What is Claude AI and How Does it Work?

If you’re happy with ChatGPT, you should probably explore one of its rivals – Claude AI. It’s a pretty new AI tool that was launched in March 2023.

Its latest release, Claude 3, was launched in March 2024 and is available in three variants – Haiku, Sonnet, and Opus. It was developed by Anthropic, a company backed by Google and Amazon.

There are many chatbots out there but only a few of them make news. Claude AI is definitely one of the most talked-about. Its latest iteration, Claude 3, has set new industry benchmarks as it covers a wide range of cognitive tasks.

In this article, I’ll discuss Claude, Anthropic, and how it all works.

What is Anthropic AI

Anthropic is an AI safety and research company based in San Francisco, California. Led by Dario Amodei, a former VP at OpenAI, Anthropic focuses on creating innovative AI models that prioritize safety and ethical considerations.

Anthropic first made news in May 2021 when it raised $124 million in funding from leading Silicon Valley investors. The team at Anthropic consists of researchers in the field of AI such as Daniela Amodei and Chris Olah.

Anthropic’s primary focus is on developing AI systems that are safe, ethical, and beneficial to humanity. The company aims to tackle the potential risks and negative consequences of AI technology. It prioritizes safety, transparency, and alignment with human values.

Just at the beginning of 2024, Anthropic raised a whopping $750 million, which made news all around the world. Let’s look at some details about Anthropic.

Anthropic AI Funding

Anthropic AI has raised a total of $7.3 billion in a year. It had raised $450 million by May last year and another $100 million by August. In September 2023, Amazon said it would invest $4 billion in Anthropic. This was followed by Google investing another $2 billion in it.

These figures will only rise by the end of 2024. With big names and big investments, Anthropic has gained a lot of attention in a short period.

What is Claude AI?

Claude is the AI chatbot model by Anthropic that was initially launched in early 2022. The first model of Claude was simply called Claude or Claude 1. It underwent iterations over the next few months and the current iteration is called Claude 3.

While Claude 1 wasn’t as advanced as the current model, it was a step forward in Anthropic’s mission to create safe and ethical AI systems.

Claude 3 comes in three different flavors.

Claude 3 Haiku

This is a lightweight model. While it may not have the full capabilities of the larger models, Haiku offers impressive speed and can perform conventional tasks.

Claude 3 Sonnet

Sonnet is the mid-sized model in the Claude AI 3 family. It’s a balanced model that combines speed and accuracy.

Claude 3 Opus

It is the largest and most advanced model in the Claude AI 3 lineup. Opus can generate text and solve complex real-world problems.

All three models have a 200k token context window. You can select the right model depending on your need for speed and accuracy.

Claude AI Training

Claude 3 is trained on data up until August 2023 and it cannot access the internet. But it isn’t exactly isolated from the online world.

While Claude cannot access the internet directly, it can do so through Anthropic’s proprietary API integrations. This way, Claude can get trained on filtered data.

With limited internet access, Claude gets filtered data for training so it does not self-learn from unfiltered sources. Anthropic wants to make sure that Claude remains a harmless, helpful, and honest AI assistant, which is the primary vision of Anthropic.

To demonstrate the accuracy of Claude 2, the AI model took the bar exam and answered 67% of the questions correctly. This is above the passing threshold of the exam. Compare that with GPT-3.5 which scored 49%. While GPT-4 scored the highest, Claude 2 came in second.

Claude 2.1 was released with a 200k token context window. This defines how much information Claude can retain from previous information exchange in the conversation. It helps the bot understand the context of the question and provide relevant answers.

Claude 3 models also have a 200k token context window. Plus, they can handle complex tasks that need to process a lot of information.

How Claude AI Works

Claude AI works on the Claude large language model (LLM). The Claude LLM runs on the transformer architecture.

It forms associations between words to understand the context of the prompt. Transformer architecture is also used in other popular chatbots such as ChatGPT and Google Gemini.

Claude was trained with filtered data so it stays away from harmful and hateful content. Anthropic focuses on harm reduction and it aims at developing a model that’s honest, helpful, and harmless. This approach makes Claude different than other AI tools.

Claude runs on a “constitution” that lets it refuse to answer certain questions. As the model trains with its constitution, it learns how to avoid messages with toxic or discriminatory behavior. The constitution of Claude promotes freedom, equality, respect, accuracy, and empathy.

The Claude 3 Family

The Claude 3 model can be used to drive live customer support, auto-complete features, data extraction, etc.

Haiku is the most compact model and can read research papers with 10K tokens. It can process graphs and charts in under 3 seconds.

Sonnet is almost 2x faster than the previous Claude models and displays high intelligence levels. Opus is the smartest and delivers speeds similar to Claude 2.

All Claude 3 models come with vision capabilities and can process several visual formats. This includes graphs, charts, photos, and diagrams.

This is helpful for users who have information encoded in different formats such as flowcharts, PDFs, tables, and presentation slides.

Claude 3 is more sophisticated than Claude 2 and Claude 2.1 and can understand real harm. It avoids false alarms and there are fewer refusals as it recognizes harmful prompts with more accuracy.

Is Claude AI Right For You?

Are you an avid ChatGPT user? If yes, then you should try Claude AI. Especially since it’s free to use. The Claude 3 Sonnet is available for free and lets you generate 20 messages in about 5 hours.

You can use the free version and decide if you want to stick with the free version or upgrade to the Pro plan.

Claude AI has garnered attention in a short period and definitely deserves a trial at least once. While it’s not better or worse than ChatGPT, it is certainly different in its own way. It has a welcoming interface and a large context window that makes it understand the context of the questions very easily.

Claude AI Pricing and Tokens

The current version is Claude 3 and has three pricing models. All three models have a 200k token context window.

• Haiku: Input at $0.25/MTok and output at $1.25/MTok
• Sonnet: Input at $3/MTok and output at $15/MTok
• Opus: Input at $15/MTok and output at $75/MTok

You can purchase Claude Pro for $20/month, which is pretty much the same price as ChatGPT Plus. With Claude Pro, you get 5 times the usage you’ll get with the free plan. Plus, you’ll get to access the Opus model with the Pro plan.

You can also access the Claude legacy models.

There are three legacy models available.

• Claude 2.1: Input at $8/Mtok and output at $24/Mtok
• Claude 2.0: Input at $8/Mtok and output at $24/Mtok
• Claude Instant: Input at $0.8/Mtok and output at $2.40/Mtok

Given all these pricing models, the cheapest will be Haiku which might not be very accurate but is fast and capable of generating text that’s more general in nature.

And since it has a context window of 200k tokens, it can remember the previous data exchange and generate results accordingly.

Difference between Claude AI and ChatGPT

While Claude AI and ChatGPT are both AI chatbots, there are some differences between them. They both have a free plan that lets you generate text and solve problems. Here are some areas of differences between them.

Moderation

Both Claude AI and ChatGPT discourage harmful content but Anthropic is more transparent about its ethical practices.

ChatGPT uses human reinforcement learning, which means the responses from ChatGPT are checked by humans. They read the responses and flag them if they have bias or harmful tones. This somewhat limits the ability of the model to filter unsafe content.

Claude uses self-moderation, which means as it trains further, it’s able to filter out unsafe content without a lot of human intervention. It trains on a constitution that regulates what’s toxic and what’s not.

Training Cutoff

The training data of Claude is up to August 2023 while the cutoff for ChatGPT 4 is April 2023. The cutoff for ChatGPT 3.5 is January 2022. This makes Claude the most updated AI model.

Features

If we compare the free models of Claude and ChatGPT, the former offers better functionality as it has visual capabilities and can read files and images. But when we compare the paid plans, ChatGPT is more powerful as it has image generation and voice capabilities.

Context Window

Claude has a larger context window than ChatGPT. With the ability to process up to 200k tokens, Claude 3 can process large documents.

Compare that with ChatGPT which has a context window of only 32k tokens. In simpler terms, while Claude can process about 150,000 words in one go, ChatGPT can process up to 22,400 words.

Integrations

Claude 3 is a rather powerful tool but it doesn’t have the same integration abilities as offered by ChatGPT. The latter has several integration options including Dall-E image generation, internet access, custom GPTs, and third-party GPTs.

ChatGPT also offers plugins that allow it to be used with other platforms. Clause is somewhat new so it doesn’t have the same plugin powers as ChatGPT.

Conclusion

Claude gives strong competition to ChatGPT. No wonder it has gained popularity in a short time. It has a large context window and the support of some large corporations.

One of the most prominent features of Claude is its ability to read and analyze long documents of up to 150 pages. This lets users summarize large files with ease.

While it’s an exceptional tool, Claude cannot access the open internet, which can lead to some inconsistencies in its results. But other than this limitation, Claude is a powerful tool that can soon replace ChatGPT as the number one AI chatbot.

Frequently Asked Questions (FAQs)