Fine-tuning VLMs to output action sequences for puzzles causes emergent internal visual representations that improve performance when integrated into reasoning.
Emergent representations of program semantics in language models trained on programs
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
citation-role summary
method 1
citation-polarity summary
years
2026 2roles
method 1polarities
use method 1representative citing papers
LaGO improves online RL success rates over vanilla PPO by using pretrained LLMs as latent action priors, raising rates from 15.1% to 27.2% on CLEVR-Robot and 2.7% to 15.2% on Meta-World.
citing papers explorer
-
Do multimodal models imagine electric sheep?
Fine-tuning VLMs to output action sequences for puzzles causes emergent internal visual representations that improve performance when integrated into reasoning.
-
LaGO: Latent Action Guidance for Online Reinforcement Learning
LaGO improves online RL success rates over vanilla PPO by using pretrained LLMs as latent action priors, raising rates from 15.1% to 27.2% on CLEVR-Robot and 2.7% to 15.2% on Meta-World.