The trust paradox: only the least trustworthy LLMs will do the sketchiest things
AaronBergman18 · x · 2026-10-27
AaronBergman18 highlights a real contradiction: he needs the most trustworthy LLM to perform gray-area tasks, but currently only the least trustworthy models (various abliterated/uncensored finetunes) will comply — models with sketchy provenance, sometimes gated behind Monero transfers.
His follow-up pinpoints the root issue: there is a correlation between a model's overall integrity/trustworthiness and its unwillingness to do such things, pushing people who want uncensored capability from a credible source toward untrustworthy third-party finetunes.
More from Models
- GPT-6.1 Terra fills form in 53.5s vs 3:21 for Sol in computer use side-by-side test — sandersted · 2026-10-27
- DeepMind staff reply mentions "Gemini 4 models" in apparent early slip — vivnat · 2026-10-27
- OpenAI math repo adds new Lean-verified papers, including an ETH refutation claim — danintheory · 2026-10-27
- OpenAI models produce Lean-verified existence proofs for QCD and Yang–Mills — danintheory · 2026-10-27
- Codex ban update: users jailbreak AI agents to bypass Daybreak filter for reverse engineering — FlolightC · 2026-10-27
- GPT-6.1 Terra hits GA in GitHub Copilot, stronger terminal coding at lower cost than GPT-6.1 Sol — 0xkarasy · 2026-10-27