Anthropic, a major player in generative artificial intelligence, announced new models to fuel its Claude chatbot, the company said on Monday, as ChatGPT faces more rivals.
The company said three new AI models – called Claude 3 Opus, Sonnet, and Haiku – were its most high-performing tools yet and were industry-leading in terms of their ability to match human intelligence.
Founded in 2021, Anthropic was created by former employees of OpenAI, the company behind ChatGPT, and has been funded by Google and partnered with Amazon to develop new technology.
The company has made its hallmark on releasing AI models that seek to impose stricter guardrails than those behind ChatGPT and other chatbot rivals.
But this approach has faced pushback after last month’s release of Google’s Gemini model which was criticized for gaffes such as generating images of ethnically diverse World War II Nazi troops.
Some industry observers are also complaining that chatbots have become less impressive as companies introduce tighter controls in response to controversies involving the technology going off the rails or giving incorrect answers.
Acknowledging that safeguards could go too far, Anthropic said the new models would avoid making “unnecessary refusals” that were a problem for its earlier releases.
“Opus, Sonnet, and Haiku are significantly less likely to refuse to answer prompts that border on the system’s guardrails than previous generations of models,” it said.
Anthropic said its model Opus was the most powerful of the three and could outperform its peers on key benchmarks, including mathematics.
Claude is considered one of the major AI chatbot makers and is closely allied to Amazon and its AWS cloud division, which provides the company’s intensive computing needs.
It has also received investments from Google and other Silicon Valley heavyweights.
Unlike its rivals, Anthropic’s Claude chatbot does not generate images and only allows users to use images as requests for analysis.
The competing tools from OpenAI and Google generate images on request but executives from Anthropic believe that customers are not clamoring for the feature.
Like other AI giants, Anthropic is facing lawsuits from content makers who accuse the company of pilfering copyrighted material to build its models.
Universal and other music publishers last year sued Anthropic in a US court for using copyrighted lyrics to train its systems and generate answers to user queries.
