Happy 3rd birthday to OpenAI's longest-running model, Dec 15

[_j]

The oldest model listed on the models endpoint just had a birthday.

Can you guess what model it is?

text-embedding-ada-002

• Replaced five separate models for text search, text similarity, and code search* (actually, about a dozen..)

Announced December 15 2022 (almost missed the date!)

https://openai.com/index/new-and-improved-embedding-model/

…outperforms our previous most capable model, Davinci, at most tasks, while being priced 99.8% lower.

With no deprecation yet to be seen, and the investment it takes to run embeddings on documents for your semantic search, there is a good chance it will have a fourth birthday also.

(gpt-2 of course runs forever, being open-source)

[vb]

Nice catch.

The relationship with ada is a bit special, not only because it has a real name, but also because back then we would literally dissect models, explore all their features, quirks, and specialties, and then happily nerd out over them.

These days, it feels like we have shifted toward running quick evals and moving on, which is clearly the more practical approach.

Happy birthday, text-embedding-ada-002!

[yaroslav_ya]

Still use just it for my daily coding tasks and not even thinking to switch. All the next one we’re soulless