Anthropic unveils Claude 2: An enhanced Large Language Model for coding, text analysis, and composition writing

Anthropic has launched the latest version of their large language model (LLM), Claude 2, for beta testing. The updated model, designed for code crafting, text analysis, and composition writing, is available for free access to users in the US and the UK on a new website. Some well-known chatbots like Poe use Claude's LLM.

The enhancements in Claude 2 are based on user feedback and include improved conversational skills, clearer explanations, enhanced memory, and fewer harmful outputs. The model exhibits proficiency in coding, math, and reasoning abilities, as shown by its performance on the Bar exam multiple choice section (76.5%) and GRE reading and writing exams (above the 90th percentile). Claude 2 also supports longer inputs and outputs, enabling the analysis of large documents and generation of longer compositions.

Claude 2's coding capabilities have significantly improved, with its score on the Codex HumanEval Python programming test increasing from 56% to 71.2%. Its proficiency in grade-school math problems, as tested with GSM8k, has risen from 85.2% to 88% (as a related note, OpenAI recently released Code Interpreter Beta to all ChatGPT Plus users.). The model is also twice as effective in providing harmless responses compared to the previous version, Claude 1.3. However, despite its ability to process complex works, Anthropic advises against using Claude 2 as a factual reference or in situations involving physical or mental health. You can see all the changes that this new version brings here.