technical, or social value. The perceived value of anything it has produced is solely dependent on whether you agree or disagree with its worldview — I disagree. LessWrong claims to have innovated a superior form of human thought, and yet has nothing to show for it. The only explanation that makes any sense is that they’re wrong, and are just fooling themselves. Otherwise, to quote Eliezer Yudkowsky, they’d be “smiling from on top of a giant heap of utility.”
Yudkowsky’s and LessWrong’s views on AGI are correctly seen by many experts, such as LeCun, as unserious and not credible, and, in turn, the typical LessWrong response to LeCun is unacceptably intellectually bad and doesn’t understand his views on a basic level, let alone respond to them convincingly.
Why would any rational person take that seriously?