opt retweeted
I can't believe some X engineer actually sat down and implemented "play a random video as soon as the video you clicked on ends." If you implemented this without an opt-out, you should be ashamed of yourself.
opt retweeted
Who called it losing chain of thought monitorability through opaque internal reasoning and not “weight, weight, don’t tell me”
opt retweeted
The more time I spend working with coding agents, the more convinced I am that they make software engineering even harder
We can do amazing things with them, but unlocking their full potential requires extraordinary discipline and knowledge
Bury your head in the ground at your own risk. I aim to not jump on any hype trains, but since Opus 4.6 and GPT-5.2 + the harnesses it was clear that these things can write code nearly as good as I can in my best language; better in other languages.
But we won't see non-devs push production code for a long time; possibly forever, if you ask me. Software still needs to be built in robust ways, and it's our profession to do this well, and use the new tools we have, which create new and interesting (+tough!) challenges
opt retweeted
Knowledge is power hence inherently dangerous. To mitigate the danger we propose gathering all knowledge around the globe, legally or otherwise, and compresing it into a monolithic black box that we can guard from bad actors. Luckily for you, we are the good actors of course.
opt retweeted
THIS.
You idiots want "more time" but you wouldnt be able to fill 100 years with a single worthwhile idea.
opt retweeted
i don't see enough of this, so here again:
we're all wondering what the future of work looks like with AI progress, but I'm in constant agony trying to extract OK work from frontier models at "Ultra" settings
with levels of detailed context and elaborate feedback that i’ve never given a human in my life (… considering that my job is literally to give feedback)
at least up to Sept 2026, public frontiers models are - as they were in 2020 - systems that know a whole lot but are fundamentally quite dumb and unadaptive
opt retweeted
If LLMs solve ∃ but fail at ∀, I'm not convinced math academia is going to change much. Perhaps conjectures becomes less common. Less negative results go through peer review. But largely it can stay the same and whatever reasons were supporting it before will remain valid.
okay so. when a mathematician proves a random conjecture you are supposed to politely clap. this is because it would be rude and embarrass the mathematician to point out that most conjectures are not that important or meaningful
when an LLM proves a random conjecture mathematicians are caught in a funny position where they want to say "these are not very important or meaningful conjectures and having magic proofs or counterexamples come out of nowhere is not particularly interesting or useful to anybody, except possibly the person who proposed the conjecture and some of their colleagues" but saying that puts them in an awkward position - why didn't they say that when humans were doing it? was nearly all mathematics actually pointless this whole time and mathematicians were just politely avoiding ever pointing this out so they could all still have jobs?
if the best defense you can come up with of mathematics as a human endeavor is, effectively, "it's really fun for us to come up with these baroque puzzles and then solve them, don't screw this up for us," you are completely abdicating your responsibility as a public endeavor that taxpayers fund. i have occasionally argued that pure mathematics lost the plot decades ago and gave up on having any kind of meaningful relationship to the rest of the human quest to understand reality. this was a slow-burning tragedy, and mathematicians have been able to hide this state of affairs from everybody else because nobody else knows enough mathematics to critique what mathematicians are doing
i saw a famous professor point out a few years ago that the machine learning that underpins AI is powered by 18th or maybe 19th century mathematics, you don't need much more than multivariable calculus, linear algebra, a bit of probability. he said this as if it was a dunk on ML and AI; imo it's a much bigger dunk on mathematics. you're telling me the most important scientific revolution of our lifetimes doesn't involve any of the math that mathematicians have been working on over the last century or two? you see how that's worse, right?
(tbc i'm not saying that stuff is all useless, you need a bunch of stuff from the first half of the 20th century to do physics, e.g. functional analysis, lie theory, representation theory all feed into the standard model. after 1950 or so things get more dicey)
if mathematicians still want to have jobs they are going to have to do a better job than they have for the last few decades convincing the public they do something that actually matters, and figuring out what that actually is or should be. this is an opportunity for the field to do some very serious soul-searching, as serious or more serious than the foundational crisis from a century ago. this kind of whining is absolutely not going to cut it
opt retweeted
Sometimes I look out over a body of water and think about pixel shaders — superimposed waveforms, fresnel effects, intra-pixel maximum finding and analytical anti-aliasing. In the age of gen-AI rendering, this is like the old mechanics working on WW2 era piston planes. A craft of a prior era.
opt retweeted
Something happens to my brain after agentic coding that I can’t describe. It’s like cognitive offloading which folks have already written about, but even more. It feels like I can’t think through problems anymore. Like a fog. Using agentic but losing my hard-won agency.
opt retweeted
every new hype cycle there are obvious looking claims that turn out to be very wrong. this is the main one with ai
main culprit is equating "things that look like games" with games
the consensus, the shared reality, the formal constraints ... all the things that make a game are missing in improvised pixel slop, even if world is persistent (or even multiplayer)
and once you add them back in, you have game made with software 1.0 (note that board games are also software, just not digital)
