adi

making agents more useful, @ucberkeley alum

[adi @adbstack]

@alexandr_wang impressed with how quickly MSL has been able to reach frontier performance with Muse, but would urge you to look into this, even if some facts may be untrue/outdated.

imo the most significant problem to solve would be employees (managers as well as ICs) optimizing for promotions/perf rather than what’s actually beneficial for the business.

rooting for you!

[adi @adbstack]

@benhylak curious what differentiates you from other AI observability cos like judgement labs / Braintrust / Arize? and why have so many entered this space and pivoted?

[adi @adbstack]

@sytelus @code_star Generally agree that GitHub history can be really good signal. Curious if you think it’s reasonable to expect from candidates who are currently employed? I’d imagine that’s a large pool of talent that may not have time to make meaningful contributions to side projects

[adi @adbstack]

DMed. Tried it for one of my use-cases and believe it there's a lot of potential with enterprise. One suggestion though - might be worth adding more judge/rubric calibration support to the quickstart. I tried the creative writing flavor and got high-scoring outputs that weren't actually great (tbf this is a domain I've found is hard to get right)