Mythos 5.2 is definitely a step above 5.1. It outperforms Gemini Carbon 4 Cyber as well, which had a brief period as the top model in our benchmarks.
Cost is slightly higher even though token cost is lower, but it's finding more things so it's not really apples to apples.
@GeoffreyHuntley shit I knew I shouldn't have used hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2hunter2
@Steve_Yegge It seems to be hard to get right now but worth it if you do. You have to give the agent doing the work a bit of confidence.
"Feel free to push back on the review" usually works for me.
I know I have psychosis but I think it's plausible that a foundational open source project could be supplanted by an agent-developed alternative.
There's so much time / energy spent on existing process designed to solve problems that often don't matter anymore (if your goal is to build better software), so much social and coordination tax, resistance to change, etc.
(I don't know where this goes or what it means for open source.)
I think CVSS scoring is dead. CVEs should instead come with a prompt that helps your agent assess if your infrastructure is vulnerable.
We're wasting so much time on this as an industry. Lets give up. Give me the prompts.
Dear Github,
This new agent button is cool. But we would all much rather you fix the spinning merge thing that doesn't work. It's right above it.
Signed,
Everyone
So we’re rebuilding the unsecurable parts, at every layer. Two rules: bugs get found slower than you can ship, and every change is small enough for an agent.
Memory allocators, regex engines, XML libraries, and template engines so far. More coming.
We hae to patch the planet. We also have to rebuild it.
Vulnpocalypse or vulnerability human centipede?
We’re all sewn in somewhere: labs finding, maintainers fixing, vendors packaging, operators deploying. Nobody likes their spot. Last week showed everyone downstream what’s coming.
https://t.co/PbiKegyzYa
Even on latest, patching doesn’t win. Most of what we’re finding is the same few bug classes, and you can’t patch a class. You replace the design that keeps producing it.