Image credit: Anthropic
Anthropic has announced the release of Claude 2, its latest AI model. The upgraded version brings substantial improvements in performance and expands the model's ability to generate longer responses. Designed to function as a personal assistant, Claude 2 now has enhanced capabilities in areas such as coding, mathematics, and reasoning.
With Claude 2, users can now input up to 100K tokens in each prompt, enabling the AI to work with lengthy technical documents or even entire books. The new model also has the ability to write longer documents of a few thousand tokens in a single go, improving its capability to create comprehensive documents such as memos, letters, or stories.
Claude 2 now has better performance, longer responses, and improved safety measures.
It has shown improved skills in coding and mathematics, scoring 71.2% on the Codex HumanEval, a Python coding test, and 88.0% on GSM8k, a large set of grade-school math problems.
The new model is twice as good at providing harmless responses compared to Claude 1.3, according to internal evaluations.
Claude 2 is powered by extensive red-teaming and various safety techniques to ensure its outputs are safer.
The company plans to deploy more capability improvements to Claude 2 in the upcoming months.
Anthropic is working with businesses that use the Claude API, such as Jasper and Sourcegraph, who have reported positive impacts on their services due to the implementation of Claude 2.
The Claude 2 model is available for use via API and on the new beta website, claude.ai, available to users in the US and UK. Source