Another day, another idiotic AI headline that people are losing their minds over.

Though the headline is eye catching, very few of the hype articles that ran it actually linked the full study. Ironically, running this report through ChatGPT I get a better summary and analysis than any of the articles out there. Also, a good data pro tip, if the percentage ends in a 0 or 5 you should immediately be suspect, as the real world doesn’t often output numbers like this.

This report also shows the type of (incorrect) mindset that businesses still have, where if they make a specific technology investment problems will be solved. In the case of generative AI (or any technology), the implementation is really the first step. You have to remember that people need to use these tools, and if there is no training or development internally it’s a useful as a paperweight. The study itself recognizes this fact:

Organizations aren’t seeing the transformative effects of generative AI because they are not adopting them in that manner, but are focusing on improving productivity first. And uniquely in this case, employees are using their own personal LLMs at a much higher rate than internal enterprise LLMs (90% personal LLMS vs. 40% enterprise LLMs). As I had mentioned in my last newsletter, LLMs are benefitting more from the public, bottoms up approach.

The other big takeaway from the study is where the funding is going and where the true ROI is. 50% of budgets are going to the flashier sales and marketing initiatives, but the back office automation has a higher ROI. It follows a similar trajectory as traditional AI: everyone wants to apply it to moonshots but would be better served by applying it to the “boring” work.

A friendly reminder: any time you see a hype headline, go to the primary source of the study and you’ll be ahead of 94.2% of your peers.

AI diagnoses are coming — Microsoft unveils its plan for medical superintelligence, combining imaging, language models, and reasoning to move AI closer to doctor-level diagnosis. Read more →

Alpha School is reinventing education — Just 2 hours of AI-personalized learning outperform traditional classrooms. The rest of the day? Leadership, life skills, and curiosity-building. This is what school could look like. Read more →

Data centers are the new office towers — In a wild shift, U.S. data center construction is now nearly equal to office construction. The physical backbone of AI is taking shape. Read more →

Effective prompting can yield thousands of hours in productivity and millions of dollars in value. Starting is the hardest part, and I would argue setting up a repeatable framework is the second hardest part.

Here is a simple framework to level up your prompting game:

Role → Task → Context → Format → Constraints → Iteration

I’ve included a real world example below, that you can copy and paste into your LLM of choice. As you develop your prompting, you don’t have to put “Role, Task, etc” into the prompt itself, however in the beginning can be useful so that you don’t forget those parts.

Role → You are a strategic communications coach with expertise in finance and executive storytelling

Task → Help me prepare a 3 slide executive summary Context → for a budget meeting with my CFO focused on cutting operational costs by 10% without layoffs

Format → The format should be slide headlines with 2 - 3 bullet points per slide

Constraints → Keep it under 150 words total, use plain language, and avoid jargon

Iteration → After your first version, ask me follow-up questions to refine it

Applying AI to tax prep, sound boring? Remember, applying it to the back office is where the ROI is at.

KPMG took this to the next level by creating a 100 page long prompt for their TaxBot agent. It acts as a digital tax agent, going through hundreds of scenarios, questions and edge cases, and providing answers that would normally take hours of manual review.

It shows the power of prompting and the ability to take monotonous, yet complex work and package it into an agent. This turns a two week process into a one day process for the company. Now imagine if you do this for 10 processes? 100? 1,000? That is where the real value is.

Being in AI, weeks can feel like decades and decades can feel like weeks. One minute generative AI seems like a false idol, and then the next minute there is a massive breakthrough. That is the nature of an exponential technology, and there isn’t a predictable increase in a capabilities over the long term.

What we’re not hearing about enough are the individual stories of people making their lives better by using LLMs. Saving an hour or two a day at work, having it help them organize their personal life, and freeing up time to use as they see fit. Those individuals plugging away at improving their AI skills are creating an ever widening gap with their peers that are viewing this as a flash in the pan.

All AI is truly lifechanging when utilized day-to-day and integrated into what you do. There isn’t a magic bullet for improving, and practice makes perfect. So my challenge to you is to spend 20 minutes a day with your LLM of choice, struggle through the learning phase, and experiment with it to make your life (and the lives around you) better.

-Ylan