Anthropic, Alphabet-backed AI, Releases A Big Language Model That Is Claimed To Be Better Than OpenAI

Anthropic, an artificial intelligence company backed by Alphabet Inc. (photo: twitter @AnthropicaI)

JAKARTA - Anthropic, an artificial intelligence company backed by Alphabet Inc, on Tuesday, March 14 released a large language model that competes directly with products from OpenAI supported by Microsoft Corp, creator of ChatGPT.

Big language models are algorithms taught to produce text by giving them training texts written by humans. In recent years, researchers have obtained more human-like results with those models by increasing the amount of data provided to them and the amount of computing power used to train them.

Claude, as the Anthropic model is known, was built to perform tasks similar to ChatGPT by responding with human-like text to requests, either in the form of editing legal contracts or writing computer codes.

However, Anthropic, founded by Dario's twin brothers and Daniela Amodei, both former OpenAI executives, focused on producing artificial intelligence systems less likely to produce harmful or inappropriate content, such as instructions for hacking computers or making weapons, than any other system.

Concerns about the safety of such artificial intelligence are increasingly being noticed recently after Microsoft announced that it will limit demand for the newly launched Bing search engine after a New York Times columnist discovered that chatbots display alternative personalities and result in unpleasant responses during long-standing conversations.

Security issues have become a complicated problem for tech companies because chatbots don't understand the meaning of the words they make.

To avoid producing harmful content, chatbot creators often program them to avoid some topics at all. However, it makes chatbots vulnerable to "promp engineering," where users talk in certain ways to avoid restrictions.

Anthropic has taken a different approach, giving Claude a set of principles at a time when the model was "trained" with a huge amount of text data. Instead of trying to avoid potentially dangerous topics, Claude was designed to explain his objections, based on principles.

"Nothing scary. That's one of the reasons why we like Anthropic," Richard Robinson, CEO of Robin AI, a London-based startup that uses artificial intelligence to analyze legal contracts given Anthropic initial access to Claude, said in an interview with Reuters.

Robinson said his company had tried to apply OpenAI technology to contracts, but found that Claude was better at understanding dense legal language and was less likely to produce strange responses.

"If there is anything to do, the challenge is to make it looser in the right use," Robinson said.