Aurora Borealis
banner
queerkigaard.bsky.social
Aurora Borealis
@queerkigaard.bsky.social
Night light
The first thing the audience learns is that the Empire DESPERATELY needed this one specific engineer they knew hated them to work on the death star. Sure he ends up destroying it but he also could have *not built it to begin with* and the result would have been the same. But that's not the message
November 24, 2025 at 10:21 PM
This type of self-defeating hope and love COULD be the stage for a poignant tragedy, but that's not what Rogue One is. At best it's narrative verisimilitude in a series defined by every major character in the universe being a secret blood relative of some other powerful figure.
November 24, 2025 at 10:21 PM
And as always parasite free!
November 24, 2025 at 11:45 AM
Is there—is there balm in Wuhan?—tell me, stand and deliver!”
Quoth the Raven, “Nate Silver.”
November 24, 2025 at 9:26 AM
Till the dirges of his Hope that melancholy chill shiver,
Of ‘Nate—Nate Silver’.”
November 24, 2025 at 9:24 AM
On the morrow he will leave me, as an arrow from a quiver.”
Then the bird said, “Nate Silver.”
November 24, 2025 at 9:23 AM
Shrek
November 24, 2025 at 5:54 AM
I can't believe this is how I'm finding out that Rebirth Angra reunited
November 24, 2025 at 3:18 AM
I think we're going to see more and more awareness from these companies on how they can steer LLM bias as time goes on, and while there isn't proof of that intentional bias yet, social media's scandals have proven AI companies are unlikely to suffer meaningfully if they're caught thumbing the scale
November 24, 2025 at 3:15 AM
I guess a better example of my fears would be that socially liberal engineers accidentally took part in building the alt-right pipeline through the youtube algorithm.
November 24, 2025 at 3:08 AM
I'm saying that once that crowd gets bored and moves onto something else, will probably be when many of the situations they described will start to become more realistic problems and so I'm pushing back against the idea that these things *can't* happen. We should be thinking about these problems.
November 24, 2025 at 2:54 AM
Once enshittification hits, it will be easier for companies to hide specific instructions on how to act. It will also get easier with time due to emerging tech.

I understand you're saying a lot of the fearmongering related to LLMs right now is uninformed and frustrating.
November 24, 2025 at 2:54 AM
Or they do what every freemium online product ever does once it secures a user base, make the free version useless and then ban anyone who tries to jailbreak it
November 24, 2025 at 2:42 AM
That depends who's more responsible for the mess we're in, Donald Trump or Roger Ailes?

Yes I think conservatives are capable of being subtly manipulative, I think thats a conservative existing in the world thinking theyre just being normal and mistreating anyone who doesnt fit their idea of normal
November 24, 2025 at 2:39 AM
Yeah and I think theyre just our generations Area 51 conspiracy theorists. I dont bother with persuading anyone on this because I think the most fanatical haters want something to hate more than they care about any specifics
November 24, 2025 at 2:37 AM
Are we going to start using world models that scrape the internet for info to bootstrap "memory" into these models so they can have a jury rigged database of true facts and false statements instead of searching the internet directly? That opens more doors for subtle fuckery.
November 24, 2025 at 2:35 AM
And my response is that while yes, Claude, ChatGPT and Gemini all seem clear of those for now, part of your expressed sense of safety is "it would have to be a big conspiracy" and my worry is that emerging technology is pointing to smaller and smaller methods
November 24, 2025 at 2:33 AM
which part? The first part, no, it seems like thats not happening yet. The API part is inevitable. APIs never last.
November 24, 2025 at 2:30 AM
I am with you 100% here that its a bad thing that people in our communities are being scared off of understanding the technology. But I think youre imagining a very specific type of manipulation and theres other methods that bias can be transferred from an executive to an audience through an LLM.
November 24, 2025 at 2:29 AM
Which is what I said up front though about multimodal agents which will eventually revoke API access. Creating a model that could operate in that environment is a lot more plausible that training an entire agent to adopt your POV. If you just need an agent that avoids some topics that's doable now
November 24, 2025 at 2:25 AM
"Subtle manipulation" =/= consistent affirmative statements that an AI has to say. Its a lot broader than that.
November 24, 2025 at 2:24 AM
Is it going to say "trans people don't deserve rights?" No. It is going to be sure to always adopt a very hands off description of any crime where Trans people are a victim, in the same way the New York Times does? Thats a lot easier to add into the mix
November 24, 2025 at 2:23 AM