Aashish Pahwa on Instagram
Aashish Pahwa · Instagram post · source
3,239 likes · 1,767 comments · September 3, 2026
Comment “HUI” and I’ll send you the links.
Major open models are getting unofficial uncensored versions, and the way they’re being made is pretty interesting.
HuiHui AI is using a technique called abliteration to weaken the refusal behavior built into these models.
That means models like GPT-OSS, Qwen, DeepSeek, Llama and Gemma can behave very differently from their standard versions.
You can even run them locally with Ollama or LM Studio.
Just remember, fewer guardrails also means more questionable outputs and less reason to blindly trust the model.