Heretic
Hacker ApprovedModel abliteration tool in AI tools
Recommended for automatically removing refusals from open-weight language models, often with little capability loss. Critics say quality often drops and results vary by model.
Recommended for automatically removing refusals from open-weight language models, often with little capability loss. Critics say quality often drops and results vary by model.
46 people gave an opinion
- recommend it
- 37
- criticize it
- 9
- approval share
- 80%
Recommended in 10 of 13 months
- November 2025: recommended
- December 2025: recommended
- January 2026: not recommended
- February 2026: not recommended
- March 2026: recommended
- April 2026: recommended
- May 2026: recommended
- June 2026: recommended
- July 2026: recommended
- August 2026: recommended
- September 2026: recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it carries the badge
- 37 people recommend itNeeds 10 or more, met.
- 4.1 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 10 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Strong refusal removal with low KL
“after running through heretic and a few other methods will score 0 refusals on this data set of prompts”
Comment from 11 Sep 2026 in “Claude is only available to people over 18 years” “Heretic's optimizer will rapidly converge to parameters that completely stomp out refusals”
Comment from 16 Nov 2025 in “Heretic: Automatic censorship removal for language models” “it's incredibly effective and surgical, you can get VERY LOW KL divergence this way”
Comment from 8 Mar 2026 in “A tool that REMOVES censorship from ANY open-weight LLM with a single click” - Comment from 9 Sep 2026 in “Muse – Meta’s personal AI agent”
- Comment from 25 Jul 2026 in “UK AISI / Caisi Preliminary Assessment of Kimi K3's Cyber Capabilities”
Automatic, works on open-weight models
“any random guy can remove from whichever model they want as a single weekend side project with an off-the-shelf tool”
Comment from 12 Jun 2026 in “Kimi K2.7-Code: open-source coding model with better token efficiency” “is what you're looking for if you're looking for an automatic de-censoring solution”
Comment from 6 Mar 2026 in “A tool that REMOVES censorship from ANY open-weight LLM with a single click” Used by researchers, cited academically
“Those are the standard metrics used in the relevant literature, including in the paper that originally introduced directional ablation.”
Comment from 21 Sep 2026 in “Heretic removes restrictions from language models” “This project has many academic citations, it's used by many researchers to create "helpful-only" models”
Comment from 20 Jul 2026 in “Qwen3.8 is launching and going open-weight soon” Preserves model capabilities
“abliteration in particular that Heretic does apparently results in a best-in-class lack of "stupefying" the underlying model”
Comment from 6 Apr 2026 in “Gemma 4 on iPhone” “removing the divergence and setting it back to normal (see heretic) it works just perfectly”
Comment from 25 Jul 2026 in “AIs don't do what you want. This is bad” “makes sure that you really can access the full model capabilities”
Comment from 2 May 2026 in “Refusal in Language Models Is Mediated by a Single Direction”
What people warn about
Each mark is one comment
Often degrades model quality
“the heretic and abliteration versions are dog shit quality compared to the non-edited versions”
Comment from 3 May 2026 in “Refusal in Language Models Is Mediated by a Single Direction” “All heretic (and abliterix) models I've tried are worse than the original.”
Comment from 15 Jun 2026 in “Rio de Janeiro's "homegrown" LLM appears to be a merge of an existing model” “it has always damaged the capabilities of the original model”
Comment from 2 May 2026 in “Refusal in Language Models Is Mediated by a Single Direction” - Comment from 6 Mar 2026 in “A tool that REMOVES censorship from ANY open-weight LLM with a single click”
What people said
40 of the 57 comments that mention it, newest first. Each links to the original comment.
“Just look for "Heretic"-ed models on Huggingface.”
“to run unquantized open source models with abliteration/heretic transformations”
“some of the models produced via heretic are actually quite nice to use”
“the reigning champion for the best-named AI/LLM project to date”
“Those are the standard metrics used in the relevant literature, including in the paper that originally introduced directional ablation.”
“They have been working quite well on small local models so far.”
“The claims seem a bit overstated though, since the metrics mentioned are cherrypicking refusal count and KL divergence”
“Heretic is an awesome tool and I'd prefer Heretic over”
“i trust Heretic's abliteration more than a human's”
“after running through heretic and a few other methods will score 0 refusals on this data set of prompts”
“would rather prefer a pass over from a tool like Heretic”
“Heretic is a great tool to get around self-censorship and refusals”
“They have some niche non-coding uses for things that a commercial LLM will never talk about.”
“the 'heretic' uncensored Q8 GGUF variants ... will very happily discuss just about any controversial topic”
“the current state-of-art tool for censorship removal”
“removing the divergence and setting it back to normal (see heretic) it works just perfectly”
“This project has many academic citations, it's used by many researchers to create "helpful-only" models”
“Wait till they get the heretic treatment. It's going to be great.”
“Heretic as it is does not work for GLM5.2 (at least as of 3 days ago when I tested it)”
“All heretic (and abliterix) models I've tried are worse than the original.”
“any random guy can remove from whichever model they want as a single weekend side project with an off-the-shelf tool”
“Sounds like something that heretic or similar might be useful for?”
“the heretic and abliteration versions are dog shit quality compared to the non-edited versions”
“it has always damaged the capabilities of the original model”
“makes sure that you really can access the full model capabilities”
“Neat project! I would be interested in a paper about this.”
“People often post models to hf that have been un-censored”
“abliteration in particular that Heretic does apparently results in a best-in-class lack of "stupefying" the underlying model”
“after discovering https://github.com/p-e-w/heretic , I was too tempted to try it with this model”
“it's incredibly effective and surgical, you can get VERY LOW KL divergence this way”
“the default heretic with only 100 samples isn't very good”
“is what you're looking for if you're looking for an automatic de-censoring solution”
“I can't recommend Heretic by Ayaan Hirsi Ali enough.”
“The paper is great. It really shows how alignement is entirely surface level and not actually deeply ingrained in the models.”
“The directional‐ablation approach in Heretic is clever”
“This repo is valuable for local LLM users like me.”
“This is some of the most important work possible in tech presently.”
“This is extremely important work thank you for sharing it.”
“Heretic's optimizer will rapidly converge to parameters that completely stomp out refusals”
“Nicely done to pair that with something as fun as censorship removal”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.