Hey Reader,

This one thing really baffles me about people using AI.

I went to a friend's birthday party this weekend, and one of my old friends I used to ride BMX with said this:

Inside, I was like:

But then I nodded, and said: "Yeah, and you can do so much more cool stuff with it."

I wanna pause here for a moment, because this is critical:

Do.

Not.

Use.

Chat.

GPT.

Like.

Google.

Don't use it for information. That's the worst thing to use it for.

Here's why:

ChatGPT doesn't know when its wrong

This is because ChatGPT is just a very expensive predictive keyboard. It just calculates and says what's the most likely word a person would say next, based on:

1. The prompt
2. The chat thread previous messages
3. It's training data

And even asking it questions that it has been most likely trained on, it can give false answers, yet present them with utmost confidence.

Give this one a try: Ask ChatGPT or any LLM you use, who was the 11th man to walk on the moon. The correct answer is Eugene Andrew "Gene" Cernan, according to Wikipedia list of people who have walked on the moon.​

This happened in 1972, so ChatGPT was surely trained on this information. There is even a sentence on the wikipedia page that explicitly says: "Cernan was the first to egress but was the last person back in the lunar module, so he was the 11th to walk on the Moon and the most recent to have walked on it."

Yet this is what ChatGPT says to me:

Which is FLAT OUT WRONG! Charles Duke was the 10th man to walk on the moon.

For comparison, I also asked Claude 3.5 Sonnet:

This response is slightly better, but still wrong.

So if you don't check these responses and accept them as truth, you're being fooled, and if you use them in your communication towards others, you're fooling them as well.

So what should you use ChatGPT for?

If using it for information is unsafe, what should we use it for then?

To put it into one sentence, you should use ChatGPT to process and manipulate information.

These are the facts where these LLMs can really shine, and where they can save a lot of time in professional work.

Here's a list of things that I use ChatGPT for on a weekly basis:

1. Identify blind spots: "Here are my thoughts on a topic. Act as a critic, and ask me questions or give me feedback on what am I missing here."
2. Write documentations in certain format: "Here is a meeting transcript, act as a diligent notetaker, and give me a bullet point list on the key information that was said on the meeting. Under that, give me action items that were assigned during the meeting. Then, give a summary."
3. Prepare for workshops: "I'm doing a workshop on X topic. How could I teach the following things to my students in a practical workshop? [Learning outcomes inserted here]" - this gives me pretty good ideas to start putting it together. The more I know about the topic, the less I rely on ChatGPT to fill it with content, and the more I can just get guidance on the framework, sample exercise generation, teaching methods, etc.
4. Summarize articles, YT videos: "Here is a YT video transcript/article that interest me. What are the key ideas?" NOTE: ChatGPT can't take the transcript of a YT video from the link. To do that, you'll need to use a Custom GPT like this one that can access Youtube video transcripts via the YT API
5. Generate SOPs (or any internal documentation): "Act as an operations manager, and generate an SOP for this task: [then I provide a description of the task, steps I think we should take, or just paste in a step-by-step online tutorial that I want to do as well]
6. Content brainstorming: "I'm trying to teach people how to use AI for work, here is the course curriculum that's our key IP. Based on this, put together a content calendar, brainstorm blog post, YT video ideas, etc."

The key distinction is that I don't let ChatGPT lead me astray. I provide the key information whether it's from my brain or a reliable source, instead of letting it provide it.

Parametric data (information that's stored in ChatGPT's system because it was trained on it) is less reliable than verified sources, and this is a current technological challenge among ALL Language Models.

The tech is just not there yet.

So until we have a "know-it-all AI", don't use it like Google, don't search for information with it (unless you TELL IT to browse to find the answer online). Instead, use it to process and manipulate information into different formats.

Wondering how to do all this?

Our Prompt Master AI course is our University-approved self-paced online course about using AI for work. I teach everything you need to know about using generative AI tools like ChatGPT or Claude for work. Enroll today, and use the words "WEBINAR150" at checkout to get $150 OFF the full price, while its available. More than 1,300+ students are satisfied, will you be the next?

Best,
Dave

P.S: We have a 90-day money-back guarantee on the Prompt Master AI Course, so if you don't like it, we'll refund your entire purchase if you email us within 90-days of enrollment.