A naukri.com initiative
Unite
1w
159
Image Credit: Unite
Meta’s COCONUT: The AI Method That Thinks Without Language
- Researchers at Meta and UC San Diego have developed a new AI method called COCONUT (Chain of Continuous Thought) to help language models think in their native hidden state space rather than in natural language. This method frees the model from the constraints of linguistic glue and lets it think smoothly in its own neural way. COCONUT uses two modes: language mode when it needs to understand questions or give answers and neural patterns when it needs to think in its own space. During the training process, the model developed intuitive abilities such as considering multiple reasoning paths at the same time. COCONUT has been tested on different reasoning tasks, and it showed promising results in tasks that required complex logical planning and deduction.
- Traditional reasoning approaches using language force models to translate every thought step into linguistic tokens which bottleneck their decision-making abilities and shift their effort towards maintaining grammatical coherence rather than problem-solving. Researchers found that human brains often process complex reasoning tasks without significantly engaging language centers. COCONUT's training curriculum mirrors natural learning progression, building the model's thinking abilities gradually and adding complexity as it masters each level. The method's results showed significant improvement in complex planning, achieving 97% accuracy with remarkable efficiency compared to traditional methods. COCONUT's breakthrough implies more natural, efficient, and powerful AI forms.
Read Full Article
9 Likes
For uninterrupted reading, download the app