with Claude 3, Anthropic claims to dethrone its rival OpenAI

2024-03-04 17:49:00

Generative artificial intelligence has a new champion. Almost a year after its release, the OpenAI GPT-4 model is no longer the most powerful artificial intelligence on the market. The crown is now claimed by Anthropic and its Claude 3 model, presented this Monday.

As proof of its new domination, the startup financed by billions from Amazon and Google highlights a comparative table with the results of the best models on the market out of ten benchmarks. In other words, it tested the AIs on their reasoning abilities, textual and visual understanding, writing computer code and even mathematical calculations. Claude 3 finished in first place in every test, only beating GPT-4 by one thread in three of them.

Although evaluating very large AI models as a whole is an inexact science, the results of which can always be debated, it seems that this is indeed the first time that the race for performance is no longer dominated by OpenAI. Google Gemini and Mistral Large had come close to GPT-4, had surpassed it on certain tasks, but without really overthrowing it. OpenAI certainly remains a year ahead in its development compared to Anthropic and its other pursuers, but it is under more pressure than ever to release the highly anticipated GPT-5 and regain its crown.

Perfect rival to OpenAI

Less known than OpenAI, Anthropic is its perfect rival. Its founders, brother and sister Dario and Daniela Amodei, are veterans of the house: they were part of a handful of researchers opposed to the direction taken by OpenAI following Microsoft’s first billion-dollar investment, in 2019. Anthropic, founded two years later, was therefore a direct response to OpenAI, with similar ambitions but with a stronger emphasis on security precautions in development.

Thanks to the pedigree of its leaders, the company raised $1.5 billion in investments between 2021 and the start of 2023. Then in September, everything accelerated. Anthropic received more than six billion dollars in funding promises from Amazon (four billion dollars) then Google (two billion dollars). This is, by far, the largest total of startups in the sector behind Microsoft’s investment in OpenAI, estimated at thirteen billion dollars.

Elon Musk takes OpenAI to court for prioritizing money over the public interest

And Anthropic does not intend to stop there, as reported by New York Times. In February, it again raised $750 million last month from the American fund Menlo Ventures, and it does not rule out seeking investments again at the end of the year. Its valuation has tripled in one year and today exceeds $15 billion, a significant amount but far from OpenAI’s $80 billion. It must be said that Anthropic is lagging behind in the commercialization of its AI compared to the latter. Last year, it had barely $8 million in monthly sales, according to the NOW.

Champion of Amazon and Google

Behind the OpenAI-Anthropic battle lies a remake of the cloud war between Amazon, Microsoft and Google, respectively numbers 1, 2 and 3 in the global market. In generative AI, Microsoft struck hard by offering exclusivity to OpenAI models on its platform and taking full advantage of the ChatGPT wave. Amazon also needed a showcase for their AI strategy, and they chose Anthropic. Claude-3 has already joined their AI model libraries, on behalf of Bedrock chez Amazon Web Services et Vertex AI at Google Cloud. Result: the most powerful language model on the market is available from them, and is no longer from Microsoft Azure, which only offers those from OpenAI and Mistral.

Be careful, however, to qualify the effects of the race for performance, because the impact of the battle of models on the real market remains to be demonstrated. Moreover, a large number of specialists agree that companies will increasingly turn to the use of small, more specialized and less expensive AI models, which will only be variations of very large models. like GPT-4 and Claude-3. But the performance of the best models will still serve as a headliner to attract customers.

Can Anthropic keep its ideal of artificial intelligence?

Created in opposition to OpenAI, Anthropic has placed the security of its artificial intelligences as a cardinal value. And for good reason: its leader Dario Amodei said last year in an interview that he estimated between 10% and 25% the risk that artificial intelligence would destroy humanity. To avoid the disaster scenario, the startup has implemented in September 2023 a scale of dangerousness of artificial intelligence, which reflects the extent of damage that a model can cause in the event of malicious use.

With Sora, OpenAI enters the race for AI-generated video

At the same time, it has implemented corresponding security measures at each level, in order to contain risks. Claude 3 is placed at level 2 of this dangerousness scale, out of 4, like its predecessor Claude 2. The next iteration of the model should reach level 3, synonymous with greatly reinforced safeguards, while the startup has not even still defined the protections necessary for a level 4 model. Could this priority on security handicap it in the future race for performance with OpenAI? Or will she end up making concessions?