OpenAI CEO Sam Altman has announced that GPT-5 will be a “significant leap forward” from GPT-4. However, he also warned that there’s still much work to do.
Why this matters. GPT-5 is the most anticipated launch in the field of generative AI for both end consumers using ChatGPT and enterprise users utilizing the API.
It promises to address many of GPT-4’s current limitations and unlock new application possibilities across various fields. Additionally, it’ll purportedly demonstrate increased intelligence and responsiveness.
Some context. As reported by Windows Central, during an interview at the Aspen Ideas Festival, Altman provided insights into the development of GPT-5. These included:
- Key improvements in reasoning and reliability.
- Multi-modal capability: text, images, and video.
- A release of GPT-5 projected for late 2024 or early 2025
In addition, there’s the very anticipated and increasingly possible introduction of autonomous AI agents.
A closer look. Altman has been critical of GPT-4, describing it as “mildly embarrassing” and admitting that it sometimes makes mistakes “even a six-year-old would never make.”
The OpenAI CEO's comments highlight the company's ambition to make significant improvements with GPT-5. It's not just aiming for an enhancement of the product, but also working to overcome the shortcomings that embarrass its CEO.
Reading between the lines. The development of GPT-5 faces major challenges:
- Algorithmic and complex data-related issues.
- Significantly larger model scale than in previous versions.
- The need for a radical improvement in reasoning capabilities.
Altman has likened the process to the development of the first iPhone, saying that, while it still had bugs, it was good enough to be useful. Interestingly, OpenAI has indicated that the current version of ChatGPT will seem “laughably bad” compared to what we’ll see in the next twelve months.
This clearly points to the significant advancements that GPT-5 could bring.
What’s next? Although there was initially talk of a summer 2024 release, Altman has since downplayed these expectations. There’s no official date yet, but at best, we can expect it by the end of the year.
It’s also important to manage our expectations in advance. In the months leading up to GPT-4, many expected much greater capabilities than what it actually delivered. Onlookers based their theories solely on the number of parameters in the model, using them as an indicator of GPT-4's abilities.
The improvement was notable, but not by a significant margin. Something similar might happen with GPT-5.
Image | Wikimedia Commons
View 0 comments