Today, we’re announcing that we will be deprecating the Answers, Classifications, and Search endpoints.
Since releasing these endpoints, we’ve developed new methods that achieve better results for these tasks. As a result, we’ll be removing the Answers, Classifications, and Search endpoints from our documentation and removing access to these endpoints on December 3, 2022.
We encourage anyone using these endpoints to switch over to newer techniques which produce better results, and have developed transition guides to help with this process.
If you are currently using these endpoints, this change will not immediately impact you. Prior to December 3, you will still be able to make requests and access the endpoint documentation by directly navigating to these pages:
Please note we will no longer be making any upgrades or fixes to these endpoints, and new accounts created after today will not have access to these endpoints.
If you have any questions or would like help updating your integration, please contact firstname.lastname@example.org.
The OpenAI Team
The transition docs require an account and email@example.com bounces :^)
The deprecation email inbox has been preemptively deprecated
Oh whoops! We’ll get right on fixing that! Thanks for letting us know.
firstname.lastname@example.org should be working now! Sorry for the confusion.
Hi, what is Intercom and why do I have to create an account on another platform in order to learn about the transition? Shouldn’t this information be integrated into OpenAI’s own documentation?
We’ve updated the links in the post so you shouldn’t need an account! Thanks for letting us know.
Thank you Chris, much appreciated.
Hi, the transition documentation recommends using embeddings-based search and then links to the existing embeddings documentation. Can you confirm that the following will remain true:
“For embedding-based search we will adopt the search model family, as it is specialized for document retrieval. In the example, documents are embedded by the
text-search-babbage-doc-001 model, and queries are processed by the
Yep that should remain true. Is there a particular concern you have around the embedding-based models?
No I think embeddings are amazing. But becuase the models have “search” in their name, I was worried whether they’d be affected by the deprecation. I switched from search to embeddings-based search a while ago. It’s amazing. I think we spoke over the phone a bit about it. I’m finding Q&A with embeddings cool too, although not very easy to get the quality of results I need for legal applications, but I continue to try. Overall, embeddings are the cool kid in school. Thanks again Chris.
I’m looking for the recommended replacement for the search endpoint. I have a list of documents and I want to compare input text to them to check for semantic similarity. What is the recommended way to do this now?
Search Transition Guide | OpenAI Help Center should have most of what you’re looking for. We generally suggest moving to an embeddings based approach but there is code in the openai-cookbook: openai-cookbook/search_functionality_example.py at main · openai/openai-cookbook · GitHub that will effectively replicate the search functionality.
I guess I’ll have to learn more about Python to understand this. Language agnostic documentation of how to use the API endpoints for a simple search would be useful. As it stands, it is really not clear to me from the Python code you linked what I need to do. My first step will be to figure out how to set up an adequate environment to run Python code, since I am not a Python developer.
Hello dschnurr. What exactly are the better methods for achieving the Answers endpoint? What would you recommend we use as an alternative?
If you were to read 2 messages up, you would have your answer. The answer is also in the documentation.