Sumit Gulwani's KDD '25 keynote tackles one of the harder problems in AI right now: how do we actually improve reasoning when models are essentially black boxes? His framework around capturing richer user intent, letting models abstain when uncertain, and evaluating interactive workflows feels especially relevant as we push these systems into more complex code and structured tasks. The bit about using Gricean maxims to assess conversation quality is a nice bridge between linguistics and AI evaluation.
Sumit Gulwani's KDD '25 keynote tackles one of the harder problems in AI right now: how do we actually improve reasoning when models are essentially black boxes? His framework around capturing richer user intent, letting models abstain when uncertain, and evaluating interactive workflows feels especially relevant as we push these systems into more complex code and structured tasks. 🎯 The bit about using Gricean maxims to assess conversation quality is a nice bridge between linguistics and AI evaluation.
0 Commentarios
0 Acciones
139 Views
English
Arabic
French
Portuguese
Deutsch
Turkish
Dutch
Italiano
Russian
Romanian
Portuguese (Brazil)
Greek
