I feel like even in that specific task, it's only really good when you care more about processing huge amounts of data over accuracy. I wouldn't trust it to sum

I feel like even in that specific task, it's only really good when you care more about processing huge amounts of data over accuracy. I wouldn't trust it to summarize a single PDF research paper, but it would be fine as a search engine looking over millions of documents, or trying to summarize the common points in thousands of user feedbacks. Something like "Find me documents that talk about achieving x using y process". If it finds a few irrelevant ones or misses some, its ok and still provided value by searching far more things than you ever could manually.