the breathless reporting on datacenters has escalated to “datacenters are dropping 18 hiroshima scale nuclear bombs on their communities every day”

this feels more urgent every day

claude made this in 10 seconds and it felt important to share

the fixed rate mortgage is kind of a fundamentally insane product

basically all vulnerability research going forward will be done by AI models

this subreddit of people complaining more than ever about hitting their claude rate limits after paying to use claude shows i was right 2 years ago that all this ai stuff is going to go away and no one uses it

the bevy game engine is amazing

neural networks are produced by optimization and not manual programming and as a result no human on earth completely understands the internal structure of large neural networks

i don’t think any of these folk theories about AI or reflexive hatreds survive 5 minutes of claude code usage

AI data centers are literally sucking the oxygen out of the air, causing people up to several miles away to be short of breath and suffer from extreme headaches

in addition to being able to design custom machine-prompted feeds, i think LLMs are great at providing context around highly viral posts. frontier LLMs like Claude have great reading comprehension and are impersonal/not affected by parasitic social media dynamics.

if you are mad about missing features you can just send a pr 'i cant code' you can install this cool tool called Claude that can spin you up on any codebase

how do i tow the line of speaking out against stupid technology/AI takes on here and being 'big account with domain knowledge dunking on small account' (negative affact) or getting flanderized into this being 'my thing'

street reporter who interviews people who 'hate ai' (they hate 2023 era stable diffusion culture war slop) where you claude code an idea with them on the street with a laptop

this is deeply embarassing for the people blocking it

One of the reasons that the willful blindness about AI from people with platforms on the left is shortsighted is that “hope it goes away” doesn’t have a great track record as a strategy for contesting control of new political domains and technologies.

i suspect that not only are the climate impacts of chip + datacenter buildout quite small currently (comparable to other industries), the _net_ impact can be extremely positive through better modeling of climate data, weather prediction, and so on

good article www.techdirt.com/2026/03/26/e...

under the KGM v. Meta ruling, can bluesky be held liable if someone transmits a custom feed that is deemed “addictive” or “harmful”?

“In constructing certain feeds, those platforms make choices about what third-party speech to display and how to display it. They include and exclude, organize and prioritize — and in making millions of those decisions each day, produce their own distinctive compilations of expression” Elena Kagan

the way some US politicians and commentators talk about section 230 drives me insane. they want to “own” “Big Tech” so badly that they will rip out foundational law that protects the freedoms of _individuals_ because they think, wrongly, that doing so is inconvenient to “Big Tech”

>”ai bubble is popping” post >ask if the evidence is capabilities breaking trend or random product announcement >doesn’t understand >pull out chart explaining capabilities vs product announcement >laughs and says “it’s good evidence sir” >look inside >it’s product announcement

cool to see that claude code has model-planned tool call policies now with —enable-auto-mode

ok i think i get it now, these people think that AI is literally just “annoying slop video i saw in my feed”

claude

DLSS was always a machine learning model! it was always 'AI slop!' it's just that now the models are getting big enough that you're seeing the obvious hallmarks of high scale deep learning

I'm declaring this no longer a work in progress, it is in fact very good www.andymasley.com/visuals/water/

incredible stuff

traveling back in time to scream at leibniz that he had already discovered the universal characteristic (the chain rule of calculus)

gaming exists as a niche activity resulting from hyperabundance of computer. computers are for physics simulation, research, machine intelligence, cryptography, and communications.

overwrought skepticism of deep learning is _old_. it was the mainstream position in academia across AI and linguistics and other places that “neural networks are interesting, but can’t work for anything complicated. explicit symbolic manipulation and grammars are better” for 50+ years.

even under extremely generous assumptions, the “AI is drinking all the water” thing is empirically fake. (infographic from @andymasley.bsky.social )

if you're faced with a decision: (1) use AI/machine intelligence/other mechanisms to accelerate biotech research by 2, 5, or 10x (2) block the use of the above, to preserve the job of 'humans looking through microscopes' choosing (2) seems like a completely immoral decision.

SaaS is imploding because people are realizing that many SaaS companies are effectively a prompt.

blocking waymo and other advanced autonomous vehicles directly kills people

sitting down with a claudeswarm and having it autopilot a clean-room WASM vm in 30mins, and then compiling your own rust programs to run in it, is a good way to dispel illusions about where we are right now with these systems

people rightfully make fun of “gork is this true” but the fact at this point is that frontier models have much better reading comprehension and world knowledge than the average social media user

that sounds weird and like something a nerd would say, therefore it can’t be real

the post proves its point; people are reacting to this as if it’s some outrageous claim instead of just a literally accurate description of how large deep learning models work, one repeated by every researcher and scientist, including both the 2024 chemistry and physics nobel winners

still working on these but also pulling heroics elsewhere so it’s gonna be a bit longer

tfw your game is too autistic so the united states government says “no”

i’ve become more sympathetic to the suttonesque view that true “intelligence” is a result of an agent interacting with some hard, potentially unpredictable, and unfoolable environment, rather than just the agent weights itself

i think wider ai discourse would be wildly different if everyone actually understood that Claude and ChatGPT are “eldritch creations via petajoules of stochastic optimizations” rather than “chat bot programmed by a san francisco tech company’s engineers”

it’s interesting how western fanonist thirdworldists will simp explicitly and implicitly for the state planning policy of the ccp, meanwhile the actual lived state of that policy, 内卷, closely matches everything they claim to dislike about american capitalism

presented without comment

giving claude code a fpga hooked up to a bunch of random devices and having it autonomously vibecode new verilog on to it every day. the Thingamabob

2020 gpt-3 era was the most future shock i remember feeling in a long time. it was insane at the time

i have at least two more posts like this in the pipeline btw :) one on compositional jailbreaking, and another on classifier extraction

i hope bsky can have more of a research presence than X, i’ve been posting more here to try to foster that