What difficulty? That fight is tough.

Real talk: I cheesed the Grymforge fight by lowering the platform and just had my barbarian yeet carafes of water at the construct from the upper staircase. Gravity's a bitch, Grymforge.

Ah. Does implying there's a 100% mean that there's a measurable 50%, 85%, and so on?

What I'm driving at: is there a measurable, repeatable way to show things are improving? If so, can that metric be reported on?

(Thanks for engaging with this btw, it's interesting work)

Right?

So is the conclusion basically binary? The llm output is either good enough or it isn't?

Curious what, if anything, 'prompt coverage' looks like for something like this.

1. Is there a metric for how effective these tests are? Does such a question even make sense?
2. If so, maybe someday @codecov.bsky.social can support it 😁?

If I have to read it with my own eyes or write it with my own hands: yaml

For everything else: xml

Zuckerberg was a good listen.

I'm not even a wrestling fan and his Vince McMahon episodes were really interesting.

Every Christmas he covers a "non bastard", of these his Aaron Swartz episodes were superb.

Nice! I was just thinking about picking up such a project over the holidays. This thing open source yet, can we take a look?