uzzi38 retweeted
Nice DeepSWE score! Does it survive a lowercase filename?
Introducing MaxxBench: 89 DeepSWE tasks with small changes to the instructions or starting code. Same engineering problem, slightly different requirement.
Models clustered around 74% on DeepSWE end up 23 pp apart.
Good at coding, or good at DeepSWE? 🧵
uzzi38 retweeted
Watched the latest @LinusTech video and they are running QWEN 3.8 27B with pipeline parallelism GGUF on LM studio Q4KM and complaining it isn’t fast.
Oh my eyes.
Please PLEASE someone who’s close to
him educate him on VLLM and Strata.
He could literally be getting 5x better speeds and better intelligence.
Sure the setup is complex but it’s a one time setup with a cloud agent.
uzzi38 retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
uzzi38 retweeted
We're going to have to up the security in our facility... caught this man trying to steal RAM earlier.
uzzi38 retweeted
Thanks to @copikmarcin for bring this to my attention, it looks like @Tachyum may have filed for voluntary bankruptcy.
forbes.sk/firma-patrila-k-na…
Well... it looks like it is the beginning of the end for @Tachyum...
They are closing their head office in Slovakia due to unpaid debt...
This is after a report that the Czech branch is currently being sued for failure to pay empolyees...
e.dennikn.sk/5131152/tachyum…
uzzi38 retweeted
Hi, I'm free.
Beat up but fine. Police stole my phone. Settlers punched me in the face several times, likely broke @AdeleShoko's wrist and gave @the_andrey_x a concussion. They threatened to shoot us with an assault rifle over and over.
Will tell the full story later, but for now, I just want to be clear about what happened. We witnessed two settlers locking a military gate in Umm al-Khair, blocking the only way to more than twenty Palestinian communities: trapping families inside, preventing students from reaching school and workers from their jobs, blocking delivery of food and access to ambulances, etc.
It's a crime that the military has this ability at all--but last night, we discovered that random settler terrorists also have the key. We confronted them and began asking questions on camera, and they nearly murdered us for it. When the police finally arrived, they arrested Andrey and me and charged us with assault.
The gate is still locked.
uzzi38 retweeted
New Duo users about to get a reality check on why matte finishes on OLED (AMOLED) is bad. Sure, it hides the crease, but that looks AWFUL.
If they sold a gloss variant like with MacBooks that would be better.
Photo from the Verge's review. Multiple angles look bad.
uzzi38 retweeted
Fucking hell, that ITV investigation into Reform UK is horrific.
We can’t get desensitised to this…
Reform politicians talk about Muslims how the Nazis talked about the Jews.
Anyone that fails to condemn it is complicit.
uzzi38 retweeted
I dont care that the new EGS launcher is vibe coded. I dont care of they are using GE-Proton internally. Someone at Epic had to present these ideas and get them green lit, and regardless of reasoning be it for competition or for compatibility, an effort for linux compatibility is being made, and I can appreciate that. It's more than most have done in the past and I welcome it. I want linux to grow, not for people to chase growth out.
Holy crap I didn't even think it was possible to design something worse than liquid glass.
Gotta love AI bros sometimes, always there to prove you wrong.
Because a lot of the times AI gets used well by competent developers, you can't tell a difference.
The only times you actually know are:
1. When the developer actively tells you it helps them (emulation etc).
2. When a severe lack of QA makes it blindlingly obvious.
Same as many other modern foldable phones. There is a crease that you can see if the display is turned off. With the display powered on, you can't see it. If you deliberately run your finger over it, you'll feel it.
