A quiet, powerful update for API users of OpenAI
Daniel Schauer
At the intersection of computer magic??, AI exploration ??, and Data Science ??
OpenAI quietly released an updated model to its API users a few days ago, gpt-4o-2024-08-06. Frankly, it is kind of amazing!
With a single call, costing about a nickel, this new model does the work that previously I accomplished with a complex LangChain solution. My old solution had to make at least 40 LLM calls to accomplish the task reliably, and this took time and incurred cost-per-token.
The magic is the new model's json_schema Response Format mode. Writing the JSON schema only took me about 5 minutes.
The examples below are parsing the references listed in the famous paper Attention is All You Need.
If you have an OpenAI account, you can check it out here: https://platform.openai.com/playground/p/aL2WNkmkenqRZBv8pggLLJJV?model=undefined&mode=chat
Quite smart how they promote their new model iterations by making them stronger and cheaper. They can stop supporting older ones much faster.
https://analyticsindiamag.com/ai-origins-evolution/wait-did-openai-just-solve-jagged-intelligence/