[Adapt] [Seminar] In-context Knowledge Editing

[于赟皓]

Hi Adapters,

Since large language models (LLMs) may contain outdated or incorrect knowledge, how to effectively correct that knowledge is an important issue. Earlier knowledge editing methods like ROME will try to modify the model’s weights to accomplish that. However, these methods may not apply to the scenario of LLM, due to the computation overhead. Another potential solution to this problem is to use in-context learning (ICL).

In this talk, I will mainly focus on the paper titled "Can We Edit Factual Knowledge by In-Context Learning?". In this work, the authors explored strategies for modifying factual knowledge using ICL and compared them with gradient-based methods. In this talk, I will share their methods and experimental results with you.

Hope you find this talk interesting!

Time: Wed 10 am. - 11:30 am.

Meeting link: 
https://teams.microsoft.com/l/meetup-join/19%3ameeting_M2VmMTU5MzgtODUzOC00NmU4LTg0MzktNGFjNDdiMmIwYTI1%40thread.v2/0?context=%7b%22Tid%22%3a%225cdc5b43-d7be-4caa-8173-729e3b0a62d9%22%2c%22Oid%22%3a%221a8b9fa0-af57-4a1c-9390-22d1c201d622%22%7d


Best wishes,
Frank