Get Started with Claude AI that Ethically Rivals ChatGPT

Get Started with Claude AI that Ethically Rivals ChatGPT

12 mins read104 Views Comment
clickHere
Syed Aquib Ur
Syed Aquib Ur Rahman
Assistant Manager
Updated on Mar 5, 2024 14:11 IST

Whether Claude AI is better than ChatGPT or RoBERTa, that remains to be seen. Now that Google Gemini 1.5 has also surfaced and its pro version has increased the token window in leaps more than Claude AI, let’s explore this tool’s usefulness and discover how you can get started using the new version today.

Claude AI

Claude AI lets you chat with recently trained data up to December 2022 and early 2023. With Claude 3, the data is trained until August 2023. Now, it can summarise an entire book, complete coding, analyse images, and do mathematical tasks, and more conversationally. Also, now you can expect more 'expressive' responses as compared to earlier, considering you follow the right prompts explored below.  

Let’s get up to speed with this exceptional AI chatbot, if you haven’t already.

Key Features of Claude AI

This is a powerful chatbot built on Anthropic’s Claude large language model (LLM). 

  • Claude uses federated learning techniques, so user data stays on the device and is not centralised in the cloud. This protects privacy and reduces data security risks.
  • It reads PDFs and CSV files. It has a context window of 100,000 tokens, which roughly is 75,000 words (as long as a short novel like The Great Gatsby). With its new version, Claude 3, it has increased its token window to 200,000, which is around 500 pages of text.  
  • Claude does not experience a real understanding of human attributes like emotions, tone, and personality. It detects emotional words and language patterns to guess at user feelings. 
  • Claude AI can look at sentence structure and style to estimate the tone of conversations. Likewise, it can identify linguistic patterns over time to approximate personality traits.

Where Claude AI Shines

Claude’s main use cases cover the following areas but are not limited to these only. 

The past version of this conversational assistant was Claude 2. It was a significant improvement over the previous version – Claude 1.3. The improvements came across multiple areas of coding, maths, and safety. According to Anthropic, Claude 2 ‘scored 76.5% on the multiple choice section of the Bar exam, up from 73.0% with Claude 1.3.’
Claude 2 also recently scored

  • 71.2% on the Codex HumanEval on Python programming proficiency.
  • 88.0% on GSM8k, declaring its proficiency in mathematical skills. 

So let's find out how the new version fares. 

The new version, Claude 3, comes with advanced capabilities, particularly the Vision Analysis feature, which is something that exceptionally rivals Gemini and OpenAI. It can transcribe images, and even technical diagrams. It does not generate images, however. 

Here is a comparison table of the superior capabilities of Claude 3. 

Task Claude 3 Opus Claude 3 Sonnet Claude 3 Haiku GPT-4V Gemini 1.0 Ultra Gemini 1.0 Pro
Math & reasoning MMLU (mult) 59.40% 53.10% 50.20% 56.80% 59.40% 47.90%
Document visual Q&A 89.30% 89.50% 88.80% 88.40% 90.90% 88.10%
Math MathVista (test/init) 50.5% CoT 47.9% CoT 46.4% CoT 49.90% 53.00% 45.20%
Science diagrams AI2D test 88.10% 88.70% 86.70% 78.20% 79.50% 73.90%
Chart Q&A BotanaDataringV2 (b01) 80.8% 0-shot CoT 81.1% 0-shot CoT 81.7% 0-shot CoT 78.5% 4-shot CoT 80.80% 74.10%

 P.S.: The above comparison table is a vision analysis on Claude Sonnet, of the image below. 

Claude vision analysis comparison

Legal Solutions

The chatbot has a good understanding of legal language. Richard Robinson from a legal infrastructure business – Robin AI, mentions why they use Claude –  “We use Claude to evaluate particular parts of a contract, and to suggest new, alternative language that’s more friendly to our customers.”

Conversational Coaching

Be it, for career development or work-related advice or even for some conversations, you can find some good results. 

Have a look at the screenshot below.

Conversational coaching

By the end of the answer, you can see the Claude chatbot recognises you have asked for management-related tips. See below how it ends the answer. 

Conversational coaching intelligence of Claude AI

In this same regard, Claude’s responses read natural.

Virtual Sales Assistance

This is an incredible use case for anyone looking for products using Generative AI tools trained on the latest data. In certain cases, where a buyer is completely unaware of where to find products, Claude can be even more beneficial. 

In the author’s case, the goal was to find an extremely rare version of the Mesa Boogie Rectifier Amp – the Revision C model, which is not manufactured any more. But this result from Claude was truly informative. 

The prompt didn’t mention that this model is rare. Nor was the prompt about finding a ‘used’ model.

Claude as a virtual assistant

 

The result about Reverb.com was spot on. After a regular Google search with keyword, ‘mesa boogie rev c’, the first result was this site.

Claude AI result cross check

 


Customer Service 

Claude AI has an understanding of urgency based on customer feedback and responses. It can quickly offer solutions to keep customers satisfied. For instance, if a customer complains about a damaged product, the chatbot can understand that a refund will be the best possible solution. 

Claude AI in Action

Since 2021, Claude has so far been tested in closed alpha mode on partner platforms, including DuckDuckGo, Notion, and Quora’s experimental AI chat app – Poe. It is now integrated into products such as Zoom and Slack. 

Duckassist

Like Bing incorporates ChatGPT on top of search results, DuckDuckGo has used offerings from OpenAI and Anthropic to show generated text on search. 

Duckassist with Claude AI integration

Slack

One of the most-used and free project management tools, Slack incorporates this AI chatbot for a variety of purposes. So, simply typing @Claude in any message or thread can get the chatbot started on giving results.

Slack Claude AI

Excited to use it yet?

How to Access Claude AI

So, let’s get you started with Claude AI. We will also list some alternative ways to access other models of Claude here. 

Through Official Claude.ai Website 

If you haven’t signed up yet, just visit the claude.ai site. Click on the Continue option to Sign up and enter your phone number for the verification code.

Claude AI from Anthropic site

Claude 3 comes with three versions - Opus, Sonnet, and Haiku. 

Claude 3 pricing

Either way, you must try the free version of Claude AI. 

After signing up, you will be able to use prompts or even upload text files, including PDFs and sheets, to get them summarised. 

For one chat on the free version, there is a limit of 5 files. Now, each of these files has a maximum size limit of 10 MB. 

Click on ‘Start a New Chat’ or if you want to move to a new chat altogether, you can hit CTRL key + K. 

Through Chat.forefront.ai

Forefront is a useful AI assistant in many ways. It lets you access ChatGPT 3.5, browse the internet, use PDFs and sheets, and gives access to Claude Instant 1.2 in their free pricing tier. Beyond that, you could access the Claude+ models in the paid tiers. 

Forefront AI access to Claude

Through Poe

Poe is like Forefront, but it’s Quora’s AI app that includes many other types of chatbots. Also, you could access Claude 2 for free with some limitations here. 

Poe access to Claude

Beyond these three sites, you could also check Vercel AI Playground by creating a Hobby Account.

How to Use Claude AI for Superior Results | Prompting Best Practices

Since the LLM model, Claude, is entirely different from GPT, you can expect the responses to the same ChatGPT prompt will not be the same. Also, you must note that this model has limitations which need makeshift solutions. (We are going to go there too!)

Stick to Natural Language to Instruct

Keeping a conversational tone with naturally flowing sentences is the best way to start prompting Claude and get good responses.

To get started, here we have compiled some do’s and don’t’s of using natural vs unnatural language to instruct Claude AI. 

Do Don’t
Use full sentences Use short, robotic commands
Speak conversationally Give overly technical instructions
Use common words Assume context from previous conversations
Include relevant context Require specific keywords
Ask questions Rigid sentence structures
Use polite language Speak formally or rigidly

Example of using a conversational tone: “Claude, could you please tell me what the weather forecast is for tomorrow?”

Example of a less natural instruction: “Claude, query weather database for location coordinates 41.8781° N, 87.6298° W date 2023-10-18.”

Now, this unnatural instruction is understood by ChatGPT and Claude. 

The screenshots below show how Claude AI responds as opposed to ChatGPT with the same prompt.

CLAUDE AI

Claude AI natural language instruction

CHATGPT

ChatGPT natural language instruction

If you are interested in using AI chatbots to their fullest extent, go check ChatGPT courses

Use XML Tags

The documentation on Claude mentions using XML tags for prompts. These text can be defined for giving context, structure or focus. Tags are especially helpful for long, complex inputs where you want to organise sections or call out key phrases.

Some use cases are highlighted below.

  • Specify the topic or domain you want Claude to focus on. You can wrap a section in <domain>tags</domain> so Claude knows to interpret that content through the lens of a particular domain or topic area.
  • Clarify pronouns. Use <person> tags to disambiguate pronouns if Claude is confused. 
  • Organise a long input. You can break up long inputs into sections using tags like <section> to help Claude understand the structure and flow of your content.
  • Tags are generally not required for short, straightforward inputs and conversations. Focus on using natural language first.

Example of using tag in a prompt using Claude AI.

Prompt

<section> I’m planning a trip to Europe this summer and need some help deciding where to visit. My top choices are France, Italy and Spain. I’ll be traveling for 2 weeks by train to maximize the number of cities I can see. I’m most interested in museums, architecture, culture, and great food! </section> <section> For France, I’m considering Paris and Nice. Paris seems like an obvious choice for the art and history. But I’ve heard Nice has amazing views and food along the Mediterranean. I’m worried 2 days in Paris won’t be enough. </section> <section> For Italy, I know I want to see Rome and Florence. Rome has incredible ruins and history. But I don’t want to miss out on Florence’s art collections. I also heard Cinque Terre is beautiful. Not sure if I can fit that in! </section> <section> For Spain, Barcelona and Madrid are options, but open to other ideas too! I’d love suggestions on must-see destinations and how much time to spend in each place given my constraints. </section>

Claude XML

 


Use Documents before Instructing

Anthropic has designed Claude in such a way it gives more importance to the text at the bottom of the prompt. In that logic, the prompts’ main action words should be at the bottom, especially if you want it to read large texts and get them summarised. 

Add Examples with XML Tags

Placing examples with XML tags prevents Claude Anthropic AI from making assumptions.

Let’s illustrate two scenarios. One with examples and XML tags and the other without. 

Here are two scenarios to illustrate the difference:

Scenario 1 (with XML tag and example)

<example>

I’m looking for ideas on what to cook for dinner tonight. I have chicken, broccoli, rice, and some basic spices in my kitchen. For example, one idea I had was to cook some teriyaki chicken with steamed broccoli and rice, but I’m open to other suggestions too.  

</example>

Specific Result

Based on the specifics provided in the example, I would suggest:

– Baked chicken breasts with lemon and herbs served with roasted broccoli and rice

– Chicken fajitas with sautéed peppers and onions, served with rice and avocado

– Chicken stir fry with broccoli, carrots, bell pepper and teriyaki sauce over rice

– Chicken fried rice with broccoli, carrots, peas, eggs and soy sauce

The example helps provide context about the ingredients available and a sample idea to start. This allows Anthropic’s conversational assistant to tailor the response with relevant recipe suggestions.

Scenario 2 (without XML tag or example)

I need help figuring out what to make for dinner tonight. 

Generic Result

– Some popular simple dinner ideas are spaghetti with meat sauce, baked chicken with vegetables, tacos, salad with protein, or soup and sandwiches.

– You could look up recipes based on ingredients you have on hand or cuisine types you enjoy. 

– Consider what you have available to cook with and if you want low prep, one-pot meals or are willing to cook more involved recipes.

– Make extra portions so you have leftovers for lunches or future dinners.

– Look through cookbooks, food blogs or apps for recipe inspiration based on your needs and dietary preferences.

If interested in creating better prompts, train yourself with good prompt engineering courses!

What Makes Claude AI Unique and Safe

Claude AI has been built by Anthropic, based in the Bay Area, by former executives of OpenAI. Investors, including Salesforce and Google, are funding it. Among other big investors are Amazon and Zoom.

The uniqueness of Claude comes from making it safer for users to use AI. Ever since AI has taken over the mainstream technology culture, there have been growing scholarly debates on the risks and biases involved.

Anthropic has constantly tested and re-evaluated Claude considering human safety and how AI models should align with human values.

For that, Anthropic incorporates three components, or the HHH (Helpful, Honest, and Harmless)

It is Helpful in the best way possible for the user. The Honest component ensures that Claude AI makes little room for misinformation compared to other LLMs.  It is also Harmless, as it will not help the user perform any dangerous activity to others or society. 

These three components meet Anthropic’s mission: ‘to ensure transformative AI helps people and society flourish’. Anthropic also prioritises responsible public communication around any incidents without compromising sensitive details that could aid attacks.

This is achieved through a technique called Constitutional AI

Constitutional AI aims to create AI that has human ethics. Now, it is fundamentally built into its design through balanced subsystems, rigorous ethical training, oversight mechanisms and transparency. In fact, the goal is beneficial, trustworthy AI focused on avoiding harm.

Let’s show you one such example. Contemporary and stronger AI text generation tools such as ChatGPT and Bard make it too simple to rephrase texts by removing plagiarism with the right prompts. 

See how Claude AI brings ethics into this prompt result. 

Claude AI ethics

 

With the same prompt on ChatGPT, here is the result.

Claude AI ethics

Plagiarism this way is so easy, isn’t it? 

It may also remind you of one of the brutal criticisms of this AI bot from American philosopher, Noam Chomsky, who said it was ‘high-tech plagiarism’. 

Now if Claude AI is able to recognise the ethics involved in some kind of ‘dishonesty’ involved in readily taking pieces from some website and rephrasing the content to publish, it surely shows potential than its competition. 

What’s more, Anthropic has also created an oversight system where humans can provide feedback and corrections when the AI responds or behaves in concerning ways. This allows ongoing improvement of the AI’s decision-making.

Claude AI boasts lesser bias, better understanding of language, and provides more conversational outputs than its competition.

Mind you: Hallucinations are still inevitable in all Generative AI technology. And Claude AI is doing its best to reduce them.

About the Author
author-image
Syed Aquib Ur Rahman
Assistant Manager

Aquib is a seasoned wordsmith, having penned countless blogs for Indian and international brands. These days, he's all about digital marketing and core management subjects - not to mention his unwavering commitment ... Read Full Bio