another mildly silly article about AI disinformation, but at least this one gives you something to think about

https://www.washingtonpost.com/technology/2025/04/17/llm-poisoning-grooming-chatbots-russia

(or https://archive.ph/seDGw if you don’t have a washington post subscription)

some ones to grow on

1. we have been hearing in pop culture for the past fifteen (twenty? twenty-five? more?) years about the problems of believing what you read on the internet. in the final analysis, what the AI industry has done for us is create a person-shaped information actor who has read anything that can be read but has never read anything other than information available on the internet, actively holds itself out as an authority on everything, and will not stop talking everyone’s ear off about literally anything at literally any time.
2. first of all, it is hilarious that anyone ever thought this would end in anything other than catastrophic results for the concept of consensus reality.
3. second of all, it should be obvious — not just to experts in the field, but to the average person on the street — that the problem of ensuring that this thing doesn’t “believe” disinformation is an even harder variant of the problem of ensuring that humans don’t spread disinformation manually, the old fashioned way, with hands on a keyboard and eyes on a monitor: if we can’t reliably get our neighbors not to claim that the earth is flat with the benefit of tens of thousands of years of human rhetoric and logic, how do we expect to get an information actor made out of month-old techniques in exascale linear algebra to agree that the earth is round from first principles without resorting to the brute-force instrument of “this information is axiomatically true. disregard anything that conflicts with it?” (if we can even get that to work; as this article points out, our record is mixed so far.)
4. third of all, if we do resort to that brute-force instrument — and arguably even if we don’t — it should be obvious that the question of policing what is True to this immensely powerful information broadcasting platform deserves better than to be a completely uncared-for third-order work product of however many specialist software engineers pulling down a quarter-million dollars a year; it is a fundamentally political question that should be decided with the messy work of democracy.
5. if it’s not obvious to you, please take as evidence the fact that:
   • several tech companies making AI have, over the past year, gotten upset about the fact that their AI occasionally expresses political views to the left of Mussolini;
   • none of them stop to ask questions about whether they, as mere human information actors, have gotten things wrong in the face of overwhelming AI reasoning power and evidence;
   • they instead take this as evidence of irreducible “bias” stemming from bad input that needs to be compensated for.
   • there’s no review process, they just do it.
   • even if they make all sorts of big noise about how open-weight models are a gift to humanity or whatever.
   • (obviously, as we can all agree, it is good and proper that the reference point for the range of political views accessible to everyone on earth should be what Facebook thinks is proper as a corporation.)
6. (viewed through this lens, there’s a new, malevolent reading you can apply to the obscene lengths that tech companies are going to in order to push AI on us: after the First Vibe Shift of approximately 2014-2021 led to a rising anti-corporatism culminating in the 2020 uprisings, they envision AI as a vector by which they can go down to “ring 0” of politics. if they can make us dependent enough on “thinking machines” that they exert direct control over, they can not only make certain ideas that are inconvenient or unpleasant for them effectively disappear, but they can also sell the ability to “unthink ideas” to other powerful entities.)
7. I love the way the AI field feels compelled to invent new, cutting-edge-sounding phrases like “generative engine optimization” and “LLM grooming” to make these problems sound new rather than absurd hyperintensifications of the same problem that has existed for god knows how long.