Post #2837030
2026-05-12 23:19 UTC
@plexus@toot.cat My experience is that LLMs are most useful for tasks that are trivially verifiable. I tend to use them for small, one off scripts or for searching through documentation. I haven't found them that useful for debugging, except as a rubber duck.
Replies (1)
-
@jstepien@mastodon.social 2026-05-13 08:04
@weavejester@mastodon.social @plexus@toot.cat Seconding the importance of ease of automated verification. I've had good results with property-based testing. A small local LLM can debug quite nicely given shrunk sequences of steps leading to a failed assertion generated by a stateful model-based PBT with a test oracle.