jevPOST
Don't use abliterated models in Jev-style inference harnesses
The author reminds users that when running LLMs in an open-source Jev-style inference harness, they should not use abliterated models, as it affects decision reliability.
remy🐀
@_mattataArchived 2026.09.21Originally posted 2026.09.20
remy🐀 @_mattataOriginally posted 2026.09.20
This should go without saying, but if you’re using LLM models in an open source Jev-style inference harness, don’t use abliterated models.
x.com/_mattata/status/2101469072852435244Archived 2026.09.21 10:29