.jpg?width=1200&auto=webp)
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. “Powerful, fast, or safe: pick three”, one billboard 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 discrete, and answers only when spoken to – like a perfect Victorian child who can also plan your holidays and fix your computer faster than you can say “IT”.
At least in theory. “At this stage, it is still experimental – at times cumbersome and error-prone,” said Claude’s parent company, Anthropic. Still, the technology’s potential is immense. Claude has been developed along the large language model Claude 3.5, designed to enhance self-correction capabilities and uphold ethical principles (more on this later). Claude 3.5 has what’s 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, writing fiction and fixing programming.
