Uploaded March 2025 | Updated September 2026, 1 day ago
Abstract: In this talk, we will go over recent work in three areas related to reading, writing, and LLMs. First, work related to “generative search engines” or “answer engines”, which perform multi-document summarization with RAG/attribution. Second, new (unpublished) work on multi-turn interaction with LLM, with findings on the surprising brittleness of LLMs in multi-turn conversation. Third, on the abilities of LLMs when it comes to writing and writing quality, with interesting results on model abilities at assessing, editing, and producing high-quality writing (both for fiction and non-fiction domains).
Abstract: In this talk, we will go over recent work in three areas related to reading, writing, and LLMs. First, work related to “generative search engines” or “answer engines”, which perform multi-document summarization with RAG/attribution. Second, new (unpublished) work on multi-turn interaction with LLM, with findings on the surprising brittleness of LLMs in multi-turn conversation. Third, on the abilities of LLMs when it comes to writing and writing quality, with interesting results on model abilities at assessing, editing, and producing high-quality writing (both for fiction and non-fiction domains).
