Show HN: I built a website that generates fake users for an idea (notionsmith.ai)
Technically it doesn't have to be a product: it works with book ideas for example.
It's based on GPT 3.5 and uses chain of thought to model a target market, personalities, etc. In the process of building this out I got a really great model of how to tease creativity out of LLMs.
Once you generate a user you can also chat with them, and they'll imagine their use of whatever you're building in as much detail as possible.
You can try things like:
- "a product that only exists in a state of non-existence and disintegrates into nothingness when perceived"
- "a wet sock that is always wet"
- "a mirror of Hacker News translated into pig latin"
- "a web browser that replaces the term 'AI' with a poop emoji"
-
I get a lot of feedback asking how you can trust what the model says, but I've theorized that what LLMs perceive a product as is what most web users will when they first discover what you're building.
If it doesn't align with what you're actually building, it's a sign you'll need to address a certain degree of misunderstanding when people first hear about what you're building.
I'm building a set of tools similar to this one: the next one I have in mind is one that helps you name things. Rather than coming up with a name, it teaches you how to name something for a given domain ("how to name your YC focused SaaS product that's targeted at doctors" for example)
2 comments
[ 4.7 ms ] story [ 18.6 ms ] threadEdit: if you can also share a site or something to know when you publish the other tool, that would be great!
I've been planning an improvement to the randomness by charting out the main themes across feedback that should be represented so there's less overlap.
Even getting random names was a little tedious, I ended up using an RNG to get two letters, and telling the LLM it needed to use those two letters in the name.
And I'll be posting these tools on product hunt going forward https://www.producthunt.com/@spop