Hi HN! It struck me that LLMs should be able to very competently satirise since failure modes like hallucination, lying etc. are actually benefits in satire. That opens up the possibility of more personalised / niche versions of staple satirical publications like The Onion & The Daily Mash.

We just released The Alium as a demo, it monitors AI news only, satirises the latest headlines using GPT-4 & auto-posts the resulting articles to https://www.thealium.com/ and https://twitter.com/alium_ai. Each satirical article links back to the 'real' news inspiring the post.

We found GPT-4 to be a significant improvement over 3.5 in this domain, with the model demonstrating a more robust sense of humor, creativity a reduced inclination towards repetition.

The code is Open Source: AI News API -> few shot GPT-4 text prompt for article generation -> image prompt -> DALLE-2 image gen -> moderation step (which needs to be improved since it can sometimes pick out headlines which make a questionable source of satire) -> generate & post markdown to a static Jekyll site hosted on Github Pages.

Hopefully this makes someone laugh or becomes a more entertaining way to consume AI news.