Pineau: Neural networks generalize narrowly and produce noise far from training
Joelle Pineau · Why Meta Wants To Build Artificial General Intelligence — With Joelle Pineau, VP of Meta AI Research · Jan 24, 2024 · at 11:05
Joelle Pineau, head of Meta FAIR, explains high-dimensional learning limitations in massive parameter foundational models.
0:00 / 0:17exact quote · 17.1s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“I do believe they can generalize. I think they generalize relatively narrowly, or at least, you know, as long as you stay close, you get a good manifold of information. When you start to go really far afield from your data, because the dimensions are so large, you get all sorts of noise.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Joelle Pineau
Pineau: Halting open-source AI now prevents developing vital model safety practices
“My worry about closing the doors down now is that the models are only getting better. And so if we don't release them now, we really miss an opportunity to develop the muscle we need to make these models safer. And I don't think today's model are the ones that…”
Pineau: AI is far away from a breakthrough transformer moment for reasoning
“The transformer moment for reasoning and choosing action and being able to plan at different levels of granularity. We're still far away from doing that.”
Pineau: Reasoning models fail at multi-tier hierarchical planning
“That's the part that the reasoning models don't do. They do really well at like one level of granularity... But the going back and forth between different levels of sort of resolution of action, it's really hard. So on the technical terms, we call it hierarchi…”
Pineau: Future of AI will be many specialized agents, not one superintelligence
“I tend to actually place my bet not on the fact that we're going to reach like a single super intelligent agent, but on the fact that we are much more likely to live in a future where there's going to be many agents for many things. And so some agents will abs…”
Pineau: Meta's next AI step change requires general and world models
“We've done a ton of work on AI on the platform over the last few years. That was what I would call more narrow specialized models. We can continue to do that, but the bigger step change are gonna come through the more general model, building foundation models,…”
Pineau: LLMs are currently the fastest progressing path toward general intelligence
“It gives us a path. It may not be the path, but it gives us at least a path to move towards general intelligence. And it's an exciting one. It's one that, that we're exploring. It doesn't mean that we've stopped exploring other paths towards general intelligen…”