OpenAI has just made a number of announcements that developers working with its products ought to know about.
First, the company announced that the API and ChatGPT now have access to the latest GPT-4o snapshot: gpt-4o-2024-11-20. This snapshot features better creative writing ability, with more natural, engaging, and tailored responses. Its ability to work with uploaded files has also been improved.
Next, the company announced the ability to use the Evals tool directly in the OpenAI dashboard. Evals is a framework for evaluating LLMs or the systems built around them for different criteria.
“When developing with AI models, it’s essential to continuously test their outputs to ensure they are accurate and useful. Regularly running evaluations (often called evals) on your model’s outputs using test data helps you build and maintain high-quality and reliable AI applications,” OpenAI’s documentation reads.
And finally, OpenAI is introducing the ability to test audio capabilities in Chat Completions in the Playground. This allows developers to test out voices and iterate on the capabilities quickly.
The ability to use audio in Chat Completions was first introduced last month and enables the LLM to detect tone, inflection, and other nuances within the input, resulting in richer data than text on its own. Developers can use this audio data to generate audio summaries of text, perform sentiment analysis on a recording, or do asynchronous speech to speech interactions.