Tangentially related - I've been working on a project to 'harden' science degree work against LLM's of late (which I thought was doomed at the outset, but there we go). I had a few runs of our 'harde… (View Post)

The censorship of the web & app interface seems to be a layer 'above' the model. It'll quite often generate something contentious for a second or two then give the "Oh dear, something has go… (View Post)

Indeed. I had a cover letter from a job applicant recently which began "Certainly! I can help you write a cover letter!". Some people - even AGI isn't going to help. (View Post)

It isn't half. That must have been hard to write - never mind record to camera. (View Post)

I find the problem is it's sometimes brilliant, yet wrong, yet convincing. As stands - even the SOTA 'reasoning' models need a lot of oversight (whether by humans or other LLMs). Which is annoying. (View Post)