Tootfinder

These are very simple tasks in each domain, shown here for language. So if you are asking an LLM to read thousands of pages, it probably isn't and won't reliably count results for you at all. #AIfailuremodes

It's become commonplace to say that "LLMs can handle large amounts of data," something that is true for other computer methods including ML.
When actually tested, however, LLM performance collapses as scale increases.
#generativeAI #AIscience #AIfailuremodes