OpenAI, Google and French artificial intelligence startup Mistral have launched new versions of their cutting-edge AI models within 12 hours of each other, as the industry prepares for a burst of activity over the summer.
The unprecedented rush of releases comes as the industry prepares for the long-awaited release of the next major version of GPT, the system that underpins OpenAI’s successful Chat-GPT chatbot.
The first came just hours after Nick Clegg appeared on stage at an event in London, where he confirmed that the third version of Meta’s AI model, Llama, would be released within weeks.
Seven hours after Clegg left the stage, Google’s Gemini Pro 1.5, its competitor’s most advanced large language model, was released to the general public, with a free tier limited to 50 requests per day.
An hour later, OpenAI released its own frontier model, the final version of GPT-4 Turbo. GPT-4 Turbo and Gemini Pro 1.5 are “multimodal” systems, capable of accepting more than just text. Each can accept images, while the Gemini can also accept audio and video.
In the early hours of the morning in France, Mistral, an AI startup founded by several of Clegg’s former colleagues on the Meta AI team, launched its frontier model, Mixtral 8x22B. Unlike its two American competitors, Mixtral was launched via a simple download link to a 281GB file: the company, like Meta, takes an “open source” approach and publishes its AI systems free so anyone can download and develop them.
That approach has been criticized as potentially dangerous, as it leaves the developer unable to intervene and prevent their systems from being used for harmful purposes, or to take models offline if vulnerabilities or biases are discovered that need to be fixed. Others, including Meta, defend it because it ultimately leads to better results than systems “kept in the clammy hands of a small number of very, very large, wealthy California companies.”
Meta’s Llama 3 is expected to initially launch in its smaller, less powerful versions, leading up to the launch of the company’s most advanced frontier model this summer. However, it may face stiff competition: OpenAI is believed to be planning a similar timeline for its next GPT model, GPT-5, the company’s chief operating officer, Brad Lightcap, told the Financial Times. would come “soon”.
However, experts have wondered whether the “big language model” approach shared by all frontier AI systems could be reaching its limits. “We heard a lot of people say, ‘Oh my God, we’re going to get [artificial general intelligence] within the next year,” said Meta’s chief AI scientist Yann LeCun, responding to a claim from xAI founder Elon Musk. “It’s just not happening. We have artificial intelligence systems that can pass the bar exam, but they can’t clean the table or load the dishwasher. “We have systems that manipulate language and trick us into thinking they are intelligent but they can’t understand the world.”
after newsletter promotion
Instead, LeCun suggested, researchers needed to work on what he called “goal-driven” AI with the ability to reason and plan about the world, rather than working solely with words.
That approach could lead to artificial intelligence systems with truly superhuman capabilities, LeCun said. This “is more of a vision than anything,” he added, “but it’s making exponential progress, so I’m pretty confident we’ll get there.”