Anthropic, an artificial intelligence startup positioning itself as the builder of a safer kind of chatbot, has released a new version of its AI bot, named Claude.
(Bloomberg) — Anthropic, an artificial intelligence startup positioning itself as the builder of a safer kind of chatbot, has released a new version of its AI bot, named Claude.
Anthropic said that Claude 2 is available to anyone in the US or UK online at claude.ai, and businesses can access it via an application programming interface. The new release on Tuesday comes several months after Anthropic began offering an earlier version of Claude to businesses that wanted to add it to their products. Previously, the bot was tested by a handful of companies including Quora Inc., which built it into an app called Poe that lets users ask questions.
Like its predecessor, Claude 2 is built atop a large language model and can be used for written tasks like summarizing, searching, answering questions and coding. Both models can currently take in large chunks of text — a user can ask it to summarize a book, for instance — though Claude 2 can generate longer responses than its predecessor. Responses can reach up to about 3,000 words, according to data provided by the company. Claude 2 will also offer more accurate responses on some topics, such as coding and grade-school-level math, the company said.
Anthropic’s goal has been for Claude to be less susceptible than other chatbots to manipulation. The company aims to control the generation of harmful or offensive results in response to prompts — issues that have dogged rival chatbots such as OpenAI’s ChatGPT. The startup said that its tests indicated Claude 2 was better than the existing model at generating what Anthropic calls “harmless responses” to “harmful prompts.”
Anthropic president and co-founder Daniela Amodei said that the newer model is meant to “feel, sort of subjectively, a little bit better” than the older version. She added: “And we also know that it’s still not perfect. It still has limitations.”
Anthropic, based in San Francisco, was founded in 2021 by a handful of former OpenAI staffers, including Amodei and her brother, Dario Amodei. The startup and its leadership have urged caution about the rush to develop and release AI systems, and their potential to transform industries.
In May, the company said it raised $450 million in funding, bringing the company’s total raised thus far to more than $1 billion.
Anthropic said it plans to make Claude available to people in more countries in the months ahead.
More stories like this are available on bloomberg.com
©2023 Bloomberg L.P.