An AI-powered chatbot can “read” an entire book in one minute

Claude now has a so-called “context window” of about 75,000 words, reports anthropic. This context window is, as it were, the chatbot’s short-term memory, an additional piece of ready-made knowledge from which the chatbot can draw in addition to the pre-trained knowledge of the AI.

Example: Claude For example, you can “read” an entire book in less than a minute and then talk about every aspect of the book in plain language. This is useful when analyzing all kinds of texts, but also during conversations. Anyone who has had a longer conversation with a chatbot will notice after a few questions that the beginning is already lost. The chatbot then forgets what was discussed at the beginning of the conversation, which can lead to a circular conversation.

Memorize the meaning of the words

For example, Microsoft Bing’s chatbot, which works with ChatGPT, is limited to 20 questions. After that, the chatbot can forget the previous information, and strange and sometimes hurtful answers can appear. With Claude, that limit is now several times higher.

By comparison: Claude can now remember about 75,000 words in his “ready memory”, while in ChatGPT this limit is already about 3,000 words. However, with artificial intelligence, this memory is not measured in words, but rather in symbols: after all, not every word is the same length, moreover, it is mainly about the underlying meaning.

Not words, but symbols

So the AI ​​measures words into symbols, and Claude can store up to 100,000 symbols in its ready-made memory. With ChatGPT, there are 8,000 tokens in the paid version, or 32,000 tokens in the new limited GPT-4 version.

The new version of Claude is currently only available to Anthropomorphic Business Partners. It’s not publicly known how much access to Claude costs.

Google is said to have recently invested around $400 million in Anthropic. Google recently came out with the AI-powered chatbot Bard.