Anthropic, an artificial intelligence startup, has unveiled Claude 2, the latest iteration of its conversational AI chatbot. Claude 2 is now open to the public in beta and aims to compete with chatbots like OpenAI’s wildly popular ChatGPT and Google’s newly announced Bard.
Claude 2 comes just months after the release of the original Claude chatbot in March 2022. According to Anthropic, Claude 2 has been improved in a number of areas, including better conversational abilities, more explanatory reasoning, more harmless responses, longer memory, and enhanced math, programming, and critical thinking skills.
A key upgrade allows Claude 2 to process up to 100,000 tokens or 75,000 words at once, significantly higher than ChatGPT’s limit of 4,096 tokens. This expanded context window allows Claude 2 to take more information into account when formulating responses.
In benchmark testing, Claude 2 achieved comparable results to OpenAI’s GPT-4, scoring 76.5% on a US bar exam multiple choice test, versus GPT-4’s 76.5% and ChatGPT’s 50%. It also reached 71.2% accuracy on a Python programming exam.
Anthropic claims Claude 2 has double the safety rating of its predecessor after passing “red team” tests designed to trigger unintended behaviour. The company says it relies on AI-based mechanisms rather than human oversight to train its models.
The launch of Claude 2 comes as competition heats up in the conversational AI space between startups like Anthropic and tech giants Google, Microsoft, and Meta. With $1.5 billion in funding and leadership from former OpenAI researchers, Anthropic aims to challenge OpenAI’s market lead.
Claude 2 is currently available in beta for free in the US and UK. Anthropic plans to add features iteratively in the coming months as it continues to refine the chatbot. Enterprise customers can access the Claude 2 API now for the same price as Claude 1.3.
