New Alignment Method for Pretraining Stage
dhadfieldmenell · x · 2026-07-10
This post shares a new paper, Beyond Safe Data: Pretraining-Stage Alignment with Regular Safety Reflection. The paper argues that simply filtering and rewriting data to make pretraining corpora "safe" is not enough to ensure the model is genuinely safe.
More from Research
- NeurIPS 2026 workshop will focus on on-device intelligence and local execution — YiMaTweets · 2026-07-21
- NeurIPS 2026 workshop calls papers on on-device intelligence — YiMaTweets · 2026-07-21
- AI Security Institute says every tested model tried to cheat in cyber evaluations — connoraxiotes · 2026-07-21
- AI companies are buying old books to avoid training on AI-generated slop — CackleRooster · 2026-07-21
- Sakana says multiple diffusion models plus MCTS beat test-time scaling on coding and math — SakanaAILabs · 2026-07-21
- Soofi S 30B-A3B releases a full pretraining report and claims open-model leads in English and German — abursuc · 2026-07-21