Special shoutout to our private alpha testers: @AnthropicAI, @apolloaievals, Bridgewater AIA Labs, @EpochAIResearch, @GoogleDeepMind, HAL (Princeton Language and Intelligence), MATS Research, @METR_Evals, @PalisadeAI, @penrose_labs, @redwood_ai, @thinkymachines, UIUC.

Please let us know your feedback, bug reports, and feature requests! The best way to stay in touch is through our community Slack at We’re also going to post regular updates on our changelog and blog:.• •

Join millions who have switched to Grok.

Finally, you can visualize quantitative patterns by aggregating, slicing, and filtering using Docent's charts. For example, you can plot the number of reward hacks across training steps, or compare the prevalence of reward hacking between different models.

Docent then searches for agent behaviors that match your rubric. You can click on each result to see where it was found in each transcript, as well as an explanation of why it matched.

When you ask a question like that, Docent first converts it into a precise behavior rubric by reading through your data, asking questions about ambiguities, and suggesting concrete rewrites based on your feedback.

Second, we’re introducing new features targeting types of analysis our users are most interested in. Often, users have high-level ideas about AI behaviors they’d like to measure in their data. But a user who asks "is my agent reward hacking" might mean many different things.

We’ve also made two major categories of updates based on early user feedback. The first is general product scaling: enabling access control and multi-user collaboration, scaling transcript limits by two orders of magnitude, and reducing frictions to data ingestion.

It’s easy to get started with Docent. You can use any of the following:.• Python tracing library to wrap your LLM API calls.• Native integration with UK AISI’s Inspect.• Flexible ingestion API for writing custom logic.

Learn more at our new landing page: Sign up: Check out a sample workspace: Public alpha blog post: Quickstart docs:

Docent, our tool for analyzing complex AI behaviors, is now in public alpha!. It helps scalably answer questions about agent behavior, like “is my model reward hacking” or “where does it violate instructions.”. Today, anyone can get started with just a few lines of code!

Previous fireside chat guests include @davidbau, who joined us earlier this year to discuss the need for transparency in the AI ecosystem. Check out the video! .

There are just a few spots left, and we'll accept requests on a rolling basis.

This Friday we're hosting "From Theory to Practice to Policy", a fireside chat between Yo Shavit (@yonashav) and Shafi Goldwasser. If you're local to SF and interested in the relationship between new technologies and policy, register to join! .

RT @ActInterp: Huge thanks to Sarah Schwettmann for a fascinating keynote on "AI Investigators for Understanding AI Systems" 🤖 @cogconfluen….

At #ICML2025? Come chat about investigator agents and model behavior with @ChowdhuryNeil and @_ddjohnson at West Exhibition Hall #1012, now until 1:30pm

We'll be at #ICML2025 🇨🇦 this week! Here are a few places you can find us:. Monday: Jacob (@JacobSteinhardt) speaking at Post-AGI Civilizational Equilibria (.Wednesday: Sarah (@cogconfluence) speaking at @WiMLworkshop at 10:15 and as a panelist at 11am.

RT @cogconfluence: Building a science of model understanding that addresses real-world problems is one of the key AI challenges of our time….

Transluce is hosting an #ICML2025 happy hour on Thursday, July 17 in Vancouver. Come meet us and learn more about our work! 🥂.

RT @ZhengxuanZenWu: I'll be interning at @TransluceAI for the summer doing interp too 🫡; exciting to be in SF with this bro.