In preparation for this meeting, as well as reading Chapter 7 of the Neural Networks book (see last meeting), Peter Whitham suggests we read the following short article, which critiques the current approach to AI through LLMs, and suggests a target benchmark for a real AI.

Here is another paper, this time by two professors of business schools, looking at whether present AI (Large Language Models) can support decision processes. This is important, as these results will be taught in business schools to people who have no technical understanding of what AI is doing, and need to know what it can't do.

John McKellar has just found some free courses on LinkedIn (warning you may need a LinkedIn log-in to see all the content): https://www.linkedin.com/posts/paul-storm_google-harvard-and-more-are-offering-activity-7328766154900922386-M7GG?utm_source=share&utm_medium=member_android&rcm=ACoAAAAfsy8BE-3Gr9HozZr-wKyC7LSspFNlKzw

Meeting Report

The meeting was lead by Ian Duerdoth. There were 7 members present. We discussed the agenda, and everyone present introduced themselves.

There was a discussion of sharing, and Peter was asked to find a way of building an archive of shared files in one place, so previous sharings could be found.

There is a desire to produce a Glossary that would be useful to the Main group as well as the sub-group.

Ian presented a summary of the Neural Networks book, Chapter 7, which provoked some discussion.

Peter lead a discussion of Not on the Best Path by Gary Marcus, starting from observations from John McKellar (not present) on distributional shift. A further discussion arose from the mention of copyright, and a suggestion that all human presentation was actually summarization, which was what Chat systems (LLMs) were doing when they had ingested copyright material. This is a possible subject for the main group.

Mike Doggett is preparing interesting material for the next main group meeting, which may inform how a glossary is produced.