 "OpenAI, artificial intelligence")
Illustration: BINAY SINHA
Leading AI startup Anthropic has released a new version of its Claude chatbot, claiming it outperforms other leading chatbots on a range of standard benchmark tests, including systems from Google and OpenAI.
Dario Amodei, chief executive and co-founder of Anthropic, said the new technology, called Claude 3 Opus, was particularly useful for analyzing scientific data or generating computer code.
Chatbots like ChatGPT can answer questions, write essays, generate small computer programs and much more. They can also generate false or misleading information, just like people do. When OpenAI released a new version of its technology called GPT-4 last spring, it was widely considered the most powerful chatbot technology used by both consumers and businesses. Google recently introduced a comparable technology, Gemini.
But major AI companies have been distracted by one controversy after another. They say the computer chips needed to build AI are in short supply. Yet technology continues to improve at a remarkable pace. Anthropic claims its Claude 3 Opus technology outperforms both GPT-4 and Gemini in math problem solving, computer coding, general knowledge and other areas.
Claude 3 Opus is open to consumers who pay a subscription of $20 per month. A less powerful version, called Claude 3 Sonnet, is available for free. The company also deployed Haiku, another chatbot.
The company allows businesses to create their own chatbots and other services using Opus and Sonnet technologies.
Both versions of the technology can respond to images as well as text. These might analyze a flow chart, for example, or solve a math problem involving diagrams and graphs. But the technology cannot generate images.
Chatbots capable of imitating human conversation have become a growing priority, with rapid technological advancements fueling an investment frenzy. While chatbots themselves are not new, the technology that powers Claude and its competitors’ bots is a more powerful tool known as a large language model. But technology poses problems. For example, chatbots tend to say things that aren’t true, a problem sometimes called hallucinations. “These models are still just trained to predict the next word – it’s very, very difficult to achieve a zero percent hallucination rate,” said Anthropic President Daniela Amodei. In its latest launch, the company attempted to address this problem, a priority for Anthropic customers, Amodei said. The company said new versions of Claude software are twice as likely to provide correct answers to questions.
©2023 New York Times News Service
First publication: March 5, 2024 | 10:03 p.m. STI
