Hacker News Reader

I believe 4.5 is a very large and rich model. The price is high because it's costly to inference; however, the bigger reason is to ensure that others don't distill from it. Big models have a rich latent space, but it takes time to squeeze the juice out.

View on HN

That also means people won't use it. Way to shoot yourself in the foot.

The irony of a company that has distilled the word's information complaining about another company distilling their model...

My assumption: There will be use cases where cost of using this will be smaller than the gain from it. Data from this will make the next version better and cheaper.

The small number of use cases that do pay are providing gross margins as well as feedback that helps OpenAI in various ways. I don’t think it’s a stupid move at all.

Reply on news.ycombinator.com