AI: ChatGPT rival, Californian startup Anthropic's new tool can "think" and "reason"
In addition to providing immediate answers like its competitors, the new Claude 3.7 Sonnet model has the ability to "answer on more complex topics" according to Jared Kaplan, its co-founder.
San Francisco startup Anthropic , a rival to OpenAI (ChatGPT), launched a new generative artificial intelligence (AI) model on Monday that promises to save software developers a lot of time, as the race for AI assistants rages on. Claude 3.7 Sonnet is “much, much better” at coding programs, Anthropic co-founder Jared Kaplan told AFP. Above all, he added, it is a “hybrid” model: “It can give immediate answers to any question,” like the previous version 3.5 launched in September, “but also think for a long time, reason in order to answer on more complex subjects.” This makes the new Claude better able to follow instructions, write longer documents or perform more complex analyses, according to the executive.
The year 2023 was marked by the race for competing interfaces to ChatGPT. These AI assistants have since acquired the ability to understand and even produce images and sounds (multimodality), and they are now gaining in “reason” and autonomy. OpenAI was once again a pioneer with the release of o1 in September, the first model that “reasons” , that is, displays the stages of its “reflection” before arriving at an answer.
But Anthropic - founded in 2021 by former OpenAI employees - beat him to the punch on autonomy by releasing "Computer Use" in October, a feature that allows Claude to use computers like a human: he browses the internet, selects buttons on a site, enters text and uses different software. The creator of ChatGPT responded in January with Operator, its first AI agent with similar autonomy. The two Californian startups, like the tech giants (Google, Meta , etc.) are looking for ways to stand out in a market where their services look very similar to non-professional consumers.
“It’s true that simply better, faster, etc. models don’t necessarily have a significant impact,” acknowledged Mike Krieger, Anthropic’s chief product officer. “We’re really trying to make improvements that directly address our customers’ needs,” he said. In addition to version 3.7, the company introduced its first AI programming agent, “Claude Code.” Krieger said the tool allowed it to design a feature for a mobile app in 40 minutes, instead of hours. Amazon invested $8 billion in Anthropic, while Alphabet (Google) paid it $2 billion in 2024.
lefigaro
