Posted 26 Feb. 2023 at 15:07Updated Feb 26. 2023 at 15:27
Meta doesn’t want to be left behind in the AI race. Challenged by Microsoft, which has invested heavily in ChatGPT, the group led by Mark Zuckerberg has unveiled its own version of generative AI. But the latter is, for the moment, only open to scientists.
The objective of this tool called LLaMA is “to help researchers to progress in their work” on this subject, in particular because it does not require very large infrastructures to be studied, specifies Meta in a presentation text.
In November, the launch of the ChatGPT conversational robot from the OpenAI start-up shook up the world of AI, by allowing the general public to generate, in a few seconds, a text on a given topic or to provide an explanation on a subject. complex. Since then, all the Tech giants have reacted in the Western world and in China, fearing to miss a major shift.
The risk of precipitation
But, warn the experts, this tool also presents risks, whether it concerns factual errors, bias or also in terms of data protection. A test version of Microsoft’s Bing search engine, developed in partnership with OpenAI, quickly issued inconsistent responses, the computer program notably expressing threats or its desire to steal nuclear codes…
“Further research is needed to address the risks of bias, toxic comments and hallucinations,” says Meta, the parent company of Facebook and Instagram, which has already been singled out several times in its story due to the circulation of fake news on its platforms. In the eyes of Meta, it takes significant resources, especially in terms of computing power, to train and make the new language models work better.
Several versions available
At this stage, the lack of resources “limits the ability of researchers to understand how and why these large language models work, which hampers efforts to improve their robustness and mitigate known problems, such as bias, toxicity and the possibility of generating erroneous information”, specifies the Californian company.
This is why Meta has developed several versions of LLaMA, requiring more or less resources. As OpenAI and Microsoft limit access to technologies that power their AI, Mark Zuckerberg’s company decided to share how it built LLaMA so researchers can “more easily test new approaches to limit or eliminate ” The problems. A bet which, hopes Meta, will be a winner at the end of the race.