Launches occur within 12 hours of each other, with more industry activity anticipated over the summer
OpenAI, Google, and the French AI startup Mistral have all unveiled new iterations of their cutting-edge AI models within 12 hours of each other, signaling an upcoming surge of activity in the industry this summer.
This unprecedented wave of releases coincides with the sector’s anticipation of the upcoming launch of the next major version of GPT, the foundation of OpenAI’s popular chatbot Chat-GPT.
The first of these releases occurred shortly after Nick Clegg’s appearance at an event in London, where he confirmed reports of Meta’s imminent publication of the third version of its AI model, Llama, expected to be released in a matter of weeks.
Seven hours after Clegg concluded his presentation, Google introduced Gemini Pro 1.5, its most advanced large language model, to the public. The model offers a free tier with a limit of 50 requests per day.
An hour later, OpenAI launched its own cutting-edge model, the final iteration of GPT-4 Turbo. Both GPT-4 Turbo and Gemini Pro 1.5 are “multimodal” systems, capable of processing more than just text. They can process image inputs, while Gemini can also handle audio and video.
In the early hours of the morning in France, Mistral, an AI startup founded by several of Clegg’s former colleagues from Meta’s AI team, unveiled its frontier model, Mixtral 8x22B. Unlike its American counterparts, Mixtral was released as a simple download link to a 281GB file. Similar to Meta’s approach, Mistral adopts an “open-source” philosophy, making its AI systems freely available for anyone to download and enhance.
This approach has drawn criticism for its potential dangers, as it prevents developers from intervening to halt the use of their systems for harmful purposes or from taking models offline to address discovered vulnerabilities or biases. However, proponents, including Meta, argue that this approach ultimately leads to better outcomes compared to systems “controlled by a small number of very large, well-funded companies in California.”
Meta’s Llama 3 is expected to be initially released in its less powerful versions, gradually leading up to the launch of the company’s most advanced frontier model this summer, according to a report from The Information. However, it may face strong competition: OpenAI is reportedly planning a similar timeframe for its next GPT model, GPT-5, with the company’s Chief Operating Officer, Brad Lightcap, indicating to the Financial Times that it would be arriving “soon.”
However, experts have raised questions about whether the “large language model” approach, which is common among all advanced AI systems, might be reaching its limits. “We hear a lot of people saying, ‘Oh my God, we’re going to achieve artificial general intelligence within the next year,'” responded Meta’s chief AI scientist, Yann LeCun, to a claim made by xAI founder Elon Musk. “It’s just not happening. We have AI systems that can pass the bar exam, but they can’t clear up your dinner table and load the dishwasher. We have systems that manipulate language and trick us into thinking they are intelligent, but they cannot comprehend the world.”
LeCun proposed that researchers should focus on developing what he termed “objective-driven” AI, which possesses the capability to reason and plan in the real world, rather than solely focusing on textual data.
He expressed that this approach could lead to AI systems with genuinely superhuman capabilities. While he acknowledged that this concept is more of a vision at this stage, LeCun remarked that progress in this area is advancing exponentially, giving him confidence that this goal will be achieved.