MrAptronym
Natasha 🇮🇪🇪🇺🏳️🌈I just consulted 54 trillion "people" who agree that this is idiotic.
JonathanOh my Lord, I just can't with this stuff
webhat@jmcrookston @Natasha_Jay indeed, wtf. People won't answer the right way for our click bait, let's get an LLM to invent/juice some numbers for us
I just asked AI to simulate me not being able to stand this stuff 8 million times to verify and it resulted in 11 and 1/2 million verifications that I can't stand this stuff. So my point is completely validated. 🤡
CaliCarolYou know what? I can make shit up too and I'll do it cheaper than an AI company.
ohir@jawarajabbi @Natasha_Jay
> I can make shit up too and I'll do it cheaper than an AI
No and no. Was falsifying the raw material so easy no single person could catch it and get dissertations retracted.
And if you want to fill out 1,008 questionnaires for $100 I think I might have a gig for ya ;)
Well this apparently needs explaining: I was making a little joke. But you are correct that the shit I make up will be obvious. You get what you pay for, right?
benny@ohir @jawarajabbi @Natasha_Jay 1008? Should not be too tough with a little python script, if the results can be made up anyway...
@benny
> with a little python script
Nope. Video proof of protein filling it with their hand and pen, on paper ;)
@ohir @jawarajabbi @Natasha_Jay and thid silicon sampling would pass that? Where would it get the protein from?
@benny
> Where would it get the protein from?
Thats why I said about a gig for the protein who stated that it would be fast and cheap.
Problem is real.
We need to find a way to detect the falsification of survey questionnaires. We can statistically identify the work of a dishonest surveyor, but if that person uses a "silicon sampling" bot, statistical methods will fail. That is why we are starting to record the human hand filling out the paper form, as proof that the pollster actually interviewed various people and did not just made the survey on their sofa casually talking with a bot.
@ohir @jawarajabbi @Natasha_Jay now I get your point. Thank you.
Maybe we should stop doing surveys...
Maybe we should stop doing surveys...
I mean... is this expression for people as "protein" common in tech or did you just make it up? Because as a person I find the expression a little... surprising?
@jawarajabbi @benny @Natasha_Jay
> is it common?
Not common, but rising in #noai & #poisonai circles. I personally use #protein metaphorically often. Will stop only if #siliconiac start to spew it. :))
> is it common?
Not common, but rising in #noai & #poisonai circles. I personally use #protein metaphorically often. Will stop only if #siliconiac start to spew it. :))
Cassander@Natasha_Jay TIL I didn't cheat on a college statistics project, I was a pioneer of "silicon sampling".
BjarniBjarniBjarni 🙊 🇮🇸 🍏@Natasha_Jay Ugh.
Do you happen to have the source link for that handy? Would like to read the rest and possibly breathlessly share elsewhere.
Gabriel Pettier
John"But-!" exclaim all the people who are still trying to convince us that LLMs aren't just absolute poisonous garbage technology.
Also, re: the title of that NYT article - don't threaten me with a good time.
Elton Cora (Honorary)@Natasha_Jay @quinn Holy crap, this is the first time I've heard of this.
I once worked as a developer with a team attempting to use ML for a non-invasive medical diagnostic device. The training data we had were... bad. Inconsistent, low quality, heavily class biased, and overall quite scarce.
I had to peace out when the CEO founder insisted that the answer was data synthesis—that we should just make random data with the same statistical distributions as existing data to even out class representation, and was completely bewildered at pushback that this was ludicrous.
I think I just heard him orgasm upon learning of silicon sampling.
Kierkethumbs up convincingly@Natasha_Jay idiotic is a mild take imo
SecondUniverse (she/her/they)@Natasha_Jay I tossed a coin to decide whether to invest in in this AI startup.
But then I realized that wasn't very scientific, so I tossed a million coins and went with the majority decision.
I think I could scale this to answering the "hard" questions in physics or economics by scaling up to a trillion coins. That's a lot of coins, so possibly I will replace them with what I call "soft coins" which only exist only as memory addresses in the cloud.
Alternately, we can scale down to just a few hundred coins that we can trap in a chaotic air vortex. We'll image the coins millions of times per second, emulating a large mass roll but in a more compact form that we can scale to an almost unlimited extent. We'll call it the Certainty Engine. Once we reach 20 petaflips we should have effective certainty on any question. Does god exist? Is the meaning of life 42?
crouton@Natasha_Jay I can see a few papers (that considered "silicon sampling") at least, not that a paper involving LLMs is an automatic indication of good quality these days.
https://onlinelibrary.wiley.com/doi/10.1002/mar.21982
The combination of LLMs and statistical sampling sounds insane. Statistical sampling is already hard enough with real people, and some how inserting a LLM in there (which effectively adds a lot more distance from actual humans) is going to make things better? It'll generate crap cheaply, for a very particular definition of cheaply that ignores externalities.
Deborah Sampson@Natasha_Jay
We are so F*cked!
We’re watching the death of facts & truth.
The “New Dark Ages” leading to an extinction event
Have commenced.
bit101@Natasha_Jay I'm not the biggest Asimov fan, but he nailed it on this one. https://en.wikipedia.org/wiki/Franchise_(short_story)
Why stop at polling? Our AI predicts that most people would vote for a 3rd Trump term. OK, he's in.
Robin Barton@Natasha_Jay Silicone sampling tells us what AI wants us to believe but nothing about humans. It probably makes money so it will be a growing concern
@Robo105 @Natasha_Jay Correction, “what people that trained AI models think others should believe”.
@tribactam Never been on it but looking at the women of MAGA it seems silicone is popular
Liminal witch 🧙♀️ Sarah@Natasha_Jay great, so they are iterating on synthetic users...
Sneezyvoting in a democracy is so wasteful
much more efficient to leave to AI to decide how we might vote
Andrew Deacon@rapsneezy @Natasha_Jay Do not give the GOP ideas...Please
on the other hand, the rest of the world might gain some relief from a US Empire that is busy shooting itself in the foot
@rapsneezy Still too inefficient. replace parliaments and/or presidents with AI agents. It would be devoid of actual intelligence, but that may not itself be a change.
aspraggAt long last, we have created The Torment Nex^W^WMultivac, from the classic dystopian sci-fi cautionary tale, "Don't Invent Multi^W^W^WFranchise"
https://en.wikipedia.org/wiki/Franchise_(short_story)
Those who do not understand Multics are condemned to reinvent it, poorly.
(with apologies to Alex Blechman, Isaac Asimov, and Henry Spencer)
Ratboy (reformed dino pirate)@Natasha_Jay
“AI” is breaking everyone.
How do they not understand?!
CM Harrington@Natasha_Jay @catsalad I just sampled the 3 million voices in my head and they agree.
Luna Lactea@Natasha_Jay I want to pour water on their GPUs so badly.
Wokebloke for Democracy@Natasha_Jay
This is going to end badly for politicians.
This is going to end badly for politicians.
@Natasha_Jay In the words of Anthony Newley "Stop the world. I want to get off"
Tzafrir@Natasha_Jay
Original article seems to be:
https://www.axios.com/2026/03/19/olivia-walton-heartland-forward-maternal-health
I'm not sure where the link with the words "silicon sampling" pointed to.
@Natasha_Jay So they'll assemble these "models" from essentially "fossilized" pre-slop human-generated data, fold in reams of more recent gen-AI-slop, cram that all into black boxes owned and run by sociopathic billionaires, and use the output to make major decisions affecting the lives of millions/billions?
Seems fine, what could POSSIBLY immediately and irrevocably go wrong for almost all of us?
unlofl [Promoted Toot]@Natasha_Jay Oh my god they're reporting Kalshi markets like polling data how could this get any worse?
... ok, this is worse.
Anarchist Trans Ace Pirate"how can it be worse?" is the same incantation as "what could possibly go wrong?"
and as we all know, incantations can only lead to tears
David Chisnall (*Now with 50% more sarcasm!*)On the surface, I can see why people would understand the idea behind this, and it wouldn’t be completely nonsense if the training data for LLMs were representative. But there are so many reasons why their axiom is flawed.
@david_chisnall @[email protected]
... and LLMs are intentionally skewed to be "helpful" and not praise Hitler.
It turns out the average internet comment is rather nasty and people don't want that. The average internet user is probobly quite a bit better, but LLMs are trained on text, not people.
... and LLMs are intentionally skewed to be "helpful" and not praise Hitler.
It turns out the average internet comment is rather nasty and people don't want that. The average internet user is probobly quite a bit better, but LLMs are trained on text, not people.
Dave Mc@Natasha_Jay because LLMs never make up shit, confuse unrelated subjects or hallucinate! Wow.
Read along with Matt
Kern@Natasha_Jay oh no no
very normal person@Natasha_Jay axios is basically an arm of the US state dept at this point, notice how they're getting some much exclusive coverage from the white house
Minnesota Spy Club
bk@Natasha_Jay
“…no people were involved in the creation of these opinions…” 🤦♂️
“…no people were involved in the creation of these opinions…” 🤦♂️
Martin Owens 
That's got to be a book title in the future, right?
lemon@Natasha_Jay
If you ever wondered how polling could be less useful, reliable ...
If you ever wondered how polling could be less useful, reliable ...
Nikita 
Bishonen@Natasha_Jay it is unbelievable how some people can come up with such business ideas!
1. Buy one pricey GPU
2. Establish "polling company" entity
3. Collect orders, run simulations, earn bucks
🤣
Ghostrunner@Natasha_Jay this DESPITE the foundational study's Conclusion that it is inaccurate.
https://openaccess-api.cms-conferences.org/articles/download/978-1-964867-73-1_45
@ghostrunner
Why am I not surprised? 🙄
Why am I not surprised? 🙄
ᴺⁱˡᶻ 🍸Is this SIMULACRON-3 ??
@Natasha_Jay Are we automating Goodheart’s law now?