AI Policy
tldr: We automate certain coding tasks with AI, so that we can spend more time on the creative work that we believe human beings should be doing.
Key takeaways
- There is no AI-generated content in our app (unless it was uploaded by a user)
- Our AI use is limited to developer tooling and coding tasks; we do not use AI for design work or other creative work
- The environmental footprint of our average daily AI use is less than a typical person’s daily computer use
When I first started working on no hotdog two years ago, AI was a nascent technology that barely worked. Since then, a lot has changed.
For one, no hotdog itself has changed. The first versions that took me years to ideate and build no longer exist. (But we still have !)
But more relevant to this socio-historical moment, the field of software engineering has completely changed. That previously nascent technology? Well, let’s just say that virtually every web app in production today uses AI dev tooling. So, we have some things to talk about…
Developer tooling
When we work on no hotdog, we use the same web developer tooling that everyone else uses, which means that, yes, the code was written with AI assistance.
That being said, we do not use AI as a substitute for creative work. That is where we draw the line. All the product decisions, visual design, writing, and so on, were created by human minds. I personally spent hundreds if not thousands of hours painstakingly crafting our website and apps.
Some of my favorite copy in the app was inspired by aspects of my own life. For example, the recurring motif that “the earth was supposed to be fun” was inspired by the opening sentence of my favorite book, Feed:
I don’t take AI usage lightly – I think tech bros have been ruining the world for decades and AI is their latest installment – but we also need to be realistic about how AI tooling has impacted software development. no hotdog is a really small operation, and I wouldn’t have been able to give it the craft and care it deserves had I spent all my time manually coding database migrations and backend deployments. AI coding tools allow me to spend my time on the creative aspects that I actually enjoy (and that our users actually see) — and it is this creative work that I believe human beings should be doing.
Some people worry that AI dev tooling, even at the base coding level, impedes creativity, and I’m really sympathetic to this concern. But you have to understand that, before AI dev tooling, people didn’t manually code things like you might be imagining. They used pre-built libraries that imposed their own structure and constraints on your code and design. So, I think the solution is not to eschew coding libraries or AI dev tooling altogether, but rather to exercise an explicit conscientiousness about how our developer frameworks influence the design choices we make.
AI usage footprint estimates
These are just ballpark estimates meant to show that our AI usage footprint is very low. Don’t take the exact numbers too seriously, as there’s lots of uncertainty here.
What is an AI usage footprint? And what is ours?
Energy footprint
When you ask ChatGPT something, it sends a query to a computer in a datacenter, which does some computations, and then sends the result back to your computer. These computations use some electricity and require some cooling, resulting in some environmental impact.
In order to estimate our impact, we can look at our “tokens” used. A token is basically a chunk of text about a word long and the models write their answers one token at a time, each one costing some computation. So, the number of tokens basically tells you how much work the computers did.
Anthropic tells us we have used about 40M tokens over the past few months. I doubled the number to 80M tokens, because we did some development before this period using other misc. tools. We can then multiply this by the average energy cost per token. The energy cost per token was kind of hard to find, but most sources imply about 2 joules per token. I tripled that to 6 joules to be cautious.
So, the total energy cost to produce our entire app was equivalent to about 5 days of one average home’s energy use. When you convert this to daily energy expenditure, it’s less than your own personal computer use, assuming you are average.
Water footprint
All of our AI usage combined adds up to about one bite of a hamburger.
Carbon footprint
One passenger on a single transatlantic flight produces about 16 times more carbon than all of our AI usage combined.
Feedback
Our policy is ever-evolving as we adapt to this new technology, so we always appreciate feedback on our approach. To provide constructive feedback, please reach out to ben@nohotdog.love.