Ask HN: What are the hottest areas of *non*-LLM AI work currently?

10+ years ago, "AI" would likely refer to work in RL, evolutionary/genetic algorithms, etc.

Nowadays most of the spotlight seems centered on LLMs, CNNs, and other methods that are either human-labeled or at least reliant on human-created data, and have a static separation between "learning" and "inference".

I know that there are still non-LLM, non-CNN, non-anthropocentric topics of AI development currently, in RL and in other areas. Which would you say are the most prominent or promising today, or likeliest to come to fruition?

4 points | by dira3 2 hours ago

1 comments

  • dira3 2 hours ago
    BTW, on the topic of the fading of genetic algorithms, here is an interesting recent take: https://statmodeling.stat.columbia.edu/2025/04/17/what-happe...
    [-]
    • muzani 19 minutes ago
      The quoted paper is from 2018. Evolutionary programming is IMO similar to a search algorithm. The biosphere is not. It's not searching; it's removing unfit matches. Which is far less efficient.

      I'm a big fan of evolutionary programming; it's just inefficient in the past. I think LLM agents might just be the little advantage they need, like guided missiles with GPS.

      Evolutionary programming is hard though. I think it might answer OP's question - it's something that's difficult enough for most people to avoid, but there would be greatly increased interest in it.