The AI models are constantly evolving, and so are the prompting techniques. So, let’s look at some techniques that have changed since the beginning when GPT 3.5 first arrived.

Working with examples

The best way to get to your ideal outcome faster is to use examples.

Providing examples was always a popular technique.

But as the AI models are getting more intelligent, they are also smarter at replicating the examples that you provide. Sometimes it’s also much easier to provide an example than trying to write it out.

For example (no pun intended), if you want to replicate your writing style when generating an email it could look like this:

Write an email to John about [the context of your email]. And follow the structure and writing style of this example:

<example>
A previous email that you have written.
</example> 

You can provide more than one example, but just make sure they are similar to not give mixed signals.

Use XML tags

As the models are getting more intelligent they can also process more input text. This is called the ‘context-window’ and tools like Claude can already process roughly 150.000 words or 500 pages.

It’s essential to structure your prompts in a way that separates your input data from the instructions.

Like I did with the <example> </example> tags above, those are XML tags.

Another common way is to use it like this:

<article> article text </article>

<instructions> Provide a summary of the article in less than 200 words</instructions>

It helps the AI better understand how to process your prompt, thus leading to much better output.

If you want more prompt engineering tips, I’ve created a new guide on prompt engineering that you can check out.

If one thing bothers me it's the default voice of ChatGPT.

You see it everywhere on LinkedIn and the rest of the internet.

It's extremely repetitive and monotonous, but maybe worst of all—it's extremely hard to read.

For example, let's look at a common prompt's output. The prompt would be something like this:

"Write an announcement for LinkedIn that we have a new product launched at Org called Product X

And the output will look like:

The sentences are long full with complex words, the passive voice, and adverbs.

Great for your English paper, terrible for writing on the internet.

Why simple writing matters

In the USA, 54% of adults have literacy below a 6th grade level. By using ChatGPT's default voice, you could lose out on connecting with a huge audience. The same is true in many other countries.

So how can we make more impact with ChatGPT?

The answer is simple: write simply.

Consider what Jeff Bezos, Warren Buffett, and viral LinkedIn writers like Sahil Bloom have in common. They all write extremely simply, often at a 7th grade level on average.

Simple writing demonstrates clear thinking and the ability to make complex topics easy to grasp. That is a superpower.

A simple trick to make ChatGPT write simple (without making it childish) is to add:

Write at a Flesch reading score of 80 or higher

This will result in output at a 6th grade level without sounding childish. It's perfect for business writing, which needs to be direct, concise and clear.

If you want it to take one step further to create powerful and clear messages, try out the Power Writing prompt:

Use clear, direct language and avoid complex terminology.
Aim for a Flesch reading score of 80 or higher.
Use the active voice.
Avoid adverbs.
Avoid buzzwords and instead use plain English.
Use jargon where relevant.
Avoid being salesy or overly enthusiastic and instead express calm confidence.

This will work wonders to make your writing simple, and to the point, but not simplistic.

Especially if you use ChatGPT, this modification is very needed.

While ChatGPT is the most famous LLM (AI Chatbot), it’s not the best option anymore.

Anthropic’s Claude took the throne this week on the Chatbot Arena Leaderboard.

At the moment I also recommend Claude Sonnet (free) or Claude Opus (paid).

Especially if you do a lot of creative work like writing. Claude has a great natural writing style and the ability to listen to your instructions.

For programming, data analysis, and strategy the best options are either GPT-4 or Claude’s Opus, with a slight preference to Opus.

My most important recommendation is to stop using GPT 3.5 and trade that in for another free option like Claude Sonnet or Gemini. You’ll experience a world of difference.

If you want to stay up-to-date on what the best chatbots are, I would check out this resource on Workwiz. It gets updated every time there is a new update.

🛠️ New tools:

• Hume - The first voice AI with emotional intelligence (also try out the demo!)

• Dora AI - Instantly create a website from a single prompt.

• Arcade - Create interactive product demos in minutes.

• Creatie - Create professional designs in minutes with one-click.

• DataMotto - Prepare your data for analysis with AI in minutes.

• CBI Instant Insights - Instant company and deal analysis.

What did you think of this email?

You can add more feedback after choosing an option, or you can reply to this email directly.

Enjoyed this newsletter? Forward it to a friend and have them sign up here.

See you next Thursday!

— Jerry