looking for venus on the ocean floor
Joined January 2016
- Tweets679
- Following659
- Followers505
- Likes1.4K
after going over the EL updates for the glamsterdam hf, @Nethermind seems have adapted the best in the agentic era without compromising on correctness. other teams are moving at half the speed. @ErigonEth 's doing well too, but the codebase seems more buggy
the new anthropic CVP for defenders (defense access) is much more benevolent than the oai equivalent (daybreak blue) whilest supporting the SOTA (opus 5.5)
highly recommend for you to go get verified!
what's your daily driver model for light cyber tasks (triage, pocs, modeling, etc)?
i really dislike gpt6-sol, and still run gpt5.6-sol, anything better out there?
astra, fable, opus5.5 all get flagged pretty fast even with the cyber programs
i tried to demonstrate the "confirmation of an invalid transaction". the issue was about a vm incorrectly executing the prescribed semantics. i assume, my report got interpreted as a valid tx introducing invalid state change, and got closed.
earlier this year it was easy to find bounty-grade bugs by just chatting with your lmm buddy
with each model release there's a short abitrage window, but overall the model capabilty is commoditizing
you can either brute-force by 10x-ing your token usage to get only a slight advantage, or unfortunately, use the brain again
with a harness you can transcend the model. the models are great tacticians, but weak strategists.
a codebase is an exponential maze, and you will get lost making locally optimal decisions while relying on a lossy context
respect
Confession time:
@ControlZ_1337 didn't manually review a single line of code in this competition.
Everything was done through custom AI workflows that we built and orchestrated using our own platform: open·kritt.
And as the name suggests, we're open sourcing it. @Kritt_AI
We're planning to release it this week (hopefully), so if you don't want to miss it, go follow @Kritt_AI for updates.
Our hope is that open·kritt helps more researchers build powerful AI-assisted security workflows and ultimately makes the Web3 ecosystem more secure :)
can't wait for the siri magic in gpt 6.0
🚨OPENAI ALLEGEDLY STOLE APPLE’S ENTIRE PLAYBOOK
“Hundreds of billions of dollars, decades of work”
Hardware Engineering
- circuit designs, component architecture, power management
- unreleased products guarded by internal codenames
- AI/ML integration for hardware
- EMI engineering + testing methods
Manufacturing Secrets
- proprietary processes + custom machinery
- equipment Apple builds and installs IN supplier factories
- proprietary metal alloys + finishing techniques
- DFM (Design for Manufacturability) expertise
Component Technologies
- power chips, battery systems, displays, acoustics, touch
- identities of specialized sub-suppliers
- exact specs Apple demands from components
Testing Data
- failure analyses, lifecycle simulations
- “negative know-how” — everything Apple tried that DIDN’T work
Supply Chain
- supplier contracts, allocation strategies
- global logistics coordination
- systems-level integration
ALL GONE 💀
basic bug and a clever exploit. check your return values!
i partially blame this on the evm, as the codesize limit and high gas prices encouraged similar practices
unfortunately, gas golfing patterns might still be deeply entrenched in the training data, and agents will still use them even as evm evolves, and more defensive code could and should be written
On 1 June, Gnosis Pay experienced a security incident affecting card accounts. All affected balances were restored.
Post-mortem here: gnosis.io/blog/post-mortem-g…
conveniently put mythos and fable into the same column.. from quick testing, fable scores as high as 0% on cyber
Fable 5 is state-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, scientific research, and vision.
The longer and more complex the task, the larger Fable 5’s lead over our other models.
ALT Benchmark table titled Mythos 5 & Fable 5, comparing Claude Mythos 5 and Fable 5 against Claude Mythos Preview, Claude Opus 4.8, GPT 5.5, and Gemini 3.1 Pro.
