@Jason Excited for this, I’m not brave enough yet for video but I just started to share what has been working for me. Trying to be unfiltered, ungenerated, and focus on what doesn’t work so others can avoid it.
https://t.co/BRBqefsBV8
In AI if you move the goal posts, own either team, or the refs, these evals aren’t wins or losses, they’re marketing theatre. Effective eval should naturally fall out of any system doing real work and subject to independent verification.
I think it's a mixture of both in concert that produce the desired effect however I use a simple prompt rule to help determine IF I write a property-based test.
The test MUST prove the existence of a real bug prior to remediation and the bugs elimination upon test inclusion. Understanding this won't catch all bug permutations. However, I've found operating with this guidance prevents me from generating 10,000 happy path useless tests and helps constrain token spend on tests that haven't caused previous pain.
The ability of an LLM or agentic framework to refactor a code base into a new language, while impressive, does not equate to an effective software or product development lifecycle.
The years of toil writing effective tests are key. It would be an interesting experiment for teams to attempt re-writing Bun or NodeJS with all tests removed first.
The ability
to replicate something that has decades of lessons learned does not mean a mechanism
to autonomously create production grade innovative products.
Effective tests and evaluations are probably the biggest pain I continue to see and not a void the LLMs seem to fill well.
We need the @haveibeenpwned equivalent but instead as “haveibeencloned” and instead of leaked passwords it focused on leaked repo context/IP into frontier model providers. @troyhunt
Careful you’re going to end up nerd sniped and come out the other end as OSS maintainer shoveling slop. I wonder if the future state of OSS is open code, closed contributions. Like you can cut tickets, request features, report bugs, but all PRs flow through the maintainers agent and review gates.