Vent: the demo guy at the Austin AI meetup called my model 'too honest' and I can't shake it
So I showed up to this small AI builders meetup in Austin about 5 weeks ago with a little text classifier I built over a weekend, nothing fancy, maybe 500 training rows. This guy who runs a startup demo station tried it out and said flat out, 'this thing is too honest, nobody will pay for it.' He meant my model was giving wrong answers with total confidence instead of hedging, and he said investors want output that sounds sure. I've been chewing on that for a while because I built it to flag uncertainty on purpose. Now I'm second guessing the whole design, like should I add a confidence score or just let it guess? He also said his team spent 3 months on prompt wording alone, which seems wild to me. Has anyone here dealt with feedback like that, and did you actually change your model or just ignore it?