A quick chat with a coder friend changed how I see AI safety

Last Tuesday I was on a video call with an old friend who works on large language models in Austin. He told me that the biggest worry isn't the model going rogue, it's small hidden biases that show up in real world tools like hiring and loan approvals. He walked me through a test where the AI scored job applicants differently based on subtle wording changes in their resumes. That made me realize how much careful checking is needed before these systems get wide use. Has anyone else talked to someone in the field and come away with a new concern?

1 comments

Log in to join the discussion

Log In

1 Comment

rowan_butler9322d ago

Your friend in Austin described a resume test where wording changes flipped the scores. So what? A hiring manager reading the same two resumes would also get swayed by wording. That is not new, and it is not some hidden AI problem. Humans have bias too, and we have had hiring laws and HR reviews for decades to catch the worst of it. Software that scales up a weak bias is still just a tool, and we already know tools need testing before they get shipped. The real question is whether the bias is worse than what people already do, and I doubt anyone has shown that. Sounds like your friend found a bug, not the end of the world.

4