Paper introduces Ψ-0.6 vision-language world model with streams of thought
Original title: Ψ-0.6: Adaptive Streams of Thought in a Unified Vision-Language World Model
The authors say Ψ-0.6 beats leading open-weight models zero-shot on multi-view spatial reasoning and can simulate counterfactual scenes.
Researchers introduce Ψ-0.6, a unified vision-language world model that interleaves images, text, geometry, and motion through a shared token interface they call streams of thought. The paper reports zero-shot gains over leading open-weight models on multi-view spatial reasoning, with further gains when the model reasons through geometric intermediates. It also describes counterfactual simulation, such as judging whether moving an object would block a pathway. Reinforcement learning from task feedback is said to improve visual question-answering accuracy, with gains transferring to unseen benchmarks.
Unconfirmed: Comparisons with leading open-weight models are the authors' claims; model names and scores are not listed here.
- Streams can generate and interpret images, text, geometry, and motion.
- Counterfactual simulation covers imagined object movement and pathway obstruction.
- Task-feedback RL jointly optimizes selection and use of intermediate representations.
- Reported VQA gains transferred to unseen benchmarks.
From the original article
1 min read at huggingface.co
Humans reason about the physical world by flexibly combining different forms of inference as a problem unfolds. Achieving this flexibility is a major open challenge in AI: capabilities often remain separated across specialized models, limiting how they can be composed and jointly adapted to a task. We introduce Ψ-0.
By Khai Loong Aw, Klemen Kotar, Baihan Zhang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.