charles
I think this is right, and I will add to it that part of why it's galling is that it's clear OpenAI doesn't give a shit about math or the math community, they only care about their product. Whatever gets broken or stepped on in the process isn't their problem.
It occurred to me last night that another way to express what's wrong with OpenAI just dumping this slop on us is that it's not peer-reviewed. If any reasonable academic did this there would be disciplinary hearings. There's a process and there's a reason for the process!
a signal to noise ratio like you've never seen in your life
What happens when there are tens of thousands of papers a day. What happens when people start using it as a tutor. What happens when it floods resources that were formerly effort limited. You barely have to think to be ahead of the consensus
Linux is fundamentally for people who like the experience of configuring and using a computer. If you don't want to use a computer, fine, but why are you running linux?
I'm watching this guy review a Linux distro that really emphasizes "ai agents" and this looks like hell. People really live like this? This is how they manage their computer experience? Failing to change their computer wallpaper and toolbar? [I cut down this SIGNIFICANTLY from 6 minutes too]
my airgapped cpu
Air gapping apparently wouldn't work in alignment tests because, I shit you not according to this OpenAI guy, one computer could run the CPU hot and tell the other airgapped one it is in trouble. I wish I could get paid $10m a year to say the stupidest cybersecurity thing you've ever heard
charles reposted
For math, specifically, Terence Tao makes a very good case that the proof is only a small point of the whole undertaking, and skipping straight to the proof is salting the earth: mathandai.org
curious what the reaction will be like when a biologist laments that a disease was cured too quickly, without giving us a chance to really mull it over
I think the correct framing here is about how many of our cognitive processes are encoded in language itself - i have started to think LLMs are primarily sophisticated representations of language itself, and what they are good at is the parts of our cognition proper to language
This is the crux of the debate, right? Whether it's doing mimicry or reasoning. I struggle to see how you could view it as purely mimicry. It has to tie together multiple novel thoughts in a logical construction to create its responses.
I have always referred to earwigs as "dinguses" and as a result my wife has only a mild aversion to them instead of being disgusted
We name all the house centipedes in our house “Michelle”: by this tactic I converted my partner from moderate arachnophobia to tenderly looking after our little squad of Corner Spiders and Towel Michelles (they like getting lost in the bathroom)
Paper is a preprint (papers.ssrn.com/sol3/papers....) but there seems to be a growing body of evidence that free range chatbot use (i.e. the typical AI use case) is actually pretty corrosive for learning
ChatGPT is also appealing to users because it is good enough and they are lazy
there really is something wrong with them you can go a very long way explaining the appeal of ChatGPT to consumers that it is a version of the internet that's not trying to sell you something (besides more of itself of course)
I used to work with Computer Touchers and now I work with high schoolers, and a high schooler on a bad day is easier to deal with than a Computer Toucher on an average day
in conversation with a high level Computer Toucher i will sometimes ask a question like "what was your intent in making thing X function this way, could we consider changing it or would you advise against that" and invariably their response is like i asked "hey fuck head your shit fuckin sucks"
charles reposted
Any adaptation of A Christmas Carol is doomed to failure because the perfect adaptation (the muppets one) has already been made
Saw someone say that Johnny Depp didn't face any consequences from MeToo and like... it killed his career? He's getting a second chance I think because chuds now run Paramount but I strongly suspect the scrooge movie - basically the first mainstream movie he's starred in since - will bomb?
charles reposted
Good LLMs often give correct answers to questions that are posed correctly - but the LLM that most people interact with, the AI results in google search - is dogshit.
gee I’m shocked. I suspected the study was completely fucked when I saw that the LLMs only gave wrong answers. LLMs usually give right answers. surely that’s relevant to people’s expectations about answer quality lol
The language in the 10-K strikes me as borderline boilerplate: of course contamination in the food chain is a huge risk to restaurants like this, and they were right to acknowledge it as such. Any reasonable person could have come up with the same guess, especially in the era of diarrhea-by-policy!
SEC filings show Taco Bell—and other restaurants—knew cyclospora posed a risk to their businesses and the public.
charles reposted
Getting students interested in the material I'm teaching has been more effective at preventing AI cheating than any other approach I've tried so far
it’s amazing how often the solution to endlessly multiplying pedagogical problems turns out to be restoring an actual pedagogical relationship between instructors and students
charles reposted
charles reposted
This was an incredibly helpful primer on what's going on with China www.nytimes.com/2026/07/14/o...