System 2 Thinking

This month we'll look at some of the recent models that claim better reasoning and we compare some of the approaches

"Ignore all previous instructions : A different training objective" - Martin Andrews

Martin's talk will come in three stages, for beginner, intermediate and advanced. Firstly, a little about LoRA fine-tuning models; then a simple (effective) idea "do-over training"; finally a discussion of how training of a model that does inference-time computation might be a new paradigm.

"Daily AI Papers - using GitHub and Gemini" - Gabriel Chua

Gabriel will give a lightning talk about an automated tool that he built for scanning new arXiv research, creating summaries and publishing - using GitHub actions, Gemini models and GitHub pages.

"What's behind o1 and can it be replicated: Building a Poor man's o1" - Sam Witteveen

In this talk Sam will look at some of the ideas around what makes up the latest o1 model and how some of these ideas can be replicated in other models with a few hacks and modifications.

Follow-up on a previous speaker:

@niceaunties is part of an exhibition at Singapore's ArtScience Museum during September : Here's her post