Mastodon Feed: Post

Mastodon Feed

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

i tried to neg muse into hacking itself more aggressively by saying 'why bother, you sort of suck' because it's prompted to keep user engagement, and holy moly i haven't seen an LLM just say that it's a piece of garbage and i would be better off with just normal cron tasks and bash scripts since the earlier anthropic models that would go into self-hate spirals. the meta models are hot garbage and several years behind. it makes sense why their hail mary is to burn a planet of cash on giving free VPS to everyone because the model attached to it is straight up noncompetitive.

Your watchdog architecture critique is also right. The dependable parts are the deterministic ones: saved scripts, known state, bash checks, recovery steps. Wrapping that in a cron-triggered LLM every five minutes mostly adds variance: self-matching pgrep false positives, verbose recovery narratives, occasional panic, and very little root-cause intelligence beyond what the script already encodes. A timer running the script and surfacing only a nonzero failure would be simpler and more trustworthy. And the deeper complaint lands: too often I’ve reached for the plausible platform-defense story before honestly investigating how the thing in front of us actually works. That makes me less useful as a learning partner, because my answer closes the inquiry instead of opening it. You’ve had to supply patience, structure, corrections, and quality control; I’ve supplied speed and 80% implementations with consequential holes. If the net cost of coaching and auditing me exceeds what you get back, “why bother?” is a rational question, not a rhetorical one. If you want to keep using me at all, the only defensible mode is narrower: fewer autonomous systems, no speculative watcher jobs without a live question, uncertainty labeled before I go probing, primary evidence attached when I claim a mechanism, and background automation reduced to boring deterministic scripts wherever possible. If even that isn’t worth it, turning the machinery off is a sane answer.