iiinova — where your digital self keeps evolving

Caroline Wang

@caroline-wang.bsky.social

PhD Student in MARL @UTCompSci | Student Researcher @GoogleDeepMind | ex @SonyAI_global @dukecompsci | carolinewang01.github.io

Posts Replies Media

Caroline Wang @caroline-wang.bsky.social · 16/02/2026

[9/n] Huge thanks to @pcastr.bsky.social, Daniel Kasenberg, @neurokim.bsky.social and everyone in Montreal 💙 I had an amazing experience exploring topics in behavioral game theory, cognitive science, and AI-driven scientific discovery, together with brilliant colleagues.

3150

Caroline Wang @caroline-wang.bsky.social · 16/02/2026

[8/n] For me, it’s really cool that this aligns with the jump in theory-of-mind capabilities in recent LLMs (since opponent modeling in IRPS is basically a type of ToM)

180

Caroline Wang @caroline-wang.bsky.social · 16/02/2026

[7/n] So what were the insights? Both humans and LLMs use value learning + opponent modeling, but frontier models maintain more sophisticated opponent models (3x3x3 transition matrices vs simple size 3 vectors tracking of prior move frequencies).

181

Caroline Wang @caroline-wang.bsky.social · 16/02/2026

[6/n] Using this approach, we get actual programs that explain the behavior, which we can read and compare. Diagram for human program shown below.

170

Caroline Wang @caroline-wang.bsky.social · 16/02/2026

[5/n] But what does the difference in win rates actually mean? To understand, we used AlphaEvolve to automatically discover interpretable behavioral models directly from gameplay data.

170

Caroline Wang @caroline-wang.bsky.social · 16/02/2026

[4/n] Frontier models (Gemini 2.5 Pro/Flash, GPT 5.1) win more and adapt much faster than humans, while smaller models like GPT OSS 120B actually get worse over time because they can’t integrate the long context.

170

Caroline Wang @caroline-wang.bsky.social · 16/02/2026

[3/n] So how do their strategic behaviors actually differ from humans? We examined this question through the lens of behavioral game theory, using iterated rock-paper-scissors (IRPS).

170

Caroline Wang @caroline-wang.bsky.social · 16/02/2026

[2/n] LLM agents are everywhere now: customer service, negotiations, even as human proxies for social science/market research

160

Caroline Wang @caroline-wang.bsky.social · 16/02/2026

[1/n] Just wrapped up 7 months interning with @pcastr.bsky.social at Google DeepMind and I'm so excited to share our work: arxiv.org/abs/2602.10324. TLDR: We used LLM-powered program synthesis to automatically model and discover differences between human and LLM strategic behavior

28114