Goodfire
@GoodfireAI
Followers
2K
Following
136
Statuses
85
Advancing humanity's understanding of AI through interpretability research. Building the future of safe and reliable AI systems.
San Francisco
Joined August 2024
Scaling interpretability work for frontier AI alignment has never been more important. That’s why we’re launching Ember - the first hosted mechanistic interpretability API with fast inference support for generative models like Llama 3.3 70B.
16
90
554
RT @leedsharkey: Big new review! 🟦Open Problems in Mechanistic Interpretability🟦 We bring together perspectives from ~30 top researchers…
0
86
0
Job listings here: We encourage you to apply even if you may not meet all of the requirements!
0
1
10
RT @livgorton: We (@GoodfireAI) are hiring research fellows! Early career researchers and engineers can join us for ~2 months to work on an…
0
24
0
RT @apartresearch: Myra Deng: Goodfire's Interpretability Tools in Action Our Community Manager @varchanaiyer_ interviews @GoodfireAI's @m…
0
3
0
@posedscaredcity @Sauers_ @posedscaredcity Feel free to send us a DM with more info on the issue you're running into!
1
0
1
Launch post: Open source SAEs: Ember API/SDK for accessing SAEs:
We're open-sourcing Sparse Autoencoders (SAEs) for Llama 3.3 70B and Llama 3.1 8B! These are, to the best of our knowledge, the first open-source SAEs for models at this scale and capability level.
1
6
32
4/ We're excited to see how the community builds on these foundations. If you're working on model interpretability or interested in applying similar approaches to your AI systems, we'd love to hear from you!
0
0
26
0
0
1
@IntuitMachine We've launched publicly with support for Llama 3.3 70B:
Scaling interpretability work for frontier AI alignment has never been more important. That’s why we’re launching Ember - the first hosted mechanistic interpretability API with fast inference support for generative models like Llama 3.3 70B.
0
0
2
@victormustar Our API/SDK is publicly available if you want to try out more visualizations! Check out
0
0
0
@Ahmad_Al_Dahle Our team sprinted to launch Ember - a hosted interpretability API - on Llama 3.3 70B! Check out
0
0
2
Scaling interpretability work for frontier AI alignment has never been more important. That’s why we’re launching Ember - the first hosted mechanistic interpretability API with fast inference support for generative models like Llama 3.3 70B.
0
0
4
We believe that accessible interpretability tools on powerful models enables both new research and applications. We plan to open source both our Llama 3.1 8B and Llama 3.3 70B SAEs and release a detailed research report on our steering work in January.
0
0
13