In case any of yin see the "AI programmer Devyn!!!" hype, here's how I popped that hype balloon ...
The same marketing site that claims "Devyn can not just solve coding problems, but create entire applications on its own from prompts" lists its most impressive performance on SWE bench - the ability to solve code problems from a GitHub issue - at 13%.
And that's super impressive compared to other LLMs.
But if I couldn't solve 87% of documented bugs, I'd be out of a fucking job, y'all.
