Pinned Tweet
🤖 !! Model-free agents can internally plan !! 🤖
In our ICLR 2025 paper, we interpret a model-free RL agent and show that it internally performs a form of planning resembling bidirectional search.
Attending #ICLR2025 to present this research - please reach out if you want to chat about interp, reasoning, or anything else!!
🤖 Model-free agents can internally plan!
Sokoban agents develop bidirectional search planning!
🔬 We probe for planning concepts
⚙️ Investigate plan formation
✅ Verify plans impact behavior
Chat with us at #ICLR2025!
📍 Apr 24: Poster 10am + Oral 4:06pm SGT
Tom Bush retweeted
Preprint time:
“AI in a vat: Fundamental limits of efficient world modelling for agent sandboxing and interpretability”
arxiv.org/abs/2504.04608
Exploring the fundamental limits that shape the design space of world modelling for agent sandboxing and interpretability
For instance, in the example below, the agent initially plans to push the upper-left box to the center-most by pushing it down and then right. However, the agent realises this is infeasible – once the box is been pushed down, it is stuck – and then forms an alternate plan.
1
6
843
This research would not have been possible without the amazing support and mentorship I recieved from my brilliant co-authors @Stephen36910351 , @usmananwar391 , @AdriGarriga and @DavidSKrueger
Paper: arxiv.org/abs/2504.01871
Blog post: tuphs28.github.io/projects/i…
6
304
This research would not have been possible without the amazing support and mentorship I recieved from my brilliant co-authors @Stephen36910351 , @usmananwar391 , @AdriGarriga and @DavidSKrueger
Paper: arxiv.org/abs/2504.01871
Blog post: tuphs28.github.io/projects/i…
6
278




