shitposting, programming, hacking, arguing. building privacy infra w/zk. tee skeptic. decentralization maxi. descriptive, not prescriptive. fuck vcs, fund pgs.
Joined September 2021
- Tweets4.7K
- Following716
- Followers288
- Likes12.9K
thirk π¦ββ¬ retweeted
Technical was always a bundle:
β typing code
β a mental model of how software fits together
β knowing whatβs worth building.
Agents made the first one nearly free. The other two got more valuable, because thatβs what you steer with now.
So the sad engineers are mourning the typing and everyone else is wishing they had the mental model.
The typing was the least interesting part of the job. It was just the most visible.
Deepseek has essentially solved the big cost problems because of their insane cache ratios compared to other models.
Their context cache is something like 100x cheaper than frontier, making it easier to both run and have high context (and it makes sense to rent at 100x cheaper....)
Will be surprised if many models don't rearchitect their own models in this direction.
People think that Ahmad is glazing Deepseek V4.1 flash as he is "that opensource guy" but after spending 100+ hours digging through eval tasks and running closed source tasks through many frontier models I am starting to believe him.
I am confident that Deepseek V4.1 flash is often really on par with frontier models and even when its low by 5-8% it still cheaper by 13-20 times compared to frontier closed models.
Don't take my word for it, even on cursor bench deepseek V4.1 flash is 13 times cheaper than frontier but the only difference is that its fully open weights all the way to the inference stack.
The future of humanity is looking bright to me.
thirk π¦ββ¬ retweeted
a 10yr old CNN outperforms frontier LLMs on cancer diagnosis
it seems inevitable that similar training data will be available to LLMs, and it's surprising that's not already the case
the "long tails" of the real world, where precision really matters, continue to need domain specific training
lynchlab.net/llm-skin-cancer
First day of local AI handoff testing on real projects. Saved 4M output tokens and 6M input which would have been over $100 in API costs π₯³
thirk π¦ββ¬ retweeted
In light of all the ACTUAL exploits going on right now, people who post dumb shit like this should be beaten with a stick
It took a few days longer than I hoped but I'm finally sipping tokens instead of chugging thanks to Opus 5.5 directing a local Qwen3.8 Flash Next hand off instead of doing the bulk work itself.
thirk π¦ββ¬ retweeted
Age ratings tell parents whatβs in a game so they can make informed decisions for their kids.
These AI statements are just as partisan as mandatory political essays or forcing developers to list their voting history to drive cancel campaigns and public struggle sessions.
thirk π¦ββ¬ retweeted
i've seen a lot of posts like this lately, where people bemoan the death of their hobbies, and it seems to boil down to one exact thing: they derive joy not from being able to do something themselves, but from others being unable to do it. if a random stranger failing to answer your quiz is enough to ruin your enjoyment of the hobby or the community, maybe you don't love it as much as you think you do.
Recently attended an in-person handheld game console hackathon. I wrote a raycaster by hand. Someone wearing a βI β€οΈ slopβ tshirt saw mine and said βoh I wrote a ray caster too!β. They showed me. It was a ray *tracer* and claude did all the work. The person didnβt know basic graphics terms. They couldnβt answer basic questions about how it worked, could only talk about their prompting harness. Sad how AI is ruining the craft and community for people that actually like to program and have spent their lives getting good at it.
thirk π¦ββ¬ retweeted
i remember being surprised in school the huge push back of electricity, video cameras, spotify, cars and so on throughout history
and now i get to witness this first hand with ai and it's fascinating. the push back makes more sense emotionally when you see it first hand
thirk π¦ββ¬ retweeted
When I was writing Unreal Engine 1, my time was divided as follows:
- Around 30% went to seriously deep coding on hard problems: assembly language texture mappers, C++ BSP trees, data persistence, object oriented systems design, and networking.
- Around 70% was just typing and refactoring large bodies of code needed for a complete engine feature set. This was not hard, and while I was doing it my mind was mostly on the deeper coding topics, thinking and planning for what's next. It didn't feel much different from vibe coding today, except that it was just me typing code on an IBM Model M keyboard rather than prompting it and reviewing the results.
So, it has always seemed perfectly logical that we would rely on AI for that 70%, once the AI could do it with a similar quality standard as we could, which has never been perfection. It came as no surprise that, after a few years of AI model evolution, it was able to take the reigns on that 70% and free up programmers' time for more valuable tasks.
What's more of a surprise is the emergence of AI as a sort of insta-Carmack able to solve hard technical problems quite cleverly and often cleanly, and that it's able to reason about and work in complex 3D scenes both by operating the tools and by directly understanding the raw data files.
I abandoned assembly language in 1998 when out-of-order CPUs made much simpler C/C++ code faster, and never I missed it. If prompting through architectural documents, concept art, and design plans can transcend typing code and tweaking vertices, that will enable developers to spend their time on higher-level, higher value technical and creative tasks.
What matters is the end product, whether it's a game or an engine. Anything developers can do to try out more ideas and make better software and games faster is a win for everyone.
thirk π¦ββ¬ retweeted
The games we've long wanted, which didn't exist because they'd cost several times more to make than they'd earn back, will soon become possible. As will the games developers wanted to build, but couldn't with the resources available. It'll be a new golden age of gaming.
thirk π¦ββ¬ retweeted
can we start calling things human slop. i'm so sick of people acting like their code or math proof is inherently more valuable because they slaved away making it themselves. especially when the field is so fucking deterministic and we can actually check whether the result works.
what gets me even more angry is when it's a thankless effort like oss. someone spends their valuable time making something for you, for free, and you think it's okay to criticize how they go about it
and half the time, if they hadn't found an easier way to make it, the thing wouldn't exist at all. there isn't some "lovingly handcrafted" alternative waiting in the wings
there's just nothing.
each person being able to do way more means problems that wouldn't even have been given a thought can finally get solutions. somebody with a one-off bug in a game can go ahead and patch it themselves. somebody with a rare genetic condition could get together with a small group of people and have a real shot at contributing to research that nobody was prioritizing, without needing billions of dollars behind them just to get started.
people can finally do things they couldn't do before, including things nobody else was ever going to bother doing for them, and your first instinct is to get mad about how they did it.
i think that's so self-serving. when you'd rather something useful never existed than accept that somebody had an easier time making it than you would have, there's something genuinely ugly about that
Introducing Diorama: an OSS Minecraft minimap that shows your world in 3D, updating live as you build
Huge shoutout to @8Blits, this is heavily inspired by his map for @HallowPrison π
Out now on CurseForge for Fabric, Minecraft 26.3 & 1.21.11
thirk π¦ββ¬ retweeted
under half an hour for claude to apparently solve a decades long mystery about the seeds used in popular ecdsa curves. probably nothing anthropic.com/research/ecdsaβ¦
thirk π¦ββ¬ retweeted
The only group more insufferable than tech people is anti-tech bluesky users, who have been coming back to X in droves the last few weeks.
And theyβre just discovering that the entire world got rich over the last two years while they were busy arguing with each other about politics and AI.
And theyβre very, very angry.
thirk π¦ββ¬ retweeted
Wrapped zkAPI + Tor + extra cover traffic + local LLM prompt rewriting in an open-AI compatible daemon, and made a "paranoid mode" wrapper for my agent harness (pi):
Each conversation gets assigned a separate Tor circuit, your requests are masked with random cover traffic, zkAPI ensures you don't leak your identity through payments, and all prompts are locally rewritten (bring-your-own-local-model; I'm using qwen3.8-flash-next)
github.com/vbuterin/zkapi/trβ¦
My agent harness is wrapped with a bwrap sandbox (also experimenting with MicroVM) and wrapped again with torsocks, so everything it does goes through Tor.
Here's what me asking it to do a simple research task looks like.
It's getting quite usable! Though you definitely still pay quite a bit in increased latency, especially when starting a new conversation.
Also, Tor is far from optimal here for privacy; the cover traffic mechanism is okay against passive adversaries but much weaker if the LLM provider is colluding with the guard and actively manipulating timing. This kind of task really requires a more specialized mixnet, probably take the tradeoff of a higher bandwidth multiplier and built-in constant cover traffic.
Also also, much cheaper zkAPI deposit and withdraw gas costs are coming to mainnet soon.
thirk π¦ββ¬ retweeted
I expected this latest Epic AI announcement to surface some new interesting anti AI arguments and insights but I can feel the regression in quality of the anti AI camp
the main argument is βAI coding is bad qualityβ π
this isnβt even close to even ground any more. GG gamers.