The newest AI can arrange your holiday- but will it be a strictly woke one?

A lightning-quick artificial megabrain with an appetite for social justice? WILLIAM HOSIE has a chat with Claude Al

- WILLIAM HOSIE

If you've been in London these past few weeks you will no doubt have come across an advert for Claude AI, the chatbot coming for ChatGPT's crown. It's just like GPT, only sleeker, more capacious and with a distinctly ethical core that some are already labelling as woke (more on this later). You sign up, select your payment plan or (like most) opt for the free version, and ask away. "Powerful, fast, or safe: pick three," one billboard advertising Claude on City Road said. "The AI with people skills," another ran. I've always found the idea of people skills annoyingly amorphous, and the sort of people who say they have them to be either tone deaf or socially inept. Claude, fortunately, is neither: he is discreet and answers only when spoken to like a perfect Victorian child who can also plan your holidays and show you how to fix your computer faster than you can say "IT".

At least in theory. "At this stage, it is still experimental," says Claude's parent company, Anthropic. Still, the potential is immense.

Claude has been developed using the large language model Claude 3.5, which has been designed to enhance the chatbot's selfcorrection capabilities and to uphold ethical principles. In terms of processing power, Claude 3.5 has what's technically called a context window of 100,000 tokens, while GPT-4 (the large language model behind the latest version of ChatGPT) has a window of only 32,768. This means Claude 3.5 can handle much longer pieces of text, making it ideal for tasks like summarising lengthy documents and fixing programming issues.

It's a tough break for OpenAI, the parent company of ChatGPT, which is already facing an antitrust lawsuit from rival firm, Elon Musk's XAI. To rub salt in the wound, Anthropic's founders -siblings Daniela and Dario Amodei are ex-employees of Open AI.