One of the major challenges of general artificial intelligence is to develop agents capable of performing scientific research and discovering new knowledge. Although boundary models are already used as tools for human scientists, eg for synthesizing ideas, coding, or prediction tasks, they still only drive a small part of the scientific process. This paper presents the first comprehensive framework for automated scientific discovery, which enables large-scale linguistic models to independently perform research and communicate their findings. Introducing the AI Scientist, who generates novel research ideas, writes code, runs experiments, visualizes results, explains his findings by writing a full scientific paper, and runs a simulated review process for evaluation. In practice, this process can be repeated to develop ideas iteratively in an open manner, acting as a human science community. We demonstrate its versatility by applying it to three different machine learning applications: a distribution model, a transformer-based language model, and learning power. Each idea is implemented and developed into a full paper at a cost of less than $15 per paper. To evaluate the produced papers, we design and validate an automated reviewer, which we show achieves near-human performance in evaluating paper scores. An AI Scientist can produce papers that exceed the acceptance threshold at a top machine learning conference as judged by our automated reviewer. This method means the beginning of a new era in the discovery of science in machine learning: bringing the revolutionary benefits of AI agents to the entire process of AI research itself, and bringing us closer to a world where endless creativity and innovation can be revealed to the world’s most challenging problems. Our code is open at this https URL
That appears in a new paper by Chris Lu, Cong Lu, Robert Tjarko Lange, Jakob Foerster, Jeff Clune, David Ha. Note that this relates to some earlier work in economics by Benjamin Manning of MIT (and co-authors).
I’ve said it before, and I’ll say it again. A by-product of LLMs is interacting with well-prepared, complex people at their peak, not when you ask them random questions for fun.
The post Okie-dokie, solve the equation appeared first on Marginal REVOLUTION.
Source link