Anthropic Launched Claude 2
Anthropic, a startup founded by former OpenAI executives, today announced the arrival of Claude 2, the second version of the AI chatbot. Available today in beta on the Web and API (Application Programming Interface), this bot promises better performance and a more limited ability to generate false or malicious content.
The Claude 2 bot, described by Anthropic as “helpful, harmless and honest,” can provide summaries of texts, write code, translate texts, and perform a host of other semantic tasks. Its functionality is described as similar to Google Bard or Microsoft Bing, but Anthropic claims it is built differently. His communication style is more conversational and human than his “brothers”, and he is also supposedly endowed with a sense of humor. Claude 2 is guided by a set of principles called “constitution” by its creators, which it uses to validate its responses without involving human moderators.
Claude 2 has greatly expanded its capabilities compared to its predecessor. In addition to being able to create longer responses, the chatbot is now slightly better at math, coding, and reasoning compared to the previous model. So, Claude 2 scored 76.5% in the bar exam, while the predecessor received only 73%. According to Anthropic, Claude 2 is much better at “providing harmless responses” that do not contain malicious content, although Anthropic does not rule out the possibility that the chatbot could be provoked.
Anthropic expanded Claude 2’s context window to about 75,000 words. The user will be able to upload dozens of pages or even a whole novel to the chatbot for analysis. Due to this size of the context window, Claude 2 can create a summary of a complex and very long research paper.
