Loading activity…
@goodfireai
Using interpretability to understand, design, and align AI.
AI
Goodfire researcher Nick Cammarata argues that modern tools reveal most neurons as legible in models up to trillions of parameters.
AI
GoodfireAI says EVEE scored 4.2 million ClinVar variants in April, with MEPE c.209_212del scoring particularly high.
AI
Goodfire says its activation monitors use signals from inside models to monitor for prompt injection and sensitive data exposure.
AI
GoodfireAI says participants will build models using a 150-million-cell atlas, with top teams’ hypotheses tested in Prima Mente’s lab.
AI
Goodfire says external red-teaming from @farairesearch found its monitor cut universal jailbreak success to 0%.
AI
GoodfireAI lists papers on manifold steering, whether SAEs capture concept manifolds, and Llama’s base-10 reasoning about cyclic concepts.
AI
GoodfireAI says its monitors use protein model embeddings for millisecond-speed screening and produce fewer false positives than frontier-model safeguards.
AI
Goodfire CEO Eric Ho argues that understanding models’ internal mechanisms is necessary to shape what they learn during training and verify what they’ve learned afterward.
AI
Goodfire says sparse autoencoders are useful for exploring a model when you don’t know what to look for. For a specific target, such as reward hacking, it recommends a probe.
AI
Goodfire is inviting interpretability, alignment and life sciences researchers to meet its team over vinyl and drinks.
AI
Goodfire says its probes catch cheating that other AI monitors miss. On Kimi K3, it reports a 90% cut in the cost of monitoring with large language models, with about a 1% drop in precision.
AI
Base Labs says it will publish methods for training and monitoring open models, while Baseten will integrate the work into its deployment infrastructure and offer it as a managed service.
AI
The launch announcement says frontier agents still cheat frequently, and describes an evaluation spanning math, coding, knowledge work, visual tasks and more.
AI
Goodfire shares 92-second clip of Chief Scientist on ML Street Talk.
AI
The AI company invites interpretability, alignment, and life sciences researchers to join its team over vinyl and drinks.
AI
The AI interpretability company announces access to its research tool.
