I enjoyed your presentation, thank you for going deep into it ! I have a massive app using ada-v2, and it's true, the similarities are often a little too close, even when the subjects are not related. But in your results, you say that the similarities in the Large Model are lower, ranging from 0.01 to 0.45.... Don't you find it strange ? A strong similarity should go to > 0.70. Is it a bug ? Can similarities with v3 be "loud and clear" ?
@codingcrashcourses8533
7 ай бұрын
The absolute values should not matter too much I think, but yes, I found that interested. I also found it interesting that in ada-002 totally unrelated values score das high as 0.7
@chakerayachi8468
5 ай бұрын
really thanks for the video but i have one question how to pass the dimensions size with azure open ai text embedding
@codingcrashcourses8533
5 ай бұрын
This works exactly the same, the models are the same: learn.microsoft.com/de-de/azure/ai-services/openai/concepts/models#embeddings
Пікірлер: 4