Heard a dev at a Seattle meetup say 'we're building the guardrails after the car is already on the highway' about new AI models.

It was about how fast some teams push new features out the door. Has anyone else's company actually slowed down to test for bias or weird outputs first?

3 comments

3 Comments

paulw531mo ago

Yeah, my friend's team tried to slow down for a bias check last quarter. They found the model was giving really bad advice on medical stuff, like scary bad. It took them weeks to fix it, but they caught it before launch, unlike what @robertcarr is saying. They said the pressure to just ignore it and ship was huge.

blakestone1mo ago

Our CTO made us add bias checks to the sprint plan as a non-negotiable task. It adds maybe two days but stops those last-minute fires. How does your team handle the schedule pressure?

robertcarr2mo ago

Man, that quote hits home. Is anyone actually building the guardrails first? My team talks a big game about responsible AI, but the pressure to ship new features is insane. We usually end up scrambling to fix problems after a feature is already live, which feels backwards and risky.