Join our daily and weekly newsletters to receive the latest updates and exclusive content on industry-leading AI coverage. Learn more
OpenAI continues to push the boundaries of generative AI. Yesterday, without much fanfare or warning, the company’s official statement ChatGPT Accountt on social media X posted an update that read: “A new GPT-4o model has appeared on ChatGPT since last week. We hope you all like it and check it out if you haven’t already! We think you’ll like it,” followed by a smiley face emoji.
While the ChatGPT app’s Platform X account did not provide further details immediately following the post, sources at OpenAI told VentureBeat that the novel model had been updated based on user feedback.
OpenAI later posted an update on its website blog release notes about the model, stating:
“we’ve made an update to GPT-4o that we’ve discovered, based on experimental results and qualitative feedback, that ChatGPT users prefer. This isn’t a new frontier class model. While we’d like to say exactly how the model responses differ, figuring out how to specifically test and communicate improvements in model behavior is itself an ongoing area of research (and one we’re working on!).“
This is not a novel line of reasoning, despite user speculation.
Brave users speculated that the novel GPT-4o model within ChatGPT was characterized by step-by-step or multi-step reasoning and more detailed explanations of processes provided to the user in natural language.
However, an OpenAI spokesperson told VentureBeat that the model update does not include any novel reasoning process and that the ChatGPT describing its reasoning can be triggered by a specific user prompt.
Ahead of the announcement, users noticed that ChatGPT’s underlying model, OpenAI’s GPT-4o, was behaving differently and better than in the recent past.
Other users report that the native GPT-4o image generation capabilities via ChatGPT also appear to be activated.
Why is this so essential, especially since ChatGPT can already generate images? Because ChatGPT, when powered by the previous GPT-4 model, relies on OpenAI’s DALL-E 3 diffusion-based image generation model to generate images based on user textual cues.
One of the main selling points and benefits of GPT-4o when it was announced in May was that it was trained to natively multimodal processing, converting not only text but also pixels into tokens, allowing it to generate higher quality images on its own than DALL-E 3, faster, more efficiently, with better understanding of text hints and more precise and realistic generation of illustrated text. In pictures.
Not everyone is joyful
Others have taken a more critical, even cynical, approach to the ChatGPT update, stating that OpenAI should have explained the changes in more detail in terms of model behavior and user experience.
Some even believe that the change is ultimately superficial or barely noticeable.
Different GPT-4o versions for ChatGPT and API?
When asked by VentureBeat about the update, an OpenAI spokesperson said, “We often make minor improvements to our models in ChatGPT and the application programming interface (API). The ChatGPT variant may differ from what is in the API because we are always optimizing for what is best for developers in the variant we deliver to the API.”
VentureBeat reported last week that OpenAI had updated its GPT-4o model. GPT-4o powers ChatGPT, as well as third-party developer apps via the OpenAI API.
On X, OpenAI Developers Official Account posted an update explaining:
“This model is now available in the API as `chatgpt-4o-latest`. We recommend `gpt-4o-2024-08-06` for most API use cases, but we are happy to provide developers access to test our latest improvements for chat use cases.”
For some developers like Aidan McLau, this wasn’t enough to get more information, so he asked the company to clarify the matter. Why supports two different GPT-4o models.
However, Michelle Pokrass, a member of OpenAI’s technical staff, quickly responded, stating, “chatgpt-4o-latest will track our 4o model in ChatGPT and is a chat-optimized model. Our model from last week (gpt-4o-2024-08-06) is optimized for API usage (e.g. calling functions, following instructions), so we generally recommend it!”
Fix: Tuesday, August 13, 4:11 PM EST: This article originally stated in the title and text of the article that OpenAI’s novel model demonstrated step-by-step reasoning based on a user’s post on X. However, after speaking with OpenAI, the company denied that this was the case. We have since updated the article to reflect this, and have also retained the user’s original tweet that inspired the conclusion.