For artificial intelligence, 2024 will go down as the year of all records. Hundreds of models have been made available: Gemini from Google, LeChat from the French company Mistral, or even o1, the latest generation of ChatGPT capable of “reasoning”. Not to mention the promises of AI agents, which have excited the sector.
Above all, investments have never been so important in the sector: in the United States alone, venture capital companies have raised no less than 56 billion dollars in 2024. A daunting sum, which has enabled many start-ups to explode their valuation – first and foremost OpenAI, the company behind ChatGPT. The company, led by Sam Altman, completed a $6.6 billion fundraising round in October. To see its valuation rise to $157 billion. A stratospheric level.
Compared to its competitors, xAI, created in March 2023 by Elon Musk, still looks like Tom Thumb. Initially launched with a team of 12 people, mainly engineers poached from OpenAI, Google and others, xAI had around 100 employees as of June 2024. By comparison, OpenAI would have 4,500 employees, as would Deepmind, the artificial intelligence division from Google. A very long list of job offers on the xAI website confirms that the company still needs many new recruits.
A slightly conspiratorial “fun” mode
However, xAI’s roadmap, drawn up by its media founder, does not lack ambition. xAI’s mission is to “understand the true nature of the Universe”, while positioning itself as an “anti-OpenAI”. Understand: without establishing the barriers put in place by Sam Altman’s laboratory to compensate for the biases of artificial intelligence. ChatGPT was trained not to discriminate and to provide cautious answers to certain controversial questions – an unbearable maneuver for Elon Musk, who criticized the chatbot as “woke”.
In response, xAI therefore deliberately developed “completely free” artificial intelligence, according to Elon Musk’s wishes. This chatbot called “Grok” offers two operating modes: one “classic”, and the other “fun”, incorporating jokes. At the risk of shocking sensitive souls?
Despite the desire to make Grok a neutral tool, in fact, the AI ”still has a very liberal editorial line”, explains Laurent Cordonier, director of research at the Descartes Foundation and specialized in issues of disinformation. This influence means that the chatbot can sometimes make dangerous speeches, especially when it is in “fun” mode: a Vice investigation previously revealed that Grok was telling users that certain conspiracy theories could be true.
A race for the number of users
It must be said that it trains on data coming from X (formerly Twitter), where a lot of fake news circulates. Based on this information, the AI produces erroneous summaries. “Grok cannot differentiate between what is true or false, including with humorous content,” explains Laurent Cordonier. Not to mention the fact that it is very easy to use Grok to generate fake news, given its lack of barriers.
Despite Elon Musk’s promises, Grok is not yet “particularly efficient”, indicates Mehdi Triki, head of institutional relations at HubAI. It falls short of its competitors on a number of points, notably on the understanding of the longest queries, the structuring of responses, and of course, on the precision of the facts. Nevertheless, “he is impressive in several areas.”
For example, he is highly appreciated for his speed, his ability to code or to solve mathematical problems. It scores 87.5% on the MMLU test, which compares the performance of large language models. It is therefore placed ahead of ChatGPT-4 which obtains “only” 86.4%.
Its performance has particularly improved with the release of Grok 2 in August. And for good reason: by being connected to X, it can count on the site’s 336 million regular users to have access to even more data and information. “With Grok, Elon Musk is not in a race for performance, but for adoption. His philosophy is not to surpass OpenAI on a technical level, but in number of users,” underlines Mehdi Triki.
A unique advantage in the world
For comparison, OpenAI has 300 million active users for ChatGPT. Therein lies the strength of the richest man in the world: he can very easily integrate Grok into his entire galaxy of companies. According to Hanan Ouazan, generative artificial intelligence expert for the consultancy Artefact, “this strategy will also allow Grok to obtain impressive visual understanding capabilities, by linking with Tesla”. The manufacturer has been working on the ability of vehicles to perceive their environment for years. He enjoys a huge lead in this area. If Grok accessed the driving data collected by Tesla cars for years, it would have a unique advantage in the world. “The only competitor that has a foot in driving is the Gemini model, from Google, which could possibly be integrated into certain vehicles connected with Android.”
“The second obvious synergy for xAI is with the Optimus robots,” continues Hanan Ouazan. These humanoids, presented during the last Tesla event, amazed those present, even if their dexterity remains questionable. “They are already impressive, but if tomorrow they are powered by Grok, it will be revolutionary.” This perspective undoubtedly explains the disproportionate valuation of xAI, while its chatbot has yet to prove itself. “Elon Musk’s vision transcends traditional company boundaries: xAI, soon TeslaX”, judge Hanan Ouazan.
The competition is lagging behind: OpenAI announced in March that it had entered into a partnership with Figure, a manufacturer of humanoids, but the company has not yet given a precise date for an arrival on the market. For his part, Elon Musk has already announced that a thousand Optimus robots would be deployed in his factories by 2025.
Of course, this statement should be taken with a grain of salt. Elon Musk is used to exaggerations and very optimistic launch dates: full autonomous driving, which the businessman has been promising since 2016, has still not arrived. “We must always be wary of publicity effects,” says Mehdi Triki. “In the tech sector, they always bring in a lot of investment, and entrepreneurs play a lot of it.”
Above all, advances in artificial intelligence, which have been very rapid in recent years, could be more modest in the future. Language models, which need gigantic amounts of data to train, have already digested most of the Internet. They could soon run out of fuel. Especially since computing power could also stall, due to lack of electricity or sufficiently numerous chips on the market.
Historically, the businessman has succeeded in carrying out projects initially considered crazy. But this madness is not always contagious. Hired with great fanfare from OpenAI, certain xAI employees took the opposite route after a few months, unconvinced by the project. The war between the two AI giants has only just begun.
.