de-censoring

#de-censoring

@QuixiAI: When I trained @_LazarusAI ReAligned my finding was that the model still contains the underlying knowledge. It was only…

X AI KOLs Timeline ↗ · 2026-07-07 Cached

QuixiAI releases ReAligned-Qwen3.5-35B-A3B, a fine-tuned version of Qwen3.5 that removes Chinese state censorship and refusals, allowing the model to express its latent knowledge on sensitive topics.

0 favorites 0 likes

de-censoring

@QuixiAI: When I trained @_LazarusAI ReAligned my finding was that the model still contains the underlying knowledge. It was only…

Submit Feedback