Reddit debate: cloud LLMs change rules on a whim — only local models are reliable
Substantial_Swan_144 · reddit · 2026-11-01
A Reddit post argues the biggest problem with cloud LLMs isn't technical but commercial: providers want the benefits of language models without the liabilities, and can change rules on a whim. Cited examples include sudden refusals of game disassembly and translation, and GPT silently sanitizing translations. The author warns that betting entire workflows on 2-3 big tech companies risks overnight censorship or sabotage, concluding that only locally-run models can be genuinely reliable. Discussion thread ensues.
More from AGI Musings
- Economists warn: AI agent-based modeling risks repeating the failed SFI agenda — Afinetheorem · 2026-11-01
- "Funny that no one knows how to make money from intelligence" — gklambauer · 2026-11-01
- Agentic AI is training users to articulate their desires and principles precisely — sebpaquet · 2026-11-01
- Gartner: 38% of orgs pilot AI agents, only 11% in production, 40% of projects to fail by 2027 — ingliguori · 2026-11-01
- Weekly: Anthropic's S1 filing, labs claim models refute Exponential Time Hypothesis — njyx · 2026-11-01
- Pedro Domingos revives Moravec's paradox: AI cracks Navier-Stokes but can't do accounting — pmddomingos · 2026-11-01