OpenAI quietly added new GPT-4o model to ChatGPT



On August 13, 2024, OpenAI, the developer of the AI chat service ChatGPT, announced that 'a new GPT-4o model has appeared in ChatGPT since last week.' Since then, some information about the 'new GPT-4o model' has been revealed.




ChatGPT updated to new model based on user feedback | VentureBeat
https://venturebeat.com/ai/openai-updates-chatgpt-to-new-model-based-on-user-feedback/

According to the release notes provided by OpenAI, the 'new GPT-4o model' is an upgraded version of the existing GPT-4o model, but it is unclear how it has changed specifically. OpenAI said, 'We would like to tell you exactly how the model's response is different, but figuring out how to benchmark it at a fine-grained level and communicate improvements is itself an area of ongoing research.'

'Through experimental results and qualitative feedback, we found that ChatGPT users tended to prefer different features, so we introduced the update to GPT-4o,' OpenAI said.

One user thought that the new GPT-4o model uses multi-step inference and explained how it does so, but OpenAI denied this, explaining that it did not introduce a new inference process with the update and that even if it did explain its inference, it may simply have been following user prompts.

Another user reported that image generation through ChatGPT by GPT-4o has also been enabled. When it was working with the previous GPT-4 model, images were generated using OpenAI's DALL-E 3 diffusion-based image generation model, so if it was actually using GPT-4o, it would be a big step forward.




According to OpenAI, the API version and ChatGPT version may differ. At the time of writing, the latest version is provided as an API under the name 'chatgpt-4o-latest', and although 'gpt-4o-2024-08-06' is recommended for most API usage, some developers are provided with access to the latest version.

Related Posts:

in Software, Posted by log1p_kr