DW flist

Having had a nice day's sleep I woke up thinking that Legends of Tomorrow, characters departing from January 2016, encodes in its treatment of Gideon a specific set of assumptions about AI. Ones that do not quite hold in 2025.

Basically the characters treat Gideon like a voice interface, in early seasons, not an independent person with opinions.
And then they treat what Gideon says as Fact, especially if it has pictures.

Which is interesting because Rip Hunter is *clearly* manipulating them by showing them the pictures and only once they're having feelings telling them that's a simulation.
Simulation is another word for guess, even if Gideon does put a percentage next to it.

If you start now? You'd get a set of people who still believe Gideon because look, pictures and a percent from a computer, true facts. But you'd get other people who are aware that deepfake graphics do not make your report more reliable, and who would call the AI out for hallucinating.

The show does interrogate its own starting position but by making Gideon unquestionably an independent person and then making it a right mess that people did not treat her that way. Making it just as weird that Rip reprogrammed Gideon as that the Time Masters reprogrammed Mick. And that's always a good set of questions, are we treating this person as a thing.

Plus there's the gigo problem or the deliberately being lied to encoded in what the Time Masters fed their computer to get Rip hopping.

But asking questions about error rates and reliability were not the first thought of anyone in this story, and that's super weird, coming from a context where I've seen reports that the LLM currently being called AI have an error rate where between 80% and *all* the answers in that study had errors. No I don't have a source to cite, it was social media screenshots, so obviously I shouldn't believe it. I could google for it. Wonder what google says about reliability of AI... well it's not letting me copy paste the answer, that's interesting, but it says Not Fully Reliable and links to three items with the most recent being May 2025... and one having answers from 2023, how is that any use this year? Google reckons the majority of answers are accurate, where it links to things that say 31% of answers were errors. But then I rephrased and asked for a percent, "what percent of AI answers are correct" and it said AI answers were incorrect between 60% and 94% of the time. It is talking about studies without linking them. The links it does go to are news summaries not studies, and they're from March 2025. Oh, here's a study I probably saw because BBC, https://www.bbc.co.uk/mediacentre/2025/new-ebu-research-ai-assistants-news-content , news summarised by GenAI wrong 45% of the time across all AI, some AI as bad as 76% wrong. So a few minutes with the google AI get me two contradictory poorly sourced answers, where the sources have different numbers than the AI summary, which is what usually happens when I read the google AI bit. And my memory is also poorly sourced! And skimming the links proffered tends to show a lot of podcasts and video content which is hard to check and... ugh. (This is me demonstrating very sloppy research on my part and how AI isn't helping. I feel embarrassed now. I am in fact capable of poking several pages of links, it's just the show always takes the first AI answer, and lo, it is messy.)

I do not trust AI because no matter how precisely defined the question is about stuff I already know, ie Pathfinder first edition, I always get an answer that is muddled with irrelevant sources and some degree of just plain wrong, ie about DnD 5e and using fandom wiki about computer games instead. Yes the terms are very similar but I have yet to be able to narrow its attention down to just Archive of Nethys or Paizo sources *and that used to be easy*. It used to be quicker to google search than to search on AoN or PFSRD or the paizo boards. Now it is an error prone mess that keeps trying to send you to DnDBeyond even though that is a completely different game. ... different edition branched off the same source yes I can see where the confusion is coming from but the confusion is *new* and *annoying*.

So: Legends Gideon. No one spends time questioning her sources or her error rate or how deepfakes are actually helping this situation or how many *other* scenarios for a future she finds equally plausible right now. And that stuff is absolutely essential. It's the core of how Rip gets led around by his training. And it's not just Legends where it matters, there's a whole thread of the Flash that goes back to the historical record and how it (mis)represents him, the reliability of sources and how history was written by his wife, without her admitting the connection. ... this will turn into my Eobard Thawne Has A Point Actually speech, and I don't think I watched the last season of Flash yet so I'll leave be. But really. Computer 'intelligence' and the authoritative misrepresentation of mediated reality: A Thread.

Pretty important in the real world as well.