I was feeding my AI wrong for 6 months and a guy at a Seattle meetup made me see it

Charlie Roberts · 2026-10-02T00:20:03.460899+00:00

I kept wondering why my image classifier kept guessing cat when I fed it pictures of dogs, thinking the model was just bad. A guy named Dave at a Seattle AI meetup showed me I was normalizing pixels to a 0 to 255 range when the training script wanted 0 to 1, which basically threw off every label by a mile. The fix took about 3 minutes and my accuracy jumped from 58 percent to 91 percent on the same test set. Has anyone else found a dumb preprocessing mistake that was silently wrecking a model for months?

Topic Shares

Loading…

I was feeding my AI wrong for 6 months and a guy at a Seattle meetup made me see it | AI Innovations - Topic Shares