Isabel Silva Corpus (@isabelcorpus.bsky.social)

@isabelcorpus.bsky.social Oct 9, 2026, 2:55 PM

thank you and interesting hypothesis! Like Mor suggested, we think AI suspicion was low during the period we focus on.. but this points to an interesting open question around how suspicion (and social penalties) of AI use have changed over time. Tricky to measure historically!

@isabelcorpus.bsky.social Oct 9, 2026, 2:51 PM

this is an interesting idea! author comments are rare, and also hard to interpret as a proxy for commitment... for ex. some authors comment as a reply to someone else's comment, in which case the behavior is conditioned on some level of external engagement

@informor.bsky.social Oct 8, 2026, 3:02 PMReposted by @isabelcorpus.bsky.social

🆕 📝 New paper officially out in @nathumbehav.nature.com!

In 2023, Change.org built an AI-assisted writing tool into their platform. We offer causal evidence that access to the AI tool changed petition text (word count, language)... but may have actually hurt outcomes (signatures, comments).

@isabelcorpus.bsky.social Sep 30, 2026, 2:46 PM

This was joint work with the amazing @allisonkoe.bsky.social @eegilbert.org and @informor.bsky.social

@isabelcorpus.bsky.social Sep 30, 2026, 2:45 PM

In 2023, Change.org integrated an AI-assisted writing tool into their platform. We found that access to the tool changed platform trends in petition text, but that outcomes (signatures and comments) did not improve...

Excited that this paper is out, check it out!

🔗: www.nature.com/articles/s41...

@madesai.bsky.social Sep 28, 2026, 8:08 PMReposted by @isabelcorpus.bsky.social

Excited to share our new paper, “What AI Benchmarks Actually Measure: Adapting Convergent and Discriminant Validity to Interrogate Fifty-Six AI Benchmarks,” accepted as an oral at COLM! arxiv.org/pdf/2609.08812

Heatmap of average correlations between model rankings on benchmarks grouped into 11 assigned concepts: four capability concepts (reasoning, knowledge, comprehension, summarization) and seven safety concepts (over-refusal, refusal, safety detection, ethics, bias, privacy, unsafe behavior). Diagonal cells show within-concept correlations, ranging from 0.87 (knowledge) and 0.72 (over-refusal) down to 0.20 (bias) and 0.02 (safety detection). Reasoning, knowledge, and comprehension correlate with each other at 0.69 to 0.78, higher than reasoning's and comprehension's own within-concept values (0.66 and 0.68). Ethics correlates more with knowledge (0.70) than with itself (0.55), and bias correlates more with capability concepts (0.41 to 0.45) than with itself (0.20). Privacy and unsafe behavior correlate negatively with reasoning, knowledge, and comprehension (−0.41 to −0.49). Over-refusal and refusal correlate at −0.42.

@kpaxdonahue.bsky.social Sep 1, 2026, 2:23 AMReposted by @isabelcorpus.bsky.social

I’m recruiting students this upcoming cycle at UIUC CS. I’m excited about Qs on societal impact of AI, especially human-AI collaboration, agentic teams, and benchmark design. Also, check out my grad seminar on these topics! tinyurl.com/cs598aisinth...