Log inSign up
Image
user avatar
Goodfire
@GoodfireAI
Using interpretability to understand, learn from, and design AI.
San Francisco
goodfire.ai
Joined August 2024
31
Following
25.9K
Followers
RepliesMedia
  • Pinned
    user avatar
    Goodfire
    @GoodfireAI
    Jul 14
    > replicate J-space on GLM 5.2 > train a reward model and run RL to reduce hallucinations > show me how this model makes cancer predictions Using our platform Silico is like having a team of AI researchers ready to run experiments like these. Private beta is open now. 🧵 (1/6)
    Image
    00:00
  • user avatar
    Goodfire
    @GoodfireAI
    Jul 30
    You ask an AI model a question. Why does it answer the way it does? Using the model's own neurons, we can trace how it makes decisions — and steer it toward better ones. Case study: we tested an LLM and found that it sometimes endorses drunk driving.🧵 (1/5)
    Image
    GIF
  • user avatar
    Goodfire
    @GoodfireAI
    Jul 29
    We asked our agentic research platform Silico to replicate three classic causal interpretability papers without the authors' code - Interpretability in the Wild, ROME, and Mixing Mechanisms. One prompt each, then we let it run autonomously. (1/3)
    Image
  • user avatar
    Goodfire
    @GoodfireAI
    Jul 28
    In a recent hackathon, we trained an LLM to label its own neurons. But our RL run collapsed. 94% of labels opened with the word "texts." Instead of reworking our data + retraining, we edited the weights directly - dropping “texts” from 94% to 5% with minimal side effects. (1/5)
    Image
    GIF
  • user avatar
    Goodfire
    @GoodfireAI
    Jul 27
    There's a "bouba-kiki space" inside LLMs "Spiky-sounding" and "round-sounding" words align on opposite ends of a specific direction in Llama & Gemma's activations, independent of their meaning, mirroring the famous bouba-kiki effect
    Image