Re:Show HN: Make a free 3.8B model as reliable as one 7× big...
by babhishek21 on Monday June 01, 2026 @02:13PM
(#48360563)
Looking at the results, I have some thoughts:
1. I understand the need to have the ceiling model be at a big enough factor to make for a good headline. But is it really fair to compare between two different family of models (phi3:mini vs mixtral:8x7b)?
2. The corpora is really small. Are the results here statistically significant?
1. I understand the need to have the ceiling model be at a big enough factor to make for a good headline. But is it really fair to compare between two different family of models (phi3:mini vs mixtral:8x7b)?
2. The corpora is really small. Are the results here statistically significant?