Willison: “No model has beaten GPT-4 on a range of widely used benchmarks like this.”
They all claim to have “near-human” abilities.
“Near-human” is marketing speak for “not as good as a human and there is no measurable scale to say how close it is so we will say it is closed”
How big of a paycheck did the “journalist” get paid on this one?
Claude didn’t get paid shit
Claude writing self-promoting articles about Claude? I’d believe it!
Just gave Claude a try and although there are similarities with the other AI and how it “feels”, it responds with a certain depth and “open-mindedness” that I don’t think I’ve experienced with the other ones. Planning on playing around with it for a couple days to see the range of its helpfulness.
I don’t have access to whatever their latest public model is and don’t know if the one on their website has updated in the past few months, but it’s by far my favorite AI model for generated text. Out of the stories I’ve had it generate, it’s by far the best compared to Perplexity and ChatGPT, at least for my standards.
The devs and shills for Claude also claimed it to be able to analyze a full document and give results, but it can’t, it just lies to you and says it can and then posts no download link for the results it said it wrote out.
