What is Claude AI, and how does it compare to ChatGPT?

These days, a lot of people are talking about ChatGPT, the AI chatbot that’s taken the world’s attention by storm. But there are alternatives on the market, and one of them is Claude AI, which is one of ChatGPT’s stronger competitors. There are key differences between the two that are worth considering, particularly in the area of token limits, costing, and AI ethics.

In this article, I’m going to demystify what Claude AI is and how it matches up to ChatGPT, including its strengths and limitations. 

What is Claude AI?

Claude, developed by Anthropic AI, is both an AI chatbot and the name for the underlying Large Language Models (LLMs) that power it. Claude is trained to have natural, text-based conversations, and it excels in tasks like summarization, editing, Q&A, decision-making, code-writing, and more.

Currently, Anthropic offers three “Claude” models: Claude 1, Claude 2, and Claude-Instant. While all are language-only models, each has subtle differences in capability. Claude is regularly trained on up-to-date information, and can read up to 75,000 words at a time. This means it can read a short book and answer questions about it!

Who is Anthropic AI?

Anthropic is an AI startup co-founded by ex-OpenAI members, notable for its focus on AI ethics. Among the founders are the Amodei siblings, Dario and Daniela, who were integral to OpenAI's GPT-3 project before parting ways due to AI safety concerns, sparking the inception of Anthropic in 2021.

If OpenAI sounds familiar, that’s because that’s the company behind ChatGPT. GPT-3 is the predecessor of both ChatGPT and Claude. Back in 2020, it was the world’s most advanced LLM — imagine ChatGPT, but less chatty and more like a glorified auto-complete tool. The Amodei siblings brought with them various researchers from OpenAI when they started Anthropic. Now valued at $20 billion, Anthropic is a public-benefits corporation that has secured substantial investments from tech titans like Google, and recently $4 billion from Amazon. 

Anthropic’s differentiator is AI safety. Unlike OpenAI's GPT-4, which has been trained on human preferences — a poorly defined metric — Anthropic pioneered a purposeful approach for their Claude models called “constitutional AI”. They have trained their AI to align with a constitutional AI document that outlines principles such as freedom, opposition to inhumane treatment, and privacy. They tout this approach as uniquely safe and responsible. 

How do I use Claude AI?

Getting started with Claude AI is simple. All you have to do is set up an account and/or login to claude.ai. You’ll see this input box greet you:

Type your question or message into the box. You can also optionally upload a pdf or text document that you want Claude to read in conjunction with your message for context. Then, start your chat! You can copy the response Claude gives you, retry your question for a slightly different answer, or provide feedback.

How does Claude AI line up against ChatGPT?

The various Claude offerings map well to OpenAI’s language model offerings: Claude-Instant is the cheaper and faster offering, similar to GPT-3.5, whereas Claude-2 is the cutting-edge but slower model, competitive with GPT-4. 

Claude does not have access to outside information outside of what is provided in their prompt, and Claude cannot interpret images or create images. 

Is Claude better than ChatGPT?

Yes and no, Claude wins over ChatGPT on some benchmarks while losing on others. Free Claude is better than free ChatGPT, but ChatGPT has superior paid subscription with expanded knowledge and capabilities compared to Claude.

The tests and evaluations we use to measure the intelligence of AI can never be comprehensive, so questions of which benchmarks to rely on remain. A simple and easy to understand leaderboard goes a long way, and LMSYS’s chatbot arena leaderboard is my favorite. It ranks models based on three evaluations:

1. Elo Rating

This is the same algorithm that ranks chess players against each other. It is based on blind side-by-side comparisons and human input to rank which response is best, and ultimately which model is best.

2. MT-Bench

This is a benchmark that evaluates models using GPT-4. The benchmark was shown to be 80% accurate to human evaluation and uses multi-turn question answer sessions as the input/output pairs that are evaluated.

3. MMLU (Massive Multitask Language Understanding)

This approach to LLM evaluation helps measure the knowledge captured during model training by evaluating question answering at various depths of understanding/ability and across a broad range of topics. See the dataset here.

Claude’s models all outstrip free tier ChatGPT, or GPT-3.5, both in Elo ratings and MMLU, but GPT-4 is clearly in the lead when it comes to the slower and more expensive tiers. Besides raw knowledge, Claude excels in reading, analyzing, and summarizing long documents with its 150 page limit, sufficient to fit small books. 

Claude Pro lacks many of the newer features ChatGPT+ offers, namely voice chat, image creation, data analysis, image understanding, and web browsing. The Claude Pro offering has a lot of ground to make up if it hopes to grow its market share at the same price point as ChatGPT+.

Can Claude AI access the internet?

No, Claude cannot access the internet. While the OpenAI-Bing partnership has been the impetus for developing internet access for OpenAI’s GPT-4, Claude has not yet partnered with a search engine in the same way. It remains to be seen whether Claude will get access to browse-functionality like ChatGPT, but for now it seems unlikely.

How much does Claude AI cost?

For the average user, Anthropic provides free access to their best Claude model via their chat interface at claude.ai, which is still in open beta testing as of October 2023. Just sign up for a free account!

They also offer Claude Pro at $20 per month, a subscription analogous to ChatGPT+. The free tier of Claude AI gives users access to Anthropic’s latest and most capable model, Claude 2. This contrasts with OpenAI which keeps GPT-4 behind their ChatGPT+ subscription. According to Anthropic’s website, Claude Pro offers:

• 5x more [Claude 2] usage than the free tier provides, with the ability to send many more messages

• Priority access to Claude.ai during high-traffic periods

• Early access to new features that help you get the most out of Claude

Anthropic sells access to their models to three different groups: standard users (e.g. users of their chat interface), developers who want to integrate Claude into their applications, and Businesses who require enterprise-level support. 

For developers eager to practically apply and expand their AI skills, Pluralsight’s AI Sandboxes offer a dynamic platform for experimentation and learning.

Developers looking to use Claude AI can get access to their models via the Anthropic API, or via Amazon Bedrock. Both Amazon and Anthropic support an on-demand approach, priced by the amount of text that you want Claude to process. 

The following table outlines the prices per thousand “tokens”, which are a measure of how much text is going into the model. A typical approximation is 750 words per 1000 tokens. Input and output tokens are priced differently due to differences in computation costs. Included is a price comparison with OpenAI’s GPT models.

For enterprise customers, AWS Bedrock also offers “Provisioned Throughput”, which reserves cloud compute capacity that is dedicated to your business. This option doesn’t come cheap, but once you have the compute capacity reserved, you can send requests to it as much as needed for the fixed cost of what you paid to reserve the cloud resources. 

A commitment of one month is the minimum time you can reserve throughput with Amazon Bedrock, and at the Claude-Instant price point, that will cost you $29,462.40. The following table outlines the price per hour of your reservation.

For the most up-to-date pricing, see Anthropic’s pricing page and Amazon Bedrock’s Pricing Page. 

Is Claude AI factually accurate?

Claude AI, like other language models, is designed to generate text based on the patterns it has seen during training. While Anthropic aims for factual accuracy, Claude is not perfect, and suffers from the same hallucination problems as GPT-3.5 and GPT-4. 

Claude really doesn't "know" what it knows; it just generates text based on statistical likelihoods. Saying "I don't know" is generally not its strong suit. So, it's always a good idea to fact-check important information. In my testing, I’ve found that Claude is more accurate than ChatGPT for summarization tasks.

How do I get access to Claude AI?

Claude AI or the Claude class models are available through various channels. The easiest way to access Claude is by making a free account at claude.ai to access Anthropic’s chatbot interface. If you are looking to use Claude for your day-to-day needs, the chat interface is great! Even basic experiments with prompt engineering can be done using the free tier which gets you access to Claude-2.0. 

If you are a developer looking for API access to Claude, you have two options:

• Anthropic’s own API is still in a closed beta phase and in order to access it, a request for early access needs to be submitted; however, there is no word on what requests they are granting or how long it will take to get access. 

• Amazon offers the full suite of Claude models through Amazon Bedrock. On-demand and provisioned throughput is generally available, so no waitlists or requests-for-access. 

Conclusion

If you’ve made it to this point, you’re now an expert on Anthropic’s Claude LLM. Claude stands out for its 100K token input limit, its uniquely transparent approach to AI safety with a “constitution”, and for the free access to the best Claude model developed yet, Claude-2. 

Anthropic has put up strong competition to OpenAI with Claude. Claude-2 is more capable than OpenAI’s free ChatGPT tier and is a strong choice for personal, developer, and even enterprise use. OpenAI’s GPT-4 still holds the crown, but with Claude on its heels, we may soon see OpenAI de-throned. 

Want to learn more about Generative AI?

Pluralsight offers a range of beginner, intermediate, and expert AI and ML courses, including dedicated courses on generative AI. Since you can sign up for a 10-day free trial with no commitments, it’s a great way to take some professionally authored courses with a set course structure.

If you’re interested in learning about using Generative AI in AWS, such as Amazon Bedrock, Amazon CodeWhisperer and Amazon SageMaker, you can also check out the A Cloud Guru course: “Foundations for Generative AI in AWS”.