this clip from the new will anderson video is actually maybe the funniest thing ive seen in years
Very funny to me that there's an AI-boosting dipshit patrolling the replies who is completely confident that the latest high end models are that much smarter.
In case you're wondering: no, the new models aren't really that much smarter.
Here's my experience. The company I work for is run by one of those true believers in AI. We are all obligated to have AI analyze all our work, and the company pays for the latest and bestest super-premium version of Claude to do so. It's a hilariously stupid expenditure that cannot possibly be sustainable, but when the CEO rubs elbows with everyone at Anthropic and OpenAI, well, I guess we've got to be on the "cutting edge" of this garbage.
I have watched Claude analyze a bash script and have almost exactly this sort of stupidly long back-and-forth "debate" with itself about a trivial matter of syntax (which a simpler static analysis tool could validate in a fraction of a second). In the end, it concluded that the script was actually MathML, and then it spat out a screen and a half of malformed and meaningless MathML syntax in reply. This happened a month ago, not years ago.
So, yeah, "first percentile" indeed. These people don't have the foggiest idea of how an LLM actually works, just wholly unearned confidence from these machines telling them how smart and clever they are.
“Omg the LLM is so much better now guys”
Yeah it’s 12% better at the, very specific task they advertised in their latest earnings call and/or press release
Everything else it’s still shit
And even for that specific thing
Usually if you spin it a bit differently it’s suddenly shit again



















