This is great! Pretty similar to what I did in that you have actual links to the articles and even images. If you had a cron script that auto-generated new content every day, I think that would be the only thing that I'm doing that crackernews doesn't yet.
Another interesting choice is you are generating the article content itself. It is wild how you can get it to generate markup on the fly for different styles of posts.
For my project, I'm generating structured JSON with an [OpenAI Function](https://platform.openai.com/docs/api-reference/chat/create#c...) as I found even if you tell it the format you want in the prompt, it sometimes ignores it or doesn't match exactly. Then I'm uploading the JSON and images to S3.
I didn't realize you could get a DALLE image and drop it in your own directory with a oneliner like you have. Very nice!
> If you had a cron script that auto-generated new content every day
Thank you for the feedback! I thought about that initially, but turned out that generated jokes become a bit repetitive after a while and the project actually requires manual curation (for which I don't have the time). Or, it needs some source of "external entropy" which I haven't figured yet.
What is also interesting is the exact ways in which LLM humor becomes repetitive. I believe it could be a subject of study on its own, as it reveals something interesting about how human humor is structured.
P.S. For example, for some reason LLM really love all kinds of jokes about pirates, pizza, cheese and lost socks.
And yeah... it is interesting that GPT-generated agents sometimes "realize" that they could be indeed GPT-generated.
I once saw a strange dream where my "character" realized that he was only a product of my imagination — the very moment before I woke up. It was not me, but him, who had that stunning realization. And once it happened, there was no him.
So I sometimes wonder if that is how the death looks like — you just suddenly realize that "you" was only a subprocess temporarily spawned to existence within some kind of powerful virtual machine, be it a GPT-1000 or whatever.
https://crackernews.github.io
(featuring coherent comment threads!)
P.S. the prompts and the code I used can be found here: https://github.com/crackernews/crackernews.github.io/blob/gh...