Anthropic has just announced Claude 2.1, the new version of its chatbot which rivals ChatGPT. The AI can now process up to 500 pages of text and its error rate is halved.
OpenAI, which created the famous chatbot ChatGPT, has just spent a week in turmoil after the board of directors fired, then finally reinstated, CEO Sam Altman. Anthropic, one of the main rivals, couldn’t have asked for a better time to reveal an update to its chatbot Claude, which goes to version 2.1.
According to the announcement, Claude 2.1 has a popup of 200,000 tokens. This is more than 50% compared to the new GPT-4 Turbo announced just two weeks ago. The popup window is all the text that the AI can take into account before generating new text. 200,000 tokens equal 150,000 words or 500 pages of text.
The number of hallucinations is halved
Anthropic also announced an improvement in the reliability of the chatbot, with half as many “hallucinations,” or erroneous statements. When the AI does not know information, it will now tend not to respond rather than giving a false answer. This new version of the chatbot would also have better text comprehension, with a 30% reduction in incorrect responses to submitted text, and cuts the number of errors by three to four when it indicates that a document supports a particular claim.
The chatbot is also getting a new feature in beta that allows developers to connect it to their own products and APIs (application programming interfaces). Finally, Claude 2.1 includes the possibility of using system queries that add context and allow the chatbot to adopt roles or personalities. However, as with OpenAI, you will need to subscribe to have access to the latest version of the chatbot. Count on $20 per month.