Log inSign up
Image
user avatar
Ryan Panwar
@RyanPanwar
Understanding artificial minds @GoodfireAI To follow knowledge like a sinking star, Beyond the utmost bound of human thought.
San Francisco
Joined February 2018
1,902
Following
380
Followers
RepliesMedia
  • user avatar
    Ryan Panwar
    @RyanPanwar
    Aug 4
    Silico integrates with some of the best open weight training infra, such as @thinkymachines's Tinker. It also can run on your own hardware. And it can combine the two! Here it ran RL on Tinker while running a probe-based reward model on our B200 cluster.
    Image
    Image
    Image
    01:00
    user avatar
    Goodfire
    @GoodfireAI
    Aug 4
    Silico, the platform for ambitious AI research, is publicly available today. AI is advancing fast. The tools to understand it need to advance even faster. Silico lets you interpret and train your models at frontier scale. Learn more + get access 🧵
  • user avatar
    Ryan Panwar
    @RyanPanwar
    Aug 4
    Frontier models are the most fascinating research subjects, and we’re building the world’s best lab to enable everyone to study them. Silico was born out of asking ourselves how could we build a harness, infra, and UI to 10x our research output.
    user avatar
    Goodfire
    @GoodfireAI
    Aug 4
    Silico, the platform for ambitious AI research, is publicly available today. AI is advancing fast. The tools to understand it need to advance even faster. Silico lets you interpret and train your models at frontier scale. Learn more + get access 🧵
    Image
    00:00
  • user avatar
    Ryan Panwar
    @RyanPanwar
    Jul 28
    A glimpse at the future of training: the Silico agent identified weight changes that caused mode collapse in an RL run and directly edited the weights to fix it
    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
    Ryan Panwar
    @RyanPanwar
    Jul 14
    Silico is a virtual lab for AI research. It designs experiments, manages GPUs, and presents results in a beautiful way. We believe it’s the future of AI research and want to enable everyone to study and understand the nature of intelligence.
    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
    Ryan Panwar
    @RyanPanwar
    Jun 25
    This is a really crazy result, you can precisely delete an entire language from a model without with almost 0 training data!
    user avatar
    Goodfire
    @GoodfireAI
    Jun 25
    We removed an LM's ability to speak German by fine-tuning on only 4 German tokens. As part of a 1-day hackathon with our product Silico, we removed a 67M-parameter language model's ability to predict German text, by tuning only a scalar factor on one subcomponent of the weights.
    Image