Sam Altman says GPT-4 “kind of sucks” as OpenAI preps to release a new model that will reportedly make ChatGPT “really good, like materially better”
What you need to know
- A new report suggests that OpenAI is set to launch a new model for ChatGPT later this year during the summer.
- According to sources with close affiliations to OpenAI, the new model is “materially better” when compared to its predecessors.
- GPT-5 will reportedly ship with the capability to call AI agents developed by OpenAI to perform tasks autonomously, ultimately making work easier.
OpenAI is reportedly set to release the next version of its model for ChatGPT in the next couple of months. This news comes a few days after CEO Sam Altman touched base with Lex Friedman for an interview, where they discussed Sora AI, Elon Musk, and the OpenAI fiasco that led to his bizarre firing and subsequent rehiring.
Interestingly, Altman indicated that GPT-4 “kind of sucks” when asked about the LLM’s capabilities. He added:
“I think it is an amazing thing, but relative to where we need to get to and where I believe we will get to, at the time of like GPT-3, people were like ‘oh this is amazing, this is like marvel of technology’ and it is, it was. But now we have GPT-4 and look at GPT-3 and you’re like that’s unimaginably horrible.”
But as it now seems, the next version of the model that runs ChatGPT is currently under development and should ship in the middle of this year, potentially during the summer (via Business Insider).
According to sources with close affiliations with OpenAI, a handful of enterprise customers recently gained early access to the latest model (GPT-5) and got to tinker around with its new features and capabilities.
How does GPT-5 compare to GPT-4?
One of the CEOs who interacted with the new model says, “It’s really good, like materially better.” The CEO also indicated that OpenAI showcased the new model’s capabilities with use cases and data unique to his company. He added that the new model also features several “as-yet-unreleased” capabilities, including the ability to call AI agents developed by OpenAI to perform tasks autonomously, ultimately making work easier.
The sources further disclosed that OpenAI’s GPT-5 model is still under development and is in the training phase. Beyond this point, the ChatGPT-maker will run internal tests on the model. The model will then be released to a handful of users, including OpenAI staffers and several outsiders who’ll put the tool to the test to identify any loopholes and setbacks before it’s shipped to broad availability.
With this in mind, it’s possible that the timeline to release the model to the public can be pushed ahead depending on the test results. ChatGPT has been riddled with several issues since GPT-4 shipped, including the decline of its user base and accuracy.
Several users have also lodged reports indicating that the chatbot has seemingly gotten dumber. This is on top of the multiple lawsuits OpenAI is battling, from copyright infringement allegations to abandoning its founding mission and choosing to go the for-profit way.
It’ll be interesting to see how the new GPT-5 model performs with copyrighted material restrictions. Sam Altman already indicated that it’s virtually impossible to create tools like ChatGPT without using copyright material.
