People are flexible, creative thinkers, and we express a lot of those thoughts -- plans, explanations, stories -- in language. So, how much of thinking can be learned just by modeling patterns in our existing language? 2/
— Katie Collins (@katie_m_collins) May 16, 2022
How then do we get LLMs to generate plans that actually work? We suggest a hybrid neurosymbolic model that uses LLMs to parse natural language goals into programs, and uses a symbolic planner to solve the planning task. Our hybrid model is more robust to adding constraints. 4/ pic.twitter.com/ZjRPwgT2S1
— Katie Collins (@katie_m_collins) May 16, 2022
If you’d like access to our data, code, and/or stimuli — they can all be found at our project repository! https://t.co/1QAdziDzEL 6/
— Katie Collins (@katie_m_collins) May 16, 2022
No comments:
Post a Comment