APPLE STUDY EXPOSES DEEP CRACKS IN LLMS' “REASONING” CAPABILITIES

Summary from the TLDR Web Dev newsletter

A new study by Apple engineers revealed that while LLMs may appear to reason mathematically, they are actually prone to failures when confronted with seemingly trivial changes to benchmark problems.