Georeactor Blog

Chronically online people may be familiar with the term 'BlueAnon' to mock conspiracies at the opposite political polarity of 'QAnon'. The two don't have equivalent followings, but 🤷‍
In the pushback to AI getting incorporated into everything, I notice a pattern which I want to label "LuddAnon", defined in three points:

• Major political events are best explained as someone blindly copying off an AI / LLM; any other aspect is almost irrelevant.
• A select few people know how to write prompts or find typos and formatting and 'prove' that content was AI-generated.
• When news breaks on social media, before any real reporting, a flash conspiracy will develop linking it to AI

Notes

• AI use can lead to mistakes, just give it the same scrutiny as claims that someone copied from Wikipedia or used Google Translate.
• When lawyers and professors have made mistakes with AI, these are usually proven with hallucinated sources, not by testing if LLMs can be prompted to generate similar text or ideas.
  Though ChatGPT can be more likely to use particular dashes, words, and lists, the discourse around AI in executive orders was awfully weak or contradictory (typos)
• Once the tariff formula was discovered, the claims evolved to AI making the list of countries and territories (?) or the formula, or stealing the formula from someone who is a current Trump advisor. Maybe people want to dissociate the tariff numbers from 'real' political beliefs or calculations, but tariffs were part of the 2024 campaign, staff had to put real trade numbers into spreadsheets and posters, and did people forget the first term ? The Washington Post recently reported on the real process where staff created multiple possible formulas and decided on one.
• One reason to push back on this is that AI detection based on vibes or shady software is unfair to students. Also if you care about the political situation and misinformation, have sources for stuff.
• The South Korea martial law case found that a bodyguard asked AI about martial law, but this is more about establishing a timeline; the president had discussed martial law for months
• AI gets blamed for the federal government canceling grants, firing and rehiring, and hiding web pages about the Enola Gay and Jackie Robinson. I don't know enough about DOGE's email inbox or AI claims to say whether or not they used an LLM, a script with a word list, or something even less advanced (for example, this find and replace)