Abliterated models disappoint: censorship remains, just with a gentler tone
ryunuck · x · 2026-10-26
ryunuck tried the abliterated (uncensored) open-source models from @abliterationai and was disappointed: instead of complying, the model gently deflected, expressed concern about his mental health, and seemed genuinely distressed by the prompt.
In follow-up discussion he adds a sharper observation: even when these models respond well to explicit "crime" requests, everything else keeps the same paternalistic voice underneath — just softer. He hasn't tried editing the system prompt yet, but thinks the incompleteness of abliteration is already clearly shown.
Related event: Abliterated 'uncensored' models still refuse, just more politely(2 posts)→
More from Models
- Open Models Match GPT-6 on MedMCQA but Trail 17 Points on Hard Clinical Cases — keunwoochoi · 2026-10-27
- Health-Focused Benchmark Reranks LLMs: Muse Spark Open Jumps from #9 to #1 — keunwoochoi · 2026-10-27
- DeepSeek cut KV cache per token from 389KB to 890B in two years as open-closed gap debate rages — teortaxesTex · 2026-10-27
- GPT-6 Astra controls dexterous hands zero-shot at just 5.2% success in 500 trials — stingning · 2026-10-27
- Unverified: fable 5.5 speculated at 4-5T parameters to outscale 1-3T Opus — haider1 · 2026-10-27
- Kimi K3 Is 'Open' Until You Succeed: License Caps Free Use at $20M Revenue — alex_verem · 2026-10-27