A typical couple of sentences consist of only 30 tokens, while a 1,500-word essay can use up to 2,048 tokens. ChatGPT, the language model, has an upper limit of processing prompts with a maximum of 8,000 tokens, while the extended version permits up to 32,000 tokens.

However, San Francisco-based AI lab Anthropic, started by OpenAI alums, has developed a new AI model, Claude, that can handle inputs with up to 100,000 tokens, which is equivalent to an entire novel. Claude can now accept full-length books like “Mrs. Dalloway,” “Treasure Island,” and “The Picture of Dorian Gray” and answer questions about their characters and plot twists with ease. Anthropic claims Claude can process inputs worth 100,000 tokens in less than a minute, which is five hours faster than a human would take to read the same text.

Reading such a vast text is just one aspect of the equation, as one needs to internalize the material to answer related questions. A human would require days or even months to remember every single word and spot modifications. Claude, on the other hand, can accomplish this feat in less than a minute.

Anthropic asserts that Claude can handle not only straightforward questions but also multi-step prompt chains, making it highly effective for businesses that need to search through large catalogs of documents for relevant information. Although Anthropic has made Claude’s superpowers available to business clients via APIs, there is no news yet of a watered-down version becoming available in the public domain, as with Bard, Bing Chat, or ChatGPT.

By Impact Lab