The Big AI Rundown - Part 2

Explore Claude and my custom prompt and find out what’s wrong with Perplexity

Welcome back, everyone! 👋

For a detailed showcase of ChatGPT and Gemini, check the previous post:

The Big AI Rundown - Part 1

Today, we’ll continue our deep dive into AI applications, starting with my next personal favorite: Claude.

Initially, I wanted to include Grok in this research. However, as I went deeper with the research with these other tools, I realized Grok can’t compete with most of them, and it’s tied to X, so for now, I’m leaving it out.

I’ll keep putting in the work and distilling what I learn/learned as a software engineer/consultant. I write weekly, no-fluff issues. Don’t miss them and sign up here:

Claude

Antropic’s AI app disrupted the market, particularly for programmers, who received significantly better output than ChatGPT. But let’s see if it also outperforms other AI apps in different areas.

Writing Capabilities

Style & Writing Aid

It’s not as seamless as ChatGPT’s memory, but it may be more straightforward if you’d like to use Claude for different tasks with different styles: newsletters, blog posts, professional emails, and cold outreach to potential clients.

You can easily switch between built-in and custom styles during prompts. You can edit a style and generate examples to see if it matches your expectations.

Use case: if you constantly battle with AI apps to brainstorm on newsletter and blog post titles and have to repeatedly tell the same things to the app, you should write up a style for that.

Here’s one thing to watch out for if you don’t want to sound like a robot.

When you create a writing style in their editor and save it, Claude analyzes it and rewrites the entire example you gave (I’m not sure if this is intentional or a bug).

The style prompt I gave it is in the top left picture. Later, when I edited it through the UI, I saw what was in the right panel—a complete rewrite of the example I gave.

Luckily, you can edit the style manually in the UI, and the difference is huge.

Check the title suggestions after manually editing the style and reverting it to what I initially used for the style prompt (left picture) instead of what it generated when I left Claude’s version to generate the titles (right picture).

The titles generated based on my original prompt read more interesting and lively, and the titles on the right are more washed out. But let me know if this is only me and you think there’s no significant difference.

You can create this prompt for yourself:

Help me write newsletter and blog post titles that make it relevant to the reader, create some urgency, can be polarizing, but aren't too clickbaity.

Here are some examples of past newsletter titles that fit this description:

<use the titles from your best-performing articles to give the AI an example>

Don’t judge me. I love to use a little clickbait every now and then, but only if I can deliver on the bait. Is it even a bait then? 🤔

Research Assitance

I found that asking AI apps for the latest version of something is a good indicator of what you can expect.

To the question, “What’s the latest version of Node.js?” Claude tells you when its last update was and what version was the latest back then and suggests checking the official site of Node.js for an exact answer.

At least it is not misleading like Gemini, which tells you the wrong version and supports it with a link.

Usability and Interface

The most annoying thing with current AI apps stems from the chat-like interface. New messages push old messages further up. Getting an overview of the most recent developments in your chat becomes tedious, with a lot of “Find on page” and scrolling. Yes, you could ask the AI app what was the thing it wrote before, but that would only extend the chat further.

Different ways were found to solve this seemingly simple challenge. I prefer Claude’s solution over ChatGPT’s canvas because it blends better into the workflow. Also, if you work with existing content, the content you paste into the chat is nicely separated.

Here’s how it looked like when I asked Claude to rewrite one of my blog posts about testing React applications where I initially used Jest but wanted to have a version with Vitest:

Customization and Control

There is no plugin ecosystem like in ChatGPT or Gemini, but it offers a RAG-like capability with Projects, available on the paid plans.

With Projects, you can give Claude documents, text, code, or other files that it’ll use as additional context to answer your questions.

Pricing and Value

Claude has a simple pricing. Their free plan has a limited usage quota that resets daily and depends on demand. Pro users pay €18+VAT/month, and Team users €23+VAT/month per user.

Each plan provides more usage quota than the previous tier, but the most interesting part is Collaboration, which is available on the Team plan.

Collaboration Features

ChatGPTs and Gemini’s collaboration focused on you collaborating with the AI on some of your documents.

On Claude’s Team plan you can collaborate using Claude chats with your team members inside the same Project environment.

Security and Privacy

You can read more about how Antropic handles your data here. Here’s a non-exhaustive list of things worth mentioning.

Data Privacy

• Antropic retains your personal data as long as reasonably necessary, however

• You can delete conversations, which will be removed immediately from the history and deleted from the backend within 30 days.

• For businesses, you can be approved for zero retention, which applies only to enterprise API customers.

• Claude may anonymize or de-identify personal data for research or statistical purposes and retain this information without further notice to you.

Compliance

Anthropic has a fantastic page at trust.anthropic.com that shows you their certifications. They constantly monitor infrastructure, organization, and product security.

Whether Antropic is GDPR compliant, the short answer is Yes. You can report GDPR-related issues to their Data Protection Officer at dpo@anthropic.com.

You’ll find the entire Privacy Policy here. For GDPR, read titles 5 and 9.

Ideal use Cases

Although Claude has become known for its coding capabilities, with Writing Styles, it can be your multi-purpose writing companion, matching ChatGPT’s GPT feature they only provide in their Pro plan. Projects and the Team plan make it ideal for AI-driven collaboration across multiple team members.

Perplexity

Despite being released around the same time as ChatGPT, it got much less attention. It has some catching up to do, as well as problems around sustainability, which explain the recent backslash they received from users after they started showing sponsored results in the chat. Let’s see how it compares to the others.