🧠➡️🤖💼📈 Want the best tips for desktop productivity for AI?
This week, I have them. Read on and learn how to improve your productivity.

AI Hype Cycle

Think back to the GameStop saga—a retail investor frenzy cheered on by an online personality, Keith Gill, known as Roaring Kitty, turned a struggling video game retailer into a Wall Street sensation. The movie Dumb Money documents it — ironically, it made $20 million on a production budget of $30 million. GameStop short sellers lost big to the tune of $1.3 billion in this past May’s rally.

Now, consider the recent surge in AI stocks. Like GameStop, the AI sector is riding a wave of unprecedented hype, driving valuations to new heights.

Ken Griffin, the founder and CEO of Citadel —one of the largest hedge funds- is hitting pause on the AI hype, saying he’s unconvinced that the tech will start replacing jobs in the next three years. Speaking to Citadel’s new class of interns in New York, Griffin acknowledged artificial intelligence had reached an “inflection point” with the rollout of large language models but disputed that LLM-based tools like OpenAI’s ChatGPT would over the next three years become more valuable than the top talent he’s recruiting straight out of universities. Though I am thinking with his deep pockets, he’s probably got access to the creme de la creme, not the C students from City Community College.

Wall Street is also becoming more skeptical of artificial intelligence hype helping to power stocks. Some analysts even questioned the central premises of the “AI revolution,” with Goldman Sachs’ Jim Covello, the bank’s global head of equity research, expressing doubts that the technology will be as profitable or influential as its boosters claim.

My take is that we are experiencing a high-tech version of the Tulip Mania but with a different twist. In the 1600s, investors flocked to the new and novel flowers because nothing like that existed. Today, you can grab a pack of bulbs at Lowes for under $10. While beautiful, there isn’t much utility in tulips.

Generative AI, unlike tulips, can create substantial value. However, a distinct line exists between utilizing technology and investing capital speculatively to jump on the next big trend early. Consider Scale.ai as an example - the startup's valuation of $13.8 billion, resulting from a billion-dollar investment round in May, is likely inflated regardless of the valuation method used. This eight-year-old startup primarily handles the more mundane tasks in AI that significant players like Alphabet, Meta Platforms, and OpenAI prefer to avoid. Many of these tasks recently involve hiring individuals with advanced degrees to assist AI models in determining the most effective user responses.

For my money, I’ll keep investing in AI skills to keep me competitive and stay away from the AI speculative mania. Read on if you want to learn about how to gain those skills.

AI TL;DR

Generative AI News, Tips, and Apps

• The big winners of the AI boom are the most boring companies imaginable - As I noted in a previous newsletter, what’s keeping Nvidia’s CEO, Jensen Huang up at night is the lack of data center space and power. That’s why it’s no surprise that these stocks are climbing. On a single-stock basis, Digital Realty Trust — the only data-center real-estate investment trust listed on the New York Stock Exchange — has jumped 38% over the past year, while the Global X Data Center & Digital Infrastructure ETF has climbed 12%. Those stock returns far exceed the 4% gain for the iShares Core US REIT ETF over the same period.

  Stocks that aid electricity use in data centers have also had a strong 2024. Super Micro Computer — whose liquid-cooling technology has been called a must-have for AI hardware — is up 200% year to date. Nvidia may be the king of AI, but even it is "only" up 150% in 2024.

  Others are asking, “Will AI Ever Pay Off?” Those Footing the Bill Are Worrying Already.

• Apple Joins Open AI’s Board as an Observer - Apple will get an observer role on OpenAI's board as part of a landmark AI agreement announced last month, Phil Schiller, the head of Apple's App Store and its former marketing chief. According to the report, an observer can attend board meetings without being able to vote or exercise other powers that directors usually have. Observers, however, do gain insights into how decisions are made at the company. The move comes on the heels of Apple's announcement in June, bringing OpenAI's chatbot ChatGPT to its devices and integrating its new "Apple Intelligence" technology across its suite of apps, including virtual assistant Siri.

• Meta’s Multi-token prediction models now open for research - LLMs work by taking our natural language, breaking it up into something called, ‘tokens”, then using probabilities to guess the next word. However, Meta has been working on a new method. The implications of this breakthrough, as AI models balloon in size and complexity, could curb their voracious appetite for computational power, which has raised concerns about cost and environmental impact. Meta’s multi-token prediction method might curb this trend, making advanced AI more accessible and sustainable. They have released several new models on Hugging Face, but don’t get too excited. We’ve yet to see them work, but it is potentially a way to increase the efficiency of LLMs.

• This is Why Non-Profit CEOS can’t have nice things - OpenAI CEO, Sam Altman, was recently spotted driving the streets in his $3 million Koenigsegg Regera. The Twitterverse, or is it the Xverse, was going nuts, trolling him for his excess.

Feature Story

10x Your Productivity with These AI Skills

Do you remember last year's headlines about how Prompt Engineering
was it going to be the hottest new job?

Jon Stewart, the host of "The Daily Show," called prompt engineers the "asks questions guy."

The prompt engineering job is a combination of skills: using well-crafted prompts to get good results, Optimizing Retrieval-Augmented Generation (RAG) applications, reducing LLM API costs by minimizing the tokens required to generate responses and ensuring the accuracy of those results.

These jobs pay big bucks. A quick search on Indeed shows that over 200 jobs are listed in the U.S., spending over $140,000. However, I think prompt engineering as a skill can yield that kind of return for business people even if they aren't creating AI applications.

I've spent the last few years getting geeky about using well-formed prompts to get accurate answers from AI. In many tasks, it's 10x-ed my productivity.

Over time, I think the need for well-structured prompts for most tasks will be negated by advances in LLMs' ability to predict results based on preferences. But until AI is implanted through some Elon Musk-esque neural device, we must "ask questions" appropriately to get good results.

In the meantime, I think the skills required to use these nascent AI systems will yield much better results. In that spirit, I’ve gathered some tips to help you significantly improve your daily work.

One of the things about AI is that it doesn’t seem to use natural language. By natural language, I feel like it writes like someone who has been practicing for a spelling bee by reading the thesaurus.

To get read of those awkward words and make it sound more like a regular person, I use this in my prompt chains all the time:

"Write this so that it will pass an AI detection checker" 

This is a quick way to eliminate overused AI words like delves, tapestry, and landscape and make your AI-generated writing sound better.

Also, I found that every time I use this with Claude, it passes the QuillBot AI-Checker (another good tool for making sure your AI-generated content doesn't sound like it was written by a robot).

Or when I am brainstorming for a topic, a tagline, or other piece of text.
After I get something close from my initial prompt, I can reduce the back and forth by typing the following into the chat:

Iterate  10x

It's a way to take a mediocre result and get a great one quickly.

As an example, I put together a short video on how to write my newsletter subject lines. Specifically, I am using inspiration from a Twitter thread from Brad Wolverton, Editorial Director of Hubspot Media, who is the editorial director of The Hustle. I used his domain knowledge to create a prompt on how to write a good newsletter. I then provided a few examples to guide the prompt (more on that later). Then, I went back and forth until I was happy with the result. Do you remember the 10x iteration tip from earlier?

Also note that the toolbar on the side is PromptForge, a tool I use to save my prompts so I can run them when needed. It’s a huge time saver and allows me to have a place to store them that’s accessible on ChatGPT (works with Claude and Gemini too).

Don’t Just Write Prompts, Improve Them

Crafting great prompts isn't just a one-and-done deal. It's more like cooking – you keep tweaking the recipe until it's good.

Whenever we fine-tune our prompts, we improve the method of expressing our desired outcomes to the model.

One way to drastically improve the results is to include examples of what you want and don't want.

Do this for repetitive tasks, like creating reports or sales outreach emails. Use an example when you have nailed the format you like. This will guide the output to be better down the road.

Here's how to structure this in a markdown prompt:

1. Start with clear, specific instructions for the desired output.

2. Add a #Good Examples section with 2-3 high-quality examples demonstrating the ideal responses.

3. Include a #Bad Examples section with 2-3 examples of poor or incorrect responses to avoid.

4. Test the prompt with various inputs and analyze the results.

5. Based on the output, refine the instructions and examples iteratively.

6. Gradually expand the examples to cover edge cases and improve performance.

Here's a template for the prompt.

# Role
[Act as an expert, E.g. salesperson, data analyst, marketer, etc.]

# Objective 
[What do you want the outcome to be.]

# Task Instructions
[Clear, specific instructions for the desired output]

# Good Examples
Example 1:
[Ideal response]

Example 2:
[Ideal response]

# Bad Examples
Example 1:
Output: [Poor or incorrect response]
Explanation: [Why this response is inadequate]

Example 2:
Output: [Poor or incorrect response]
Explanation: [Why this response is inadequate]

Prompt for Feature Image

Here’s the prompt I used in Midjourney to generate the image for this week’s feature article. Note I used the colors to match the branding on The AIE, and I chose a 35mm lens, which gives the images the appearance of a real photo. You can choose other formats like illustration, comic book, or even surrealistic painting to get different styles.

/imagine prompt: A 35mm lens photograph of an energetic hackathon scene, featuring diverse teams collaborating on AI projects. Colors #CC3333 and #3399CC add vibrancy and contrast, with participants displaying enthusiasm and determination under bright, focused lighting. --ar 4:3 --stylize 800

Prompt of the Week

Data Analysis

One of the places that ChatGPT shines is in data analysis. This used to be a separate feature, but now it’s available throughout the ChatGPT interface. This prompt will allow you to analyze any data. Just create a report or extract the raw data from almost any system. In this example, I downloaded my LinkedIn analytics and then asked ChatGPT to provide deep analysis. You could do the same with your web stats, sales reports, or whatever you like to add context to the data type.

 (H/T to Ashley Gross on LinkedIn, where I got this tip)

How to Use This Prompt

Replace the “LinkedIn impressions” with the type of report you are analyzing, and then cut and paste this into ChatGPT. You can also try converting this for Claude with this Claude 3 Metaprompt-Based Prompt Converter GPT. (Disclaimer: I haven’t used it for this specific prompt yet but Claude Sonnet 3.5 is a very capable model, so I will eventually.)

My AI Toolbox

AI Tools I am Using This Week

I decided to dedicate this section to the latest AI tools I use. It will be updated, but everything on this list is something I use or am experimenting with. I’ll keep my master list on the AIE website here.

• Krea - Krea AI is an AI-powered design, video, and photo generator that enables users to create diverse visual content such as cartoons, logo illusions, patterns, and videos1. It features a comprehensive image editor for real-time editing and is considered a solid alternative to other AI art generators like Midjourney and Adobe Firefly1. The platform stands out for its real-time image generation capability, responding swiftly to user inputs with various creative tools.

• Sider - Sider is your AI sidekick, seamlessly integrating into your daily workflow. It starts as a Chrome/Edge extension, making browsing, reading, and writing more accessible. Sider helps you read and write articles in the sidebar on all websites. It supports the GPT3.5/GPT-4o model, smart internet access, YouTube summary, ChatPDF, AI painting, and AI chatbots with ChatGPT, Claude, and Google Gemini! Be more productive effortlessly. Use this link to explore the advanced AI model Claude 3.5 Sonnet for free.

AI Toolbox Prompt

Here’s the prompt for this week’s AI toolbox. By specifying the Pixar 3D scene it looks like it could be in an animated movie, you could say, “A Marvel Style Comic” or even a “Normal Rockwell painting” and it would mimic that style.

/imagine prompt: A Pixar 3D scene featuring a playful AI toolbox in a modern office, tools like chatbots, data graphs, and robots interacting with people and a friendly dog. Colors #CC3333 and #3399CC dominate, with bright, cheerful expressions and warm, inviting lighting. --chaos 10 --ar 4:3 --stylize 1000

I appreciate your support.