Published · Updated
OpenAI has posted more than 700 AI-generated mathematics manuscripts, and mathematicians are split over how to check them.
A hierarchical world model co-written by Yann LeCun lifts success on a visual maze task from 18% to 73% with less planning compute.
Wall Street opens the week near record highs, with bank earnings and US inflation data due before the Fed meets at the end of October.
Understanding of language models narrows like a funnel, Andrej Karpathy argued, from billions who have barely used them to about 5,000 people inside the labs who watch agents work for weeks on mathematics and science.
OpenAI published hundreds of manuscripts from an unreleased model on 6 October. Roughly 42% of the headline results have Lean formalisations, and mathematicians disagree about how to check the rest.
· Nature · OpenAI on GitHubA paper co-written by Yann LeCun trains levels of latent predictors that each look further ahead. On Visual AntMaze a three-level hierarchy raises success from 18% to 73% while using less planner compute.
· arXiv · VentureBeatThe S&P 500 rose 0.6% on Friday to 7,811.54, up 1.2% on the week and close to its record from three days earlier. The 10-year Treasury yield rose to 5.24%.
· Associated Press, via La NaciónJPMorgan, Goldman Sachs, Citigroup and Wells Fargo report on Tuesday. The consumer price report is among the last key releases before the Fed decides at the end of October whether to raise rates again.
· Reuters, via KitcoWhen one model proposes questions and another solves them, rising internal reward can reflect shared errors. Scoring questions with a solver trained on disjoint documents cuts false agreement to 3.0% and 3.7%.
Levels of latent predictors each forecast further ahead and hand subgoals downwards. A three-level hierarchy raises Visual AntMaze success from 18% to 73% with less planner compute.