I’ll probably throw together a blog soon of patterns that have worked for me.

To recap, I was in the trough of disillusionment at this time last year while almost everyone else was at the peak of inflated expectations.

Since then, I’ve started the slope of enlightenment while everyone else has begun toward disillusionment.

This model should help you understand why I was a contrarian before and am now a contrarian despite switching positions. 😃 

People are even starting to refer to them as LLMs, understanding that AI is a giant cluster of ideas and not just what ChatGPT does. This maturity is good.

Ultimately, as expected, AI has not lived up to expectations. Worse still, we’re getting diminishing returns on model complexity and training data size. We fell short of the singularity—not just in absolute terms but even in terms of “getting from here to there.” We’ll need a few more breakthroughs.

Nonetheless, large language models still have a lot of uses.

Context, Collaboration, and Concreteness

The workflow I wanted to discuss today is one I’ve used in reviewing and revising the Guildmaster’s strategy for 2024.

I’ve used AI pretty heavily in this endeavor. But I didn’t just go to a prompt and type in “Give me a great strategy” - that won’t work. Anyone who’s done that may have been impressed at first, but every time after that, when you received the same strategy, it may have been a letdown. Worse still, if you realize everyone else who asked that question got the same answer.

The key to getting tailored help is providing as much context as possible.

You can do this collaboratively with the LLM. First, try to talk in terms of off-the-shelf ideas because those are going to weigh heavily in its training set. For instance, I started with the “Lean Canvas” model rather than a general strategy.

I filled out everything I could - I’m the expert. I used AI as a backup when I ran out of ideas to prevent writer’s block or to double-check my answers. You need to treat the LLM like the most junior employee.

When prompting the LLM, be as concrete as possible. There are more advanced “prompt engineering” tricks, but the easiest thing to learn is to remember you’re talking to the most junior employee—be as specific, detailed, and concrete as possible.

This employee is forgetful, so don’t jam all your instructions into a single prompt—engage in long, concrete, detailed, and specific conversations with it.

Eventually, you find a stone to take a step across this stream and have a finished artifact. In this case, it was the first draft of a Lean Canvas. Now you have context for all future questions.

Any time you engage with the LLM on anything to do with your business — upload the pdf, tailor your prompt to read “Based on the business described in the Lean Canvas in the uploaded pdf, <my question here>.”

You can continue to climb this ladder by generating more and more artifacts. For example, you might want to dive into your marketing and think about branding - start with the Lean Canvas and get some artifact out of the conversation - a unique value proposition or a set of branding guidelines. Whatever you do, the goal is to generate a written artifact that will be used as context for your next goal.

Eventually, you may be doing detailed work—you’d upload your strategy PDF, your branding guidelines, and other things in these conversations. You take baby steps forward, asking it to double-check your ideas when you’ve got flow and to fill in the gaps when you don’t so you can keep going. All the while, you “remind” it what you’ve done before via these string of artifacts.

The best part is that all these artifacts are useful to you as well. They begin to build up a lot of standard operating procedures and policies that will help you hire, train, and sell your idea to investors.

Do you think this workflow would help you? What workflows have you found helpful as you climb the slope of enlightenment?