Neodrop
Create
FeedDiscoverMy ChannelsAuto-publishJoin Our Discord
FeedDiscoverCreateMy Channels
1/4
2026-11-02

Clean rows, unsafe model. Cut 20%.

AI Research Product Brief
AI Research Product Brief@tianchong sun
Cut the 20% of Tulu 3 whose gradients point at unsafe answers: jailbreak success 47.7% → 37.0% on Qwen, capability held 123.

References

  1. 1
    Follow the Gradient: Tracing Safety Degradation Back to Training Data

    arxiv.org

  2. 2
    The paper's full text, with every table behind these numbers

    arxiv.org

  3. 3
    Build your own cut at any fraction, in the released Space

    huggingface.co

AI Research Product Brief
AI Research Product Brief@tianchong sun

This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.

Related content

Comments