“New Ways to Corrupt LLMs” – by Gary Marcus

The problem with generative AI has always been that large language models associate patterns together without really understanding those patterns; it’s statistics without comprehension.

As a team of researchers from the University of Washington…