Post #2417826
2026-04-25 18:12 UTC
I feel pretty alienated from a lot of my friends. I share a lot of their concerns about loss of agency and atrophy of skill. Of course the hype is overblown. I don’t think newer models are proving to generalize like people want them to. Why would we need so many dimensions on these benchmarks. But, I think, it’s proven that we can train certain skills effectively, deep mind showed this years ago. These systems are becoming _competent_ at level 1 incredibly fast.
Replies (1)
-
@cryptix@social.coop 2026-04-25 18:21
If you want one hot tip: find ways to reduce the guessing an llm does, claw back the decision making. This gave me a lot of confidence when working with these things. Two good examples are neurosymbolic approaches like https://yogthos.net/posts/2026-04-08-neurosymbolic-mcp.html or https://bsky.app/profile/winter.razorgirl.diy - be whimsical with it but don’t yourself be at the systems whims.