# What is Claude, Anthropic's new AI assistant?

Date: 2023-03-15 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/what-is-claude-anthropic/

**Answer:** Claude is a conversational AI assistant from Anthropic, a company founded by former OpenAI researchers, opened to businesses on March 14, 2023 through a chat interface and an API. It comes in two versions, Claude and the faster, cheaper Claude Instant, and is designed around Anthropic's goal of an assistant that is helpful, honest and harmless.

Yesterday was a crowded day for artificial intelligence. OpenAI announced GPT-4, Google showed AI features coming to Gmail and Docs, and Anthropic, a smaller lab started by people who used to work at OpenAI, made its own assistant, Claude, available more widely. Claude has been tested for several months in a closed alpha with partners such as Notion, Quora and DuckDuckGo. Now any business can request access.

Like ChatGPT, Claude is a large language model: a program trained on huge amounts of text that predicts what words should come next, then tuned to hold a conversation. You type a request in plain language and it answers in plain language. Anthropic lists summarizing, search, creative and collaborative writing, answering questions and coding among the jobs it handles.

## Two versions: Claude and Claude Instant

Anthropic is launching two models side by side. **Claude** is the full, high-performance version. **Claude Instant** is lighter, cheaper and much quicker. The choice is a familiar trade-off for anyone building a product: the bigger model for careful work, the smaller one when you need many fast, low-cost answers.

| | Claude | Claude Instant |
|---|---|---|
| Aim | Best quality | Speed and low cost |
| Good fit | Long documents, careful writing, tricky questions | Quick replies, simple classification, high volume |
| Access | Chat interface and API | Chat interface and API |

Access runs through Anthropic's developer console, where companies sign up, and through partner products. Some of you may already have used Claude without knowing it. Quora offers it inside Poe, its chat app. Notion uses it in Notion AI. DuckDuckGo uses it in DuckAssist, a feature that writes short answers drawn from Wikipedia and related sources. Juni Learning runs a tutoring bot on it, and Robin AI uses it to read and suggest wording for contracts.

## Where "helpful, honest and harmless" comes from

Anthropic describes Claude as built on its research into training AI systems that are "helpful, honest, and harmless." That phrase is a research program, not just a slogan. In December 2022, Anthropic researchers published a paper on a method they call **Constitutional AI**. Instead of relying only on people to flag bad answers, they give the model a short written list of principles, its "constitution." The model then critiques and rewrites its own drafts against those principles, and another model judges which of two answers better follows them. Those judgments become the training signal.

The stated aim is an assistant that refuses harmful requests while explaining why, rather than giving evasive non-answers. Anthropic's launch post does not spell out exactly which training steps produced Claude, so treat the paper as background on the company's thinking, not as a full recipe.

What early customers report, according to Anthropic, is that Claude produces harmful content less often, is easier to talk to and more "steerable," meaning it follows directions about personality, tone and behavior well.

## Trying the steering yourself

Steerability is worth testing, because it changes how you write prompts. Here is a prompt that sets tone, audience and limits all at once:

```prompt
You are a friendly librarian helping a 12-year-old with a school project.
Explain what a volcano is in under 120 words.
Use one comparison to something in a kitchen.
End with one question the student could research next.
```

Give the same prompt to two assistants and compare how closely each sticks to the word limit, the audience and the format. Small differences in how a model follows instructions matter more in daily use than headline benchmark numbers.

> **Try it:** Take a paragraph you wrote recently and ask an assistant to rewrite it three times: once formal, once warm and casual, once for a ten-year-old. Then check which version kept every fact from your original.

## How is Claude different from ChatGPT?

On the surface, the two do much the same work. The Verge notes that Claude can summarize, answer questions, help with writing and generate code, which is the core of what people use ChatGPT for. The differences sit in the company's priorities and in a few practical details.

First, emphasis. Anthropic puts safety research at the center of its pitch, and its partners stress conversation quality. Poe users quoted by Anthropic describe Claude's answers as detailed and natural, and one says it "feels more conversational than ChatGPT." These are customer impressions chosen by the company, so it is fair to read them as early signals, not test results.

Second, control. Anthropic highlights that you can direct Claude's personality, tone and behavior. The Verge contrasts this with the three fixed modes (creative, balanced and precise) in Microsoft's Bing chat.

Third, connection to the web. Anthropic says Claude is self-contained and cannot browse the internet. Like ChatGPT in its standard form, it answers from what it learned in training, so it can be out of date or simply wrong. DuckDuckGo's approach of feeding it trusted sources is one way partners address that. Whatever assistant you use, the habit stays the same: check names, figures and quotes before you rely on them.

Finally, availability. ChatGPT has a free public website anyone can try. Claude, for now, reaches most people through business access and partner apps like Poe and Notion. Many expect the field of assistants to keep widening this year, which makes it useful to learn prompting skills that carry from one model to the next.

## Sources

- Introducing Claude: https://www.anthropic.com/news/introducing-claude
- Google-backed Anthropic launches Claude, an AI chatbot that's easier to talk to: https://www.theverge.com/2023/3/14/23640056/anthropic-ai-chatbot-claude-google-launch
- Constitutional AI: Harmlessness from AI Feedback: https://arxiv.org/abs/2212.08073