(San Francisco) Anthropic, a competitor to OpenAI (ChatGPT), on Monday presented Claude 3, a new series of generative artificial intelligence (AI) models presented by the start-up as cutting-edge technology in terms of reliability and speed of responses.
“Claude 3 Opus is our smartest model, with market-leading performance for highly complex tasks,” Anthropic said in a statement Monday, where it unveiled three power-graded models: Haiku, Sonnet and Opus.
“Opus shows us the extreme limits of what is possible with generative AI,” says the company, referring to “human-like understanding.”
The success of ChatGPT since the end of 2022 has launched the vogue for generative AI, which makes it possible to produce texts, images, sounds, lines of code, etc., upon a simple request in everyday language.
Founded by former OpenAI employees, Anthropic is trying to distinguish itself from its competitors by putting in place stricter safeguards for this technology widely considered a new industrial revolution in Silicon Valley.
This approach had made Claude less impressive than ChatGPT. But the start-up believes that he has now overcome the obstacle without sacrificing his values, and even surpassed his rivals with his performance in various tests (equivalents of school and university exams, in general culture and mathematics, in particular).
“Previous Claude models often offered unnecessary refusals that suggested a lack of understanding of the context. We have made significant progress in this area,” says Anthropic.
“Claude 3 models demonstrate a more nuanced understanding of requests, they identify real harm and refuse to respond to innocuous messages much less often.”
There start-up based in San Francisco has raised at least $7 billion since 2021, the majority in 2023, thanks to investors such as Google, Salesforce and especially Amazon.
Sonnet is already available on Claude, its chatbot (conversational robot), while Opus is reserved for users subscribed to the professional service.
Anthropic for the first time allows users to provide images and documents for their queries to the model, but Claude does not yet produce images, unlike ChatGPT.
Despite its less reckless approach, the start-up does not escape the legal cases that accompany the development of generative AI.
Last year, Universal and other music publishers sued Anthropic for using copyrighted song lyrics to train its models.
And at the end of January, the American competition authority (FTC) launched an investigation into the massive investments of Microsoft, Google and Amazon in the main start-up of generative AI, OpenAI and Anthropic.