In a remarkable turn of events, OpenAI, Google, and the French AI startup Mistral have simultaneously unleashed their latest cutting-edge AI models, setting the stage for an upcoming surge in industry innovation during the summer.
This synchronised release activity coincides with the anticipation surrounding the imminent debut of the next iteration of GPT, the backbone of OpenAI’s popular Chat-GPT chatbot.
The sequence of launches commenced shortly after Nick Clegg’s announcement at an event in London, confirming Meta’s impending release of Llama 3. Google swiftly followed suit, unveiling Gemini Pro 1.5, its most advanced large language model, featuring a complimentary tier with a daily cap of 50 requests. OpenAI then joined the fray, introducing GPT-4 Turbo, alongside Gemini Pro 1.5, as multimodal systems capable of processing ,not only text, but also images, with Gemini Pro 1.5 extending its capabilities to audio and video inputs.
In France, Mistral, founded by former members of Meta’s AI team, introduced Mixtral 8x22B, departing from its American counterparts by offering a straightforward download link to a 281GB file, adhering to an open-source model for AI development.
This approach, akin to Meta’s, has faced criticism for its potential risks, as it lacks mechanisms for intervention in case of misuse or the rectification of discovered vulnerabilities or biases.
Why is this important?
Despite Meta’s plans to roll out Llama 3 gradually, culminating in the release of its most advanced frontier model in the summer, it faces formidable competition from OpenAI, slated to unveil GPT-5 within a similar timeframe, according to Brad Lightcap, the company’s chief operating officer.
However, some experts question the efficacy of the “large language model” approach shared by these frontier AI systems, suggesting that it may be reaching its limits. Yann LeCun, Meta’s chief AI scientist, emphasised the need for a shift towards “objective-driven” AI, which encompasses reasoning and planning beyond linguistic capabilities alone. “We have AI systems that can pass the bar exam, but they can’t clear up your dinner table and fill up the dishwasher. We have systems that manipulate language, and fool us into thinking that they are smart, but cannot understand the world.”