AI chatbot
Screenshots
Claude is an AI Chatbot and assistant similar to ChatGPT, developed by Anthropic. They’re currently testing this ChatGPT rival. Even though this browser chatbot isn’t available for regular users yet, there is enough information to understand the differences between it and other similar tools.
The principle behind Claude is to create an AI that’s less likely to be ‘primed’ with human-created labels which could lead to biased or discriminatory answers being returned. The idea is to have an AI that’s ‘regulated’ by other AI systems. This fundamentally changes how the AI behaves when faced with a question.
Meet “Constitutional AI”
Unlike other AIs trained with Reinforcement Learning, the approach used in Claude is different. Anthropic has dubbed it Constitutional AI. While a detailed explanation is impossible here, the AI has a set of underlying principles called its “constitution”.
Answers are based on these principles, which are the only part where humans intervened. This fundamentally changes how the AI behaves when faced with a question. It also provides a background for “ethical” behaviour since it can’t betray its principles.
Limited and still more reasonable
Claude is still in its AI infancy when compared to the amount of data and computational power required for more complex questions. For example, it can’t perform complex mathematical calculations and doesn’t have as much data as other AIs. Amusingly, it has no shame to recognise this and even apologises.
However, many AIs offer false information or extrapolate facts wrongly, whereas Claude doesn’t. For example, ChatGPT replies with confidence about who has the world record for walking across the English Channel. When asked the same question, Claude mentions it’s impossible to walk across it since it’s a body of water.
More human than ever
Claude is an incredible tool for information search, even with its basic functionalities being under development. Strangely enough, this AI feels more human than the rest, in no small way thanks to the way it seems to understand what it is, and what its shortcomings are.
If you thought ChatGPT was a polite AI, you will be blown away by Claude when it’s released for general use. However, it still needs improvement in some areas such as the amount of data it’s trained with, especially when it comes to maths and code development. That said, it’s a formidable tool and capable of better features verbally than others.
