Dominik Kowalski retweeted
Helping scale @AmpCode on @rivet_dev has been such a pleasure this year!
Getting to see how their team works on Amp and how they use Rivet Actors to build the future of coding is a glimpse at the next decade of engineering will look like
😲
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.…
8
Dominik Kowalski retweeted
Ataki tego typu na większą lub mniejszą skalę zgłaszają też Holandia, Irlandia, Estonia, Luksemburg i Finlandia.
xD
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
23
Dominik Kowalski retweeted
Rivet 3.0 is built on S3
But we did something completely different...
And oohhhh boy:
- 6.9 ms actor cold starts (durable)
- 0.69 ms SQLite writes (3 availability zone)
- 0.20 ms SQLite reads
Here's how we built our new storage engine, WarpDrive.
~~~
Most S3-backed database read & write straight to S3.
That means:
- Every commit & cache miss is an S3 request ($$$)
- S3 downtime takes you down (SLA allows 8 hours downtime/yr!)
- S3 Express is limited to one availability zone
This usually ends up slow, expensive, and has poor tail latencies.
For Rivet 3.0, we decided we could do better. We wanted 3.0 to ship with:
- SSD speeds with storage on S3
- Stateless nodes
- No dependencies
~~~
To do that, we built WarpDrive to durably write your data on fast disks *before* we write it to S3.
Every write on WarpDrive is:
1. Written to disk in 3 availability zones
2. Acked once a majority has them
3. Then batched to S3 in the background
That means that Rivet has:
- Writes that get durably acknowledged insanely fast
- Fewer S3 operations by batching in the background
- Survives S3 outages for new & warmed actors
- Nodes remain stateless
- S3 is the only dependency
~~~
WarpDrive is something we're incredibly proud of and can't wait to have you try for yourself.
Rivet 3.0 with WarpDrive is in preview today. It's available on Rivet Cloud or for self-hosting today.
Please send feedback on X, Discord, or DMs -- we want to hear anything and everything you have to say (please bump me if I missed a message!)
There should be no reason to vendor lock yourself to Cloudflare Durable Objects if you can have better performance, better DX/AX, lower cost, and more flexibility on software you own.
And of course, it's all open source.
Dominik Kowalski retweeted
basal-omni-pico🇵🇱 0.8B
w filmie wszystko co trzeba poniżej napisać więc nie będę pisał :) Model dostępny na HF. Oprócz tego poprawki do inference engine 2.3.2
To czego nie powiedziano:
basal-pico to pierwsza wersja - możecie się spodziewać, że:
- basal-engine 2.3.4 będzie znaczne przyspieszenie (niska optymalizacja)
- kolejna wersja będzie mniejsza a mocniejsza od tego - mam 0.6B ale widzę jeszcze jej niedoskonałości więc nie udostępniłem (model zoptymalizuję oraz trening)
Na zdrowie 🇵🇱
Kolejny wpis zatytułowałem - wszyscy mówią, że mają w domu słonia w szafie - czyli "mam najlepszy model" - co to oznacza. Teraz co drugi dzień pojawia się informacja, że ktoś ma "naj". Benchmarkuję od wczoraj wszystkie TOP modele. Raport wieczorem, z komentarzami.
Dominik Kowalski retweeted
How do I submit a model for evaluation?
Project page: basal.si5.pl/en/
Hf: huggingface.co/Remek/basal-o… and inference engine: github.com/rkinas/basal
Dominik Kowalski retweeted
The team behind @Deno_Land, including @rough__sea (he knows a bit about Node.js), is joining @Cloudflare.
They’ll help make Workers better for Node.js developers and advance workerd as a core primitive for what comes next. cfl.re/4zlzlSa
Dominik Kowalski retweeted
What most builders actually want: give us a model that's probably within a family, call it "medium" and just index it over time. Like T-shirts.
Don't make us track model releases constantly. Don't make us feel like we're mostly talking about Anthropic/OpenAI/xAI instead of talking to our customers. Let us work in peace.
"Days of Our LLMs" gets old quickly.
Dominik Kowalski retweeted
It is crazy how a technology that has the potential to cure cancer, alzheimer's, and who knows what else is so universally hated outside of rich tech.
How did Sama and Dario so thoroughly fumble this?
How did they not see telling everybody they're going to lose their jobs every few months and now we might (👉👈) accidentally kill everyone isn't a winning strategy?
Nie no jak na serwerze to git, nie było tematu 😂😂😂😂 #vibecoding
1
126
PHP + GO and you can do everything in web dev space. Those 2 langs are great companion. PHP gives you @laravelphp witch IMO is most productive web framework. Go gives you FrankenPHP and performance if you stuck with something performance related in PHP.
1
46
Hey @AirbnbHelp @Airbnb i need your help. One of your hosts is a scammer. We claim a refund of half of costs since we find out that place was full mold and condition was very bad much worse than in offer. We traveled with 2 kids and one of them starts coughing last night so
2
1
69
we checked whole place looking behind furniture and there were mold everywhere. Now this kid is sick. He said that if we will not claim refund he will not claim refund for tv which we broke(which is not true!).Its a scratch on tv and we sent photos to host after we realise what
1
45
Dominik Kowalski retweeted
🛠️ I am nearly finished with the multiple server deployments. It's already functioning locally; I just need to make some final adjustments.
With coolify.io, setting up a load-balanced service will take just 2 minutes. #buildinpublic
😮
Homebrew was good at the time, but I agree it's stale nowadays (I pushed it out in like 2009) so I built a new one pkgx.sh
109