Now that I’ve synthesized the advice, we can work backwards to figure out how to find good ideas:
Get deeply involved with a weird industrial/technological niche that nobody understands very well and get to know all the pain points in that sector from personal experience. Be trying to do stuff that nobody’s done before in some domain, where fundamental ideas have barely been reduced to practice. Make friends with the most capable people you meet along the way. Come up with ideas as often as you can, then find the cheapest, fastest ways to rule them out until you find something you just can’t get away from. Then take a month to build it and try to sell it immediately.
Right now, this might be something like services related to prompt engineering.
Maybe… a prompt engineering analytics company that compiles an enormous dataset of metrics on how well engineered prompts perform on specific tasks for a particular type of model, then sells that data to companies.
I recently came across a startup building advanced ChatGPT-cheating detection software that takes the text-entry behavior of the user into a Google Doc into account along with analysis of the text itself.
I’d also note that the whole class of “good tech startup ideas” has been under optimization pressure for decades now, which means that these ideas, which may have accurately reflected where to find alpha when Paul Graham was making his first billion, may not apply any longer. By the time you have this much advice easy consensus available by proven experts, it might be the worst sort of advice to take.
Now that I’ve synthesized the advice, we can work backwards to figure out how to find good ideas:
Get deeply involved with a weird industrial/technological niche that nobody understands very well and get to know all the pain points in that sector from personal experience. Be trying to do stuff that nobody’s done before in some domain, where fundamental ideas have barely been reduced to practice. Make friends with the most capable people you meet along the way. Come up with ideas as often as you can, then find the cheapest, fastest ways to rule them out until you find something you just can’t get away from. Then take a month to build it and try to sell it immediately.
Right now, this might be something like services related to prompt engineering.
Maybe… a prompt engineering analytics company that compiles an enormous dataset of metrics on how well engineered prompts perform on specific tasks for a particular type of model, then sells that data to companies.
I recently came across a startup building advanced ChatGPT-cheating detection software that takes the text-entry behavior of the user into a Google Doc into account along with analysis of the text itself.
I’d also note that the whole class of “good tech startup ideas” has been under optimization pressure for decades now, which means that these ideas, which may have accurately reflected where to find alpha when Paul Graham was making his first billion, may not apply any longer. By the time you have this much advice easy consensus available by proven experts, it might be the worst sort of advice to take.