Getting better at coding can make models better at reasoning
Demis HassabisDemis Hassabis — Scaling, superhuman AIs, AlphaZero atop LLMs, AlphaFoldat 2:00
From the conversation
For example, when these large models improve at coding, that can actually improve their general reasoning. So there is evidence of some transfer although we would like a lot more evidence of that. But that’s how the human brain learns too. If we experience and practice a lot of things like chess, creative writing, or whatever, we also tend to specialize and get better at that specific thing even though we’re using general learning techniques and general learning systems in order to get good at that domain. What’s been the most surprising example of this kind of transfer for you? Will you see language and code, or images and text? I’m hoping we’re going to see a lot more of this kind of transfer, but I think things like getting better at coding and math, and then generally improving your reasoning. That is how it works with us as human learners. But I think it’s interesting seeing that in these artificial systems. And can you see the sort of mechanistic way, in the language and code example, in which you’ve found the place in a neural network that’s getting better with both the language and the code? Or is that too far down the weeds? I don’t think our analysis techniques are quite sophisticated enough to be able to hone in on that.…
Summary
Large language models exhibit transfer learning across domains, where improvements in a specific area like coding can enhance general reasoning ability. This mirrors how the human brain operates — using general learning systems to specialize in particular domains like chess or creative writing, while potentially improving broader capabilities in the process. More evidence of such transfer is still needed, but the pattern parallels known mechanisms of human skill acquisition.
Watch the clip on YouTubeStarts at 2:00