Can Bölük
@can1357
BDFL @
I break software for a living, giving back by making LLMs break em less. Lifelong ring-0 resident, LA57 fan, drew a ▲ with dx12 once.
552 Following 21.3K Followers
we should have just given agents an unpriv user on our PCs. it'd keep 99% of the isolation boundary you want without any overhead & every single OS is already multi-user. /home/claude, done.
sandbox/vms add too much friction, obviously everyone ended up with "yolo" (or auto-review which is pretty much "yolo")
Show more
bash usage is up from 20% earlier this year to 50%, minified claude spew is not so much fun, but a little visualization goes a long way
we'll have some interesting things out in the next few hours 🤞
Show more
what an interesting idea
My fave addition to t3code. ability to mention the model inside a harness or provider.
> add graph components
> wait 24h
> Grafana is ported over
incredible
added draft-7 support 🫡
people of pi. typebox is no longer a pi dependency. you can still use it for tool args etc., but there's really no good reason. updated benchmarks.
Message threads are now in public beta on the Messages API.
The API caches your conversation on the server between turns, so instead of resending the full conversation, you send only new messages and the previous message ID, and Claude starts responding sooner.
Show more
This is how you know you fucked up your rate limits (guy randomly scrolling hits it, and no I'm on 5G so not sharing the net w agents)
Surprised this shit gets past QA.
Web client should be out in ~30min 🫡
I asked Kimi to enable ipv6 in the background as I was doing something else.
Turns out it was not enabled on my router so, not having the admin password, it naturally then found the firmware online, decrypted it, extracted the lua bytecode, found out the opcodes were shuffled, updated the decompiler to work with it and found a stack-overflow in a binary it launched to get access.
Still didn't succeed though, since it figured attempting to trigger it would be ""crashing = disruption"" 😭
Show more
i was curious as to why the claude code users were never reporting the "claude uses bash to edit files" issue.
apparently they take a git snapshot before every bash call and produce a fake edit view?
fix the RL man 😭
Show more
I'm happy to announce our inclusion into the Inverse Theo Index. Today I stand on the shoulders of giants: there has not been a day I've been more confident.
jev's quite good indeed!
it will be available as a provider on our next release and all our classification tasks should support it (thinking level routing, unexpected stops, etc.)
+ will be adding a new `judge` primitive in the py/js kernel for agents to invoke it programatically--also supporting regular models as its backend so a useful primitive regardless of availability
Show more
with omp 18.2.3, you can mention a model with ^ to turn it into a spawnable subagent!
we received many requests for allowing the model to be specified in Task invocations, but accepting arbitrary IDs are tricky: agents then end up picking outdated/arbitrary models when it's not specified + it also opens up the ability to use, say, openrouter over a codex subscription.
it's rather experimental, but let's see how folks like this take on the UX!
+ as it's not widely known, little protip: you can press alt+p twice to change the model backing the default subagent type!
Show more
> eu model
> glm 5.2 base but benches worse
> quantum btw
you can't make this shit up
⚛️🇪🇺 Introducing Quasar 1.1 438B, the first AI model using quantum-generated data.
Quasar 438B, the best European AI model, just got a quantum upgrade: Part of its healing set was generated by a hybrid quantum language model running on @IBM Quantum System Two in Donostia-San Sebastián, a 156-qubit IBM Heron processor. It is the first time quantum-generated data enters the CompactifAI pipeline. Quantum computing is not a label on this release, it is part of how the model was built.
Sharper, less verbose, tuned for agentic work and ready for European deployment: served by a European company incorporated under EU law and developed in line with EU regulatory requirements, including the transparency expectations of the EU AI Act.
Show more
gh seems to be pursuing creative load reduction strategies such as rendering a "More" button in the middle of the screen instead of loading the feed
could've also just deleted the page tbh, redirect to profile, ez fix
Show more
omp 18.2 is now out!
- Much speedier startup, full TUI booting faster than Codex, CC, and Pi,
- A new indicator, detecting and warning against gateways claiming mismatching frontier models,
- Ollama web search, refreshed /skill UX, tiered OpenRouter routing & much more
Show more
UA spoofing is back on baby, for only $0.00 you too can be OpenAI File Downloader, XaiImageApiFetch/1.0
Some sites like LinkedIn even remove their click-bait/paywall garbage if you're Claude-User
Show more
tfw your model is so bad at instruction following, you have to remind it to shut up every single turn (including tool-calls)
guys you do know you can just disable thinking, and instead give it a "deep_think" tool, and it will call it with internal CoT reasoning format right?
gl fixing that
