Theme of inquiry

How do dialogue systems achieve genuine conversational grounding?

A question within its area, explored through 5 lines of inquiry below — each a family of specific questions the research asks.


Do language model representations contain causally steerable task-specific features?

20 specific questions

See all 20 questions in this line of inquiry
How do transformer attention mechanisms implement memory and algorithmic functions?

31 specific questions

See all 31 questions in this line of inquiry
Can language model hallucination be prevented or only managed?

20 specific questions

See all 20 questions in this line of inquiry
What structural biases does transformer attention create in language model outputs?

25 specific questions

See all 25 questions in this line of inquiry
Should GUI agents use structured representations instead of raw pixels?

24 specific questions

See all 24 questions in this line of inquiry