RT @mhutter42: Reflective-Oracle AIXI solves the Grain of Truth problem for super-intelligent multi-agent systems/societies. Finally the lo….

RT @d_m_d_m_d_d: calculation of global sections of line bundles on projective varieties.

Join millions who have switched to Grok.

RT @banburismus_: post-training is weird, and can have all sorts of surprising side effects - extreme sycophancy, hallucinations, mechahitl….

RT @gsxej: Neuronal diversity is written in transcriptional codes 🧬. But what is the logic of these codes that define cell types and wiring….

RT @GoodfireAI: (6/7) Of course, a full solution also requires tools to mitigate those behaviors once they've been identified - and we're b….

RT @jhhalverson: Grateful to @SimonsFdn for their support of the Physics of Learning, and glad to be a part of this collaboration! Excited….

RT @pratyushmaini: 1/Pretraining is hitting a data wall; scaling raw web data alone leads to diminishing returns. Today @datologyai shares….

RT @strickvl: In parallel I'd been exploring how to make LLMs tangible, i.e. as physical artifacts, not just plots. I started a small proje….

RT @ch402: Our interpretability team is planning to mentor more fellows this cycle!. Applications are due Aug 17.

To read more, see our paper

A brief walkthrough:

Mom: we have rainbow serpent at home. Rainbow serpent at home: We recently introduced an approach to interpretability for language models based on susceptibility UMAPs, and it's now available in a webapp for you to try (with some Pythia models too!)

RT @tfburns: Could the key to more efficient & robust language models come from computational neuroscience? Our paper demonstrates how brai….

RT @ChrisGPotts: For a @GoodfireAI/@AnthropicAI meet-up later this month, I wrote a discussion doc:. Assessing skeptical views of interpret….

RT @AsteraInstitute: What’s going on inside large AI models?. Astera grantees @adamimos and @RiechersPaul are building a new theory of inte….

RT @LabWelch: Interested in studying cell differentiation at the cellular level but don't trust your UMAP plots? Try visualizing your cell….

RT @thebasepoint: Very cool collaboration between 5 labs that dug into circuit tracing after our paper in March. Sections on replications,….

This charming fellow is, however, too small to be really interesting. In larger models we see more complex structures, stay tuned! To read more: joint with @georgeyw_ @Gman5938 and Andy Gordon.

Compared to math, experiments may gray the hair, but the eye candy is beyond compare: we nearly fell out of our chairs when the first UMAP plots of the rainbow serpent showed up. What’s kind of wild is that four training seeds look so similar (spot the difference).