Twitxy Saved

Can Bölük

Verified account This account is verified. Learn more

@can1357

BDFL @ omp.sh
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.
can.ac Joined June 2012
559 Following 22K Followers 2.1K Posts

Latest posts 5

Pinned post

Can Bölük Verified account @can1357
It's been a while since I wrote anything so I wrote an article on how to discover the entire x86-64 instruction set in seconds including any hidden instructions and learn their basic properties while on it. There were some pretty interesting results!

blog.can.ac/2021/03/22/spe…
blog.can.ac Speculating the entire x86-64 Instruction Set In Seconds with This One Weird Trick As cheesy as the title sounds, I promise it cannot beat the cheesiness of the technique I’ll be telling you about in this post. The morning I saw Mark …
Can Bölük Verified account @can1357
"oh no what did we regress from pi"
> 8 tasks failed because reward hacking counts as failure (we add web search, duh)
> 3 tasks failed because we keep bg tasks scoped to the session so that people don't have runaway scripts
jebaited once again 💀
Guanlan Dai Verified account @guanlan 1d
We’re spending up to 100B tokens across 26,040 agent runs to answer one question: when you switch models, should you switch harnesses too?

Our v1.0 FrontierHarness leaderboard ranked harnesses on a single model.
V1.1 expands it to 21 harnesses × 5 models × 62 tasks × 4 attempts, 72x the size of v1.0.

Results are coming in live. Link below🧵
View quoted post on X
12 replies 3 reposts 190 likes 15K views View on X
Can Bölük Verified account @can1357
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")