Grace Liu

[Michiel Bakker @bakkermichiel]

🚨📄 New preprint! We find the “boiling the frog” equivalent of AI use. In a series of RCTs, we show that after just 10 min of AI assistance people perform worse and give up more often than those who never used AI.

w Grace Liu @brianchristian Mira Dumbalska and Rachit Dubey 🧵

[Mahsa Bastankhah @MBastankhah]

We will be presenting our poster “demystifying the mechanism behind emergent exploration in goal conditioned RL” today @iclr_conf

Time: 3:15-5:45 pm
Location: Pavilion 4 P4-#3404

@ben_eysenbach @GraceLiu78 @Dilip_Arumugam

[Chongyi Zheng @chongyiz1]

1/ Reinforcement learning is usually framed as maximizing rewards. But can we cast it as reaching the right goals?

New blog on bridging RL, goal-conditioned RL, and stochastic shortest path:

https://t.co/WBpEtx9Wiu

Also #ICLR2026 Poster: Thu 10:30 AM–1:00 PM, P4 #4611.

🧵⬇️

[Grace Liu @GraceLiu78]

I’m excited to present our poster “Demystifying the Mechanisms Behind Emergent Exploration in Goal-Conditioned RL” at the Coginterp workshop at #NeurIPS2025!

https://t.co/U3OprctltS

📅 Dec 7
1:15 PM
Upper Level Room 5AB

@princeton_rl
@MBastankhah

[Grace Liu @GraceLiu78]

🙏 This work was done with my incredible collaborator @QuYuxiao and our amazing advisors Jeff Schneider, Aarti Singh, and @aviral_kumar2!

Website: https://t.co/HyU5i5mRsS
Paper link: https://t.co/EdeABnkElL
[9/9]

[Grace Liu @GraceLiu78]

NEW PAPER: "CaRT: Teaching LLM Agents to Know When They Know Enough"!

LLMs often overthink, ask too many questions, or waste compute.
We introduce Counterfactuals and Reasoning for Termination (CaRT) - teaching LLMs when to stop gathering info and make decisions.

🧵[1/9]

[Grace Liu @GraceLiu78]

Explore our paper for more insights:
⛏️ How counterfactual pairs teach models to recognize information sufficiency?
⛏️ Why verbal reasoning acts as an implicit value function for termination?
⛏️ How reasoning stabilizes decision boundaries and improves OOD robustness?

[8/9]

[Grace Liu @GraceLiu78]

🙏 This work was done with the amazing @MBastankhah along with @Dilip_Arumugam, @cocosci_lab, and @ben_eysenbach.

Website: https://t.co/GYKjHPCrpo
Paper link: https://t.co/LtiWge8ZjM

[9/9]

[Grace Liu @GraceLiu78]

NEW PAPER: "Demystifying the Mechanisms Behind Emergent Exploration in Goal-conditioned RL"

How do RL algorithms develop sophisticated exploration strategies without explicit rewards? We provide insight into this question by studying Single-Goal Contrastive RL (SGCRL).

[1/9]

[Grace Liu @GraceLiu78]

✅ Takeaway: Understanding WHY algorithms work helps us build safer, more reliable AI systems.
Emergent exploration isn’t magic ✨ — it’s shaped representations doing exactly what the optimization objective prescribes. 🎯

[8/9]