New Progress in GPT-4 Turbo:
the API. Function calls have become more powerful, allowing simultaneous requests for multiple APIs. For example, getting weather updates for three locations at once. Perform output control: response_format to { type: "json_object" }
Assistants API Function Enhancement:
. The API provides three powerful tools: code interpreter, retrieval, and function calls.
：Execute Python coding in a safe environment, generate charts, and process complex data. The Code Interpreter was introduced earlier. ：Leverage external data sources to extend the assistant's knowledge, such as domain-specific data and user-provided documents, simplifying the document processing workflow. ：Allow the assistant to invoke and integrate user-defined function responses.
Multimodal capability enhancement:
：As previously introduced, specific details can be found in relevant materials. ChatGPT multimodal trial ChatGPT's application capabilities in the visual domain - Advanced 1 Application capabilities of ChatGPT in the visual domain - Advanced 2 Application capabilities of ChatGPT in the visual domain - Advanced 3 : Currently, DALL·E 3 only has text-to-image functionality; outpainting, inpainting, and image-to-image are still in DALL·E 2. ：The text-to-speech function discussed earlier.
Personalization and Custom Models:
The fine-tuning details of GPT-3 are now applicable to GPT-4, bringing more personalized options for model interaction. As previously introduced. The custom model program is well-suited for training datasets with tens of billions of tokens, offering the possibility to create customized GPT-4 models.
Significant price reduction:
Input costs reduced to one-third of the original price, and output costs reduced to half of the original price. Price comparison table. The API call limit per second has doubled, providing better cost efficiency for large-scale operations.
Technical Enhancement:
: Provides stronger speech recognition capabilities. : The VAE component of Stable Diffusion, which improves text processing, facial control, and line work, bringing higher quality and accuracy.