FQxI's Max Tegmark and colleagues have created an open-source tool for checking that code is free from bugs: https://github.com/sun-wendy/DafnyBench
In a separate AI study, Tegmark and other colleagues report a surprising geometric structure in LLM-learned concepts. On X/Twitter, Tegmark notes that they form brain-like "lobes," and "semantic crystals." Read more in their preprint on "The Geometry of Concepts": https://arxiv.org/abs/2410.19750.