Timeline Dilation
Have you tried playing around with Google’s ‘Nano-Banana’ image generation AI yet? I spent maybe 30 minutes this week using it and was blown away. After 2 minutes making a quick sketch on an envelope of a mansion (front, top and side view), it produced a remarkable realistic photo as if the mansion was real. With a few different words, you could change the style or materials or landscape to anything imaginable.
I also saw some uses where people would ask it to produce engineering drawings of existing structures, with force diagrams, dimensions and all. Very cool. For any designers out there, this sure seems like a game changer- especially for ideation.
Slow Progress
Who would think it would take a full month to train an AI voice agent to take phone orders for the simplest restaurant imaginable- a burger joint? After selecting a single restaurant to focus on, I’m up to over 16 pages of logic required to make the AI: 1) appear human-like in its responses (brief and efficient with its words, with sound common sense), and 2) flexible enough in its ability to correctly interpret and respond to a wide range of customer requests and language.
If the underlying AI itself was sensible enough, my prompt could be boiled down to a few sentences. But since the current state of AI is still in its infancy, this presents a very fortunate opportunity for me by creating a competitive moat, making it more difficult for others to compete.
The pages of logic I’m forced to implement in order to achieve acceptable behavior restricts the AI’s freedom. It’s like a government’s laws. All laws can only restrict peoples’ freedoms, they can never expand them. If the people are sensible, the number of laws required to keep them in line becomes much fewer and less severe. The more responsible the people become, the more freedom they earn and the smaller the government they need. In this metaphor, I’m ashamed to be the government, writing the laws for my AI agent to follow.
I skipped my update last week out of the kindness of my heart; I didn’t want to bore you all to death with the slowness of my progress, despite late nights of work.
With hundreds of pages worth of new versions of logic tested these last few weeks, each time one bug appears to be fixed, a new, often stupider bug seems to pop up. This weekend, I’m working on a thorough test plan, starting on the most common potential customer requests. In the past, I focused on solving complex examples, and was surprised to learn about problems with some of the most basic requests.
As of this moment, I am seeing no issues which would prevent me from visiting my first restaurant on Monday for a demo. But, I have thought that before. After some thorough testing this weekend of basic requests, let’s hope that remains the case so I can start doing business.
Extracurriculars This Week:
Playing cello for my newborn. He hasn’t cried while listening, which I’ll take as a great review. Favorite song right now is Greensleeves, from the late 1500s, with its driving cyclical melody
Watching Pluribus- fantastic new science fiction show by the guy who made Breaking Bad and Better Call Saul. Highly recommend.
As always, check out the book ‘The Preparation’ if you haven’t already. Follow its simple instructions while pursuing whatever you are most passionate about in life, and share it here on Substack.