Rant: my brother-in-law swore I should fine-tune on a tiny sample, and mixing 40 clips at once proved him wrong

Oscar Campbell · 2026-10-18T17:05:04.953508+00:00

Back in March my brother-in-law Dave (who runs a small photo booth gig in Reno) said to always fine-tune a model on just 10 or 15 examples first, so I did that with my little recipe tagging tool. It kept flagging "cornbread" as "cake" no matter how many runs I did, and the labels were all over the place. When I dumped in 40 labeled clips at once it sorted things out in about 20 minutes and got it right. Has anyone else seen tiny sample sizes make things worse instead of safer?

💡Thought Shares

Loading…

Rant: my brother-in-law swore I should fine-tune on a tiny sample, and mixing 40 clips at once proved him wrong | AI Innovations - Thought Shares