Anthropic’s Claude AI can digest an entire book like The Great Gatsby in seconds
arstechnica.com
arstechnica.com
We've been trialing it at work and found that it's easy to accidentally bully it into absolute uselessness with a couple wrong words in a follow-up prompt.
It's good, don't get me wrong, but as an example: We brought the slack bot in. One guy said to it "explain DevOps to me like I'm five". The answer was decent but the language would have been beyond a 5 year old. So another person followed up with "that doesn't seem appropriate for a five year old". Claude agreed and promised to never try to explain advanced topics to children. Now we can't get ELI5s.
Obviously that's a silly example but I'm not sure I'd trust its alignment with a live system the way I can with GPT3.5/4.
Edit: that's exactly what theh did, gg has 72k tokens according to the article (wc says the project gutenberg text has 51257 words) and the new model can do up to 75k (actually 100k tokens see below).
I can't wait until someone builds an AI that does this for bills under discussion in the US legislature.
Can you imagine no one being able to hide anything deep in ten thousand page bills? Even the least tech savvy representatives will be able to understand what the bill says and how any changes to it impact what it means.