Question about hitting a 90% accuracy mark with a small AI model

Everyone says you need huge models for real work, but I got a custom image classifier to 90% on my own data with just 2,000 training photos. I built it over a weekend using a free tool called FastAI, expecting maybe 70% at best. The milestone surprised me because it proves a focused, simple model can sometimes beat the 'bigger is better' idea. Has anyone else found a small project that worked way better than the usual advice said it would?

3 comments

Log in to join the discussion

Log In

3 Comments

taylor_hall5mo ago

Honestly, the real win is that you proved your specific data matters more than raw model size.

5

seth_martinez215mo ago

Remember when everyone thought bigger phones were better? Lol feels like the same vibe here.

3

fisher.reese4mo ago

Why do we always forget this with people too?

3