Between nationwide system outages, the academic war over Pangram, and trying to communicate with the various reviewersโ AI models for ICLR, the ONE thing that Iโm obsessing over today is how my boi @NoahKahan was totally ROBBED ๐คฌ
Today at #INFORMS2026 (11.15am PT at Moscone 312) I'll be presenting our work on "Using AI Agents to Automate Black-Box Audits of Personalization Algorithms at Scale", which appeared earlier this year at EC2026 1/n
Kudos to @ballotpedia for making a quick update to allow ChatGPT to cite their content, right as undecided voters are tuning in.
Ballotpedia is a critical resource, especially for downballot races -- and now it's back into a third of ChatGPT election answers.
@CNN did an interview with @AdamBerinsky and I on the LLM Election Observatory (https://t.co/Bq7I17Y9ji) where we discussed among other things what we would like voters to know when they are interacting with chatbots for election information. Check it out below!
In our experiments, adding CoT-derived groups reduces Brier error by 4.5โ15% compared to topic-only multicalibration, while preserving similar topic-level calibration. 5/n
... pointing out (hopefully kindly) to authors when they have violated the norms and standards that I hold authors to, and / or even refusing to review papers that do so. 10/n
I've been thinking a lot about AI slop lately, and there are a few thoughts that I wanted to document, not because they are finalized and wonโt change, but just in case others find them useful. 1/n
4. I donโt have a prescribed path ahead. But my path is: implement a โstandards and expectationsโ guide for AI use for my lab, encourage frequent discussions with my students about the โproperโ (in my view, at least) use of AI in research, ... 9/n