OpenAI CEO Sam Altman has predicted that GPT-5 will be a “significant leap forward” from GPT-4, but has also warned that there is still a lot of work to be done.
Because it is importantGPT-5 is the most anticipated release in generative AI for both the end consumer, ChatGPT client, and for companies that use the API.
It promises to overcome many of the current limitations of GPT-4 and open up new application possibilities in many fields, as well as gaining in intelligence and responsiveness.
The context. In an interview during the Aspen Ideas FestivalAltman gave some clues about the development of GPT-5, as reported Windows Central.
- Significant improvements in reasoning and reliability.
- Multimodal capability: text, images and video.
- Launch planned for late 2024 or early 2025.
And a long-awaited and increasingly feasible possibility: the introduction of autonomous AI agents.
In detail. Altman has been critical of GPT-4, describing it as “slightly embarrassing” and admitting that he sometimes makes mistakes “that not even a six-year-old would make.”
This underlines OpenAI’s ambition to make significant improvements with GPT-5. It’s not just about improving the product, but about leaving behind the shortcomings that even embarrass its CEO.
Between lines. The development of GPT-5 faces major challenges:
- Complex algorithmic and data problems.
- Much larger model scale than previous versions.
- Need for a radical improvement in reasoning capabilities
Altman has compared the process to the development of the first iPhone: “It was still buggy, but it was good enough to be useful.” And here’s a fun fact: OpenAI has noted that the current ChatGPT “will be ridiculously bad” compared to what we’ll see in the next twelve months.
A clear allusion to the leap that GPT-5 will represent.
And now that? Although there was initially speculation about a summer 2024 release, Altman has cooled these expectations. There is no concrete official date but we will be leaving, at best, by the end of the year.
It is also worthwhile to moderate expectations in advance. In the months leading up to GPT-4, many of us assumed capabilities were much higher than what was actually implied by its arrival, relying on the number of parameters as an indicator of its abilities.
The jump was noticeable, but not that big. Something similar could happen with the fifth version.
Featured Image | Wikimedia Commons
At Xataka | Runway beats OpenAI to the punch and launches the Gen-3 Alpha video generator before Sora: it can now be used from the browser
Add Comment