There has been much discussion about the benefits and harms of Artificial Intelligence, especially in recent years. Some of these are unique and unexpected — such as AI’s overuse of emdashes1 — but many form complementary pairs. For instance, while AI can cause environmental harms, it can also accelerate the development and spread of green technologies. At a personal level, AI could hasten your death or keep you healthy indefinitely. At the extremes, it could cause the largest mass extinction event in history (including humans) or it could preserve every one indefinitely, even bringing species back from the dead.
For creative endeavors, there is a distinct tradeoff: while it has never been easier to get projects up and running, it is also easy to let AI do too much of the thinking and have very little personal impact on the final product. If LLMs had better taste this might not be so noticeable, but for now not having many distinct frontier models leads to the well-documented phenomenon of AI slop. Their taste is improving over time and I believe this trend will continue, but there is still a fundamental tension between how easy it is to make something and how much humans will play a role in the making of it.
As an example, making a good movie is extremely hard. It usually takes a team with years of experience, technical know-how, and other hard-to-acquire qualities multiple years to produce a good finished product2. This trend has only been accelerating in recent years with movies having bigger budgets, larger effects teams, and longer times to completion. Now, you maybe can’t make a blockbuster-level film just through prompting today’s models but you can get surprisingly far, and the day when you actually can is probably not far off. At that point you can have an LLM generate the starting prompt from even simpler instructions and you’re off to the races. You can think of other artistic mediums that are harder for AI to automate from start to finish, but between developments in robotics and the ability of current AIs to come up with detailed ideas (including specs, mock-ups, etc), all of those are on similar trajectories.
The tradeoff is that the more time you spend prompting it and specifying exactly what you want, the more time it ends up taking to get a final product that you have meaningfully shaped. In the limit this is still less effort than it takes to make the product from scratch, where humans could act as a sort of film director shaping the overall vision and accepting/rejecting AI-proposed changes with feedback. However, why couldn’t AI get just as good at this role and take the human out of the loop entirely? Unless you’re in it for the love of the game, there will always be a temptation to just let the AI take the wheel and save you all the annoyances that come with ironing out every little detail to get things exactly right. Basically, you can do everything yourself but it will be incredibly hard and there will be decreasing external incentives to do so, while if you just write a short prompt and have the AI do everything then the AI is making all the creative decisions.
In the current day and age, I think there is a happy medium that varies between people and applications. For this website, I have specific ideas in mind for how I want it to look and what I want it to do, so I am happy to let LLMs do most of the coding and go through a lot of iterations to get the look exactly right. As for writing, my goal is both to clarify my thoughts and to share them with others, so I plan on writing initial drafts from scratch while turning to AI primarily for feedback. In both of these areas, I am motivated enough not to just completely vibe through everything with minimal effort on my part. However, AI will play an essential, ongoing role, and if my motivation falters there is always the option to just vibe it out.