punchlines.ai is an AI joke generation tool built on top of a large language model (GPT-3.5 Turbo). It was fine-tuned on thousands of late night comedy monologue jokes (https://github.com/brendansudol/joke-gpt).

web app is open-source, too: https://github.com/brendansudol/punchlines-ai

Not a dumb question! This is called fine-tuning; here's more info about it: https://beta.openai.com/docs/guides/fine-tuning

Process is very straightforward -- all the code (data cleaning, uploading, tracking can be found here: https://github.com/brendansudol/joke-gpt/blob/main/train.ipy...)

Here's more information on the fine-tuning process: https://beta.openai.com/docs/guides/fine-tuning

In terms of where it's incorporated in the code, the fine-tuned model id is stored in an environmental variable called `process.env.OPENAI_MODEL_ID` which then gets passed to the completion endpoint

thanks a lot! re: fine-tuning - it was about $30 using the most expensive model (davinci)

web app code on github: https://github.com/brendansudol/smartful

How do you plan to fill the database? Is it just one big database and you will cycle it or you plan to add to it every day?

I've seeded it with 4 months of data (stored statically in json files for simplicity for now), and have a script to add data for additional days, but the plan is to migrate over to a db (curious to try supabase) so that users can save / like individual entries.

SQLite might be a good choice for a project like this.

So all your sources for all categories are automated? My main worry was if you have to hunt for content manually - that might not be sustainable.

This would also be great for Telegram. You could easily forward daily content into a channel. Or even create a bot that allows people to subscribe only to some categories.

I made this with Django on the backend, and React on the frontend, and it's hosted on Heroku.

Code is open-sourced: https://github.com/brendansudol/jeopractice.com

Was just poking around and came across this one: http://www.jeopractice.com/?id=6177&x=3&y=2

No picture is displayed.

Thanks Tyler! Good catch; that does make this question a bit harder to answer... :) Will try to fix

I made a simple DIY ranking system using the same College Scorecard data last week. Here you go: http://bsudol.com/1PUPohk

You can select which variables are important and how much, and it generates a top/bottom 50 list.

This was inspired by this article from NPR (http://www.npr.org/sections/ed/2015/09/21/441417608/the-new-...)

Submitted: https://news.ycombinator.com/item?id=10480851

This is awesome! I'm happy the college score card data is being used for great project like this.

This is great. As a next step, perhaps make the sliders more mobile friendly?

Hey @thekevan: thanks for the detailed comment. I appreciate the suggestions (more details, bigger pictures, something to convey authority) and your interpretations on the minimalistic design. And you're right, you are definitely free to just browse around on Amazon for cool stuff -- the idea here is that if need some inspiration or are not sure what you want, this could be a nice starting place of pretty cool, well designed products are already laid out for you. I should try to communicate this value better - thank you for the feedback!

that's up there on my list (and a nice excuse to figure out to how to rank friends by closeness) -- thanks for the nudge!

thank you! and good call -- i def will add that in.