• Everyday AI
  • Posts
  • Don’t Make These 5 ChatGPT Prompting Mistakes

Don’t Make These 5 ChatGPT Prompting Mistakes

DeepMind's worries over Google defense deals, China looks for US chip access, Salesforce unveils new AI agents and more!

Author

Everyday AI
August 23, 2024

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: Sorry to break it to y’all. ChatGPT doesn’t suck. Your prompts do. Here’s how to fix them. Give it a listen.

🕵️‍♂️ Fresh Finds: Google Gemini API free tier expands to more countries, Kling AI releases pricing and a security breach with Slack AI. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: DeepMind workers concerned over Google defense deals, China looks to access U.S. chips via AWS and Salesforce unveils AI sales agents. For that and more, read on for Byte Sized News.

🚀 AI In 5: We’re pitting GPT-4o against Claude 3.5 Sonnet to see which is better at data analysis! See it here

🧠 Learn & Leveraging AI: Wondering why you’re not getting good outputs from ChatGPT? We’ll show you how to properly prompt to get good results. Keep reading for that!

↩️ Don’t miss out: Did you miss our last newsletter? We talked about OpenAI's major deal with CA, Microsoft Recall feature updates and a SearchGPT first look. Check it here!

Don’t Make These 5 ChatGPT Prompting Mistakes 🖐

Some people say that AI stinks. ChatGPT isn't any good.

Well we're here to break the news to you. ChatGPT doesn't suck. Your prompts do.

We're showing you what you're doing wrong and how to improve your ChatGPT prompts to improve your results and responses.

Join the conversation and ask Jordan any questions about ChatGPT.

Also on the pod today:

• Is ChatGPT getting lazier? 🦥
• Copy and paste super prompts 📄
• Looking for outputs vs building skillsets 🔨

It’ll be worth your 34 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – Paperguide helps you read, write and manage research, uBest provides fully automated recruitment service and Antispace is an AI OS that gamifies work.

Google - Gemini’s API free tier access has been expanded to 35 more countries.

Big Tech - Meta and Spotify’s CEOs have released a joint statement criticizing the EU for its AI regulations.

Trending in AI – Security researchers found a way that a prompt could trick Slack AI into sharing private information.

Future of Work - Over 56% of Fortune 500 companies have now listed AI as a risk factor in annual reports.

AI Models - Kling AI has announced new monthly and yearly subscription pricing.

AI in Medical – Experts are claiming that a national registry of AI medical tools could help increase transparency.

Read This – The AI bot running for Wyoming’s mayor election has now conceded and launched an alliance for AI in politics.

1. DeepMind Workers Raise Eyebrows Over Google's Defense Deals 👀

Over 200 employees from DeepMind have voiced their discontent regarding Google’s defense contracts, arguing that these agreements conflict with the company’s stated commitment to ethical AI. The letter highlights concerns about Google’s collaborations with the Israeli military, suggesting that any involvement in military operations could undermine their mission.

This situation reveals a potential culture clash within Google, especially considering DeepMind's acquisition in 2014 and promises made to avoid military surveillance.

2. China To Access U.S. AI Chips Via AWS and Cloud Services ☁️

Chinese state-linked entities are actively seeking access to advanced U.S. AI technologies through cloud services, with at least 11 organizations identified in recent public tenders. Notably, Amazon Web Services (AWS) is implicated, as Chinese intermediaries navigate around U.S. export bans on high-end Nvidia chips.

As demand for sophisticated AI models surges in China, the U.S. government is working to strengthen regulations to prevent unauthorized access to these technologies.  

3. Salesforce Unleashes Einstein Sales Coach 👥

Salesforce has unveiled its latest innovations, the Einstein SDR Agent and Sales Coach Agent, aimed at enhancing productivity for sales professionals around the clock. These AI-driven tools assist salespeople in engaging with prospects, qualifying leads, and practicing pitches while providing valuable feedback in real-time.

With sales representatives dedicating 70% of their time to non-selling activities, these advancements are set to significantly streamline the sales process

4. Anthropic Supports California's AI Bill with Caution 🤔

Anthropic, the AI competitor to OpenAI, has expressed support for California's revised SB 1047 bill, viewing it as a significant step toward ensuring safety in AI development. CEO Dario Amodei highlighted that the bill's new requirements, such as a mandatory "kill switch" for advanced models, could result in more benefits than costs.

Despite acknowledging some remaining concerns about certain ambiguities in the legislation, he remains hopeful that the changes will not hinder innovation in the rapidly evolving field.

5. Microsoft Salary Leak Reveals Big Earnings for AI Engineers 🤑

A leaked spreadsheet has come to light, showcasing the salaries and bonuses of Microsoft engineers, with over 500 employees sharing their compensation details. The findings indicate that top-level engineers are making impressive totals, with one individual reporting a compensation package of $1.23 million.

Notably, engineers in Microsoft’s newly established AI division are earning even more, averaging $377,611 in total compensation.

Claude 3.5 Sonnet vs GPT-4o Data Analysis showdown

In this head-to-head battle, we're pitting two of the most advanced large language models against each other:

Anthropic Claude 3.5 Sonnet vs GPT-4o from OpenAI.

Which one can crunch numbers and deliver actionable insights better?

🤷‍♂️ What’s Going On and Why It Matters:

ChatGPT doesn’t suck.

Your prompts do. 

Sorry not sorry, y’all. But that’s the truth. And that’s exactly what we talked about on today’s show.

Let’s get some things outta the way right now. 

  • Our team has been using the GPT technology since it came commercially available in Copy.AI in late 2020. 

  • Our host Jordan has been a keynote speaker in front of thousands of attendees on ways to properly use ChatGPT and recently even spoke at the world’s largest AI Summit in NYC on ways to get better outputs from ChatGPT.

  •  Almost anything you read about prompting (especially on Twitter and LinkedIn) is wrong, and it’s just someone trying to sell you a crappy prompting guide or to get your email for useless “Superprompts” ebook. lolz.

  • We’ve taught thousands of people on proper prompting, from entrepreneurs and marketers to business owners and F100 execs. And it’s all for free in our Prime, Prompt, Polish course.  

Katie cut it straight in today’s livestream comments. 👇

Back to what we learned on today’s rant.

Err… show. 

There’s common misconceptions floating around, especially recently, that ChatGPT is getting lazy. And those prompts aren’t getting the responses that they should. 

Spoiler alert — it’s because most people don’t understand the 101 of prompting a large language model. That’s why we went deep on the topic today, and also dove into: 

  • Using copy and paste or zero-shot prompting is a time waster and produces low-quality results

  • Building skill sets inside ChatGPT leads to consistently high-quality outputs

  • Multi-shot prompting produces better results than zero-shot prompting

  • Claiming expertise and years of experience does not greatly improve the outputs from ChatGPT

  • Large language models work differently from search engines

  • Treating a new chat inside ChatGPT like training an employee is essential for success

Alright, let’s grab the takeaways and get rid of bad prompting forever, shall we? 

Lez get it. 👇

🦾How You Can Leverage:

Today’s show was kinda like a 2-for-1. 

Not only did we debunk the whole ‘ChatGPT stinks because of my mediocre prompting’ piece, but we also broke down the 5 biggest mistakes when it comes to prompting inside ChatGPT. 

What do we know about prompting? 

Take it from Maybritt.

(If you want access to our free prompting course, just reply ‘PPP’ to this email.)

Alright, now let’s talk about the 5 most common mistakes when prompting ChatGPT, and what to do about them. 

1 – Using copy/paste super prompts 📜

Sorry Billy Boy. (He’s that 22-year-old “ChatGPT expert” who’s really just a failed NFT bro trying to sell you a useless prompt book.) 

Your prompts don’t work. 

Any Large Language Model benchmarking study (like the one we talked about yesterday while roasting Google Gemini) shows that multi-step prompts beat the snot outta zero-shot prompts. (That’s essentially what copy and paste prompting is) 

Try this:

For the love of Cheez Its, why haven’t you taken our free Prime, Prompt Polish course yet, like Mike? 

Reply PPP to this email and we’ll send you the secret registration page. 

2 – Outputs vs. building skillsets 🧑‍🏭️

Guess what? 

Large Language Models like ChatGPT are not search engines. If you’re thinking of LLMs like a glorified search engine, you’ve got it all wrong. 

Try this:

Erase any correlation between a single input and a single output that you have. LLMs are not search engines, and comparatively deliver hot garbage outputs when you only enter one input. 

Go read this breakdown from one of our favorite websites that shows the difference between how search engines and LLMs work. 

3 – Not using skill-based chats 💬

Don’t hit that “New Chat” button every time you go into ChatGPT. 

Naaaaah shorties. 

When using ChatGPT, you should be training each new chat on a very specific skillset. 

Try this: 

Instead of using one chat for hundreds of purposes, create hundreds of specific chats and train each chat in one very specific skillset. 

4 – Expert with years of experience 🧐

“You are an expert copywriter with 20 years of experience” 

Yuckgrossbarf. 🤮

If you think this will give you ideal outputs, think again.

Seeing as LLMs are trained on the history of the internet (and a lot more!), there’s a straight up cornucopia of hot-garbage advice on the internet coming from “experts with 20 years of experience” in field A, B, and C. 

Try this: 

Can we please stop pretending this works and filling Billy Boy’s impossibly ginormous ego?

K thanks.

Instead, when prompting ChatGPT, give ample examples of what defines its role. Give real-world examples of experts it should emulate, or define characteristics about ChatGPT’s role for a chat.  

5 – Looking at ChatGPT as a shortcut 🚫

Catch-22 here. We get it. 

ChatGPT is the ultimate biz building shortcut, but you’ve gotta put the work in to build the actual bridge to connect a new chat’s capabilities from zero to expert level. 

Try this:

That’s legit the whole premise of our free Prime, Prompt Polish course. Have you really not replied ‘PPP’ yet to this email? 

(Or your fave LLM like Claude, Gemini, Copilot, etc)

Reply

or to participate.