Lukas Kubis retweeted
i see we are continuing our fast & furious era of naming vulns like sequels no one wants. 2Shell. 4Shell. 2Fast 2Shellious.
Lukas Kubis retweeted
If you’re still building software factories instead of software industrial parks or software economic zones
ngmi
Lukas Kubis retweeted
Nov. 18, 2026, 5:16am EST
The Voyager 1 spacecraft has reached a distance of one light-day from Earth.
Lukas Kubis retweeted
repeat after me: oneshotting a threejs "game" is not indicative of real world model performance.
i'm tired.
Lukas Kubis retweeted
Jai compile times are so fast Claude can't believe it and has to double-check it's for real.
When I asked it why it double-checked - it its sense of "normal" compile times comes from C++ and Rust.
Even LLMs are amazed by what good engineering can achieve on modern hardware.
Ok I know @bennjordan also built a version of this but I need multiplatform and need it NOW for... testing purposes
1
10
GPT6.1 models still raise the issue of undefined USBC connector SHIELD termination / missing net on the PCB.
I'm assuming the same people that would tell you to use these models for serious hardware engineering would also complain about this. Good litmus test.
16
if the reward function targets solving problems in complex codebases, simple codebases will observe complexity emerge with the solutions. ironically enough, decreasing agent's agency with constraints can help sometimes, at the cost of lost capability
Instead of removing complexity, we add more levels of abstraction to cope with it. In theory AI could help with this - find non-essential complexity of software, taxes, laws, and then help removing them, making lives of everyone easier.
But instead its default reward model is to absorb the complexity and add much more bloat on top of it. So if you dare to remove it you are just left absolutely helpless.
And very few people can see through it now, and even less people will be able to in the future.
10
hey @FFmpeg I have an idea for a new codec
12

























