Software developer with 15+ years in AAA gamedev. Background in systems programming, concurrency, debugging, optimization and a little audio & 3D
Toque weather
Joined July 2008
- Tweets2.8K
- Following558
- Followers4.3K
- Likes6.9K
Jeff Preshing retweeted
SOLVING is SOLVED.
ClaudeGPQwen 7 has just DROPPED and it's INSANE.
It's BROVER.
Jeff Preshing retweeted
Yes. It takes a lot of skill to design a language that is fast to compile by design, and then skill to write and optimize the compiler for that language so that it is fast at compiling too.
But I know that's not what you meant 😉
Jeff Preshing retweeted
Principal research scientist at Google's Deep Mind 👇
xitter.cf/jon_barron/status/2106…
Jeff Preshing retweeted
I’m crying of laughter - please let this be the model that solved Navier-Stokes
(this guy’s name is hayZee, we need to get him on here!)
Jeff Preshing retweeted
A modern harness + open model will crush your token consumption but keep your performance.
Been playing with DeepSeek V4 flash 0731 all day and that's how much i spent, i don't know what to tell you but i always use subs because API cost a lot but for the first time i get really great performance for almost free and the Cache hits👌. @deepseek_ai good job on this one you really delivered.
Jeff Preshing retweeted
The limiting factor of software engineering has never been code production. It is complexity management.
Jeff Preshing retweeted
Awesome #evolution tool shows when humans separated from other species: 1bn years fungi, 6m years chimpanzees. Source: buff.ly/2l0gV64
Jeff Preshing retweeted
i don't really write code anymore. i just write "write code that does these 10 things in this order in these places" and then "actually can you change thing 7 to B instead of A" and then "ok now run a smoke test"
Jeff Preshing retweeted
Wow.
@Zai_org GLM 5.2 is a marvel! It is *at least* as good as Opus 4.8 and GPT 5.5. It's super fast, inexpensive, and not too verbose.
It responds with nuance and judgement, & handles long context VERY well.
I've never experienced an open weights model like this before.
Jeff Preshing retweeted
We’ve been saying this for a while. For the first time, an overwhelming majority of economically valuable tasks people are using LLMs for have been unlocked for open source by the rising tide of open source intelligence. I often offer people a bet that they can’t tell the difference between a kimi we post-train for their complex main agent task and opus 4.8/gpt-5.5, and these days the only time I lose is when they can tell the difference because the FT kimi is so much better
AI advantage is shifting from raw model scale to deep integration with unique organizational data, workflows, and human expertise through continual learning loops. Ideally, this will result in distributed, defensible value rather than winner-take-all dynamics around general models.
We're fully aligned around building the tools necessary (world-class inference, post-training, etc.) to make this a reality for every enterprise.
Jeff Preshing retweeted
One shots are nonsense when the corpus is well documented. Try that again, but you can’t use the word Minecraft. xitter.cf/shanselman/status/2019…
Jeff Preshing retweeted
all the best programmers i know spent effort testing the practical limits of AI. they identified a sweet spot that delivers productivity improvements while retaining control of design, implementation, and cognitive ownership so they could remain responsible and accountable.
Jeff Preshing retweeted
almost hesitating to post this so as not to sound like a broken record, but yeah folks I've been saying this for months now - no production software where people are on call for downtime is being written overnight unattended by AI agents. If it is, those pager incidents are increasing at a troubling rate.
please someone prove me wrong but I'm pretty sure its all side projects / zero-to-one teams.
Even the best founders I know are eyes on the model outputs to fend off the slop. Still locked in, still using coding agents for 99%+ of code, and still shipping like crazy. There's a broad spectrum between "code by hand" and "claude take the wheel"
Talking to smarter folks than me, I'm convinced many of the AI folks in my timeline are full of shit.
Nobody is "running 20 agents over night" and building stuff for actual users. Maybe some are building internal tools or disposable software. Maybe.
But building software people like using? That doesn't get hacked on day one or blow up after the 3rd user? Nope.
I don't even understand what that's supposed to look like. Do you work out a 57 pages document that perfectly describes what you want to build and then summon 14 agents and have them run wild for 6 hours? And what comes out on the other end isn't a broken pile of shit?
Nope. Not buying it.
PS: it may also be that I have an IQ of 82 and can't figure it out.
