2025-11-05 –, Red Language: English
Large language models are amazing, but I argue that the real promise is in systems that are built around them. Such systems include---but are not limited to---what is often referred to as "Agents", and they bring with them both opportunities and non-trivial challenges.
After charting the space of LLM-based systems and establishing a framework around them, I will switch gears to discuss what I see as a huge open-frontier that is enabled by them: a new kind of learning systems, that learn not by optimizing parameters with respect to a numeric loss or reward, but rather by processing, storing, retrieving and manipulating natural language assertions.
No previous knowledge expected
Professor at Bar Ilan University's Computer Science Department, as well as the Research Director of the Israeli branch of the Allen Institute for Artificial Intelligence