4 ms·
The performance is great, but the censorship is ridiculous for me. I tried it as a backend for my game Guessix[1], but it would refuse for ridiculous reasons li
by corlinp 1y ago
The performance is great, but the censorship is ridiculous for me. I tried it as a backend for my game Guessix[1], but it would refuse for ridiculous reasons like "Cannot answer questions about copyrighted works like Harry Potter."
1. https://guessix.com/ https://guessix.com/
- BoorishBears 1y agoUse constrained generation
- artdigital 1y agoMind explaining?
- 7thpower 1y agoCurious as well
- BoorishBears 1y agoIf you constrain the model to a JSON schema, most frivolous refusals go away. And if you finetune on a few formatted examples the effect is even greater
- corlinp 1y agoDo you mean like structured outputs? Unfortunately here the model is guided to explicitly tell you when you violate the rules and why, it can confuse it's system rules with the game rules and say you're not allowed to ask a question about copyrighted material etc.
- artdigital 1y agoTry the uncensored/jailbroken variants like openai-gpt-oss-20b-abliterated-uncensored-neo-imatrix I just tried to ask it how to make crystal meth and it generated a very detailed step by step guide
- Squarex 1y agoI have heard that uncensorted gpt-oss is not very good because of it being trained mainly on synthetic data. Is not not true?
- bavell 1y agoIirc abliteration (ablation?) can be done without "training" and is pretty quick. It finds the individual weights related to the concept you want to ablate, and modifies those weights to "deactivate" them. Precision brain surgery, to anthropomorphize.
- Squarex 1y agoThe problem with synthetic data would be that the censored information would not be in the training data at all.
- corlinp 1y agoVery interesting! Do the benchmarks hold up well or does it reduce performance in other areas too?