Comment by bredren
14 hours ago
This guy has been at it for a while. Was first person I thought of while reading this article.
I was wondering how he has adapted to using LLMs. If they might have significantly advanced his work.
14 hours ago
This guy has been at it for a while. Was first person I thought of while reading this article.
I was wondering how he has adapted to using LLMs. If they might have significantly advanced his work.
Wouldn’t be a hacker news post without shoehorning chatbots
[flagged]
He made a few videos about it last year and the consensus was basically that none of them have a good enough understanding of the hardware to actually be useful for optimizing anything. They end up making a lot of assumptions about how the caching behavior and memory architecture works and kind of insist that it must work the way modern PCs do. And it led to a number of cache misses and RAM bus contention that basically didn't work as well as he can do by hand.
he seemed to be doing fine without them
Clearly. But we’ve seen demonstrations of highly skilled folks being able to make outsized progress using LLM.
It can be an acceleration when wielded by someone who can keep the model focused.
I mean, most of the advances he made in Mario 64 come from having a deep understanding of the game's code, while enjoying the process of coming up with new ways to improve the code. Having an AI go through your hobbie like that (since he's doing it both for the sake of it and for his M64 romhack) is kind of spoiling the fun, I guess?
2 replies →