According to Google, it found a better matrix multiplication algorithm as well as improvements on other open problems.
Although it isn’t represented in natural language, it seems to still technically be an LLM… :(
You must log in or # to comment.
It’s a combination of LLMs with what’s basically a genetic algorithm where it evolves towards a solution space over time. I expect that this sort of stuff will be how AI research evolves going forwrd. LLMs have useful properties, but they need to be combined with other approaches to get really useful results.
