These are the players:
Sneedclub 
began as an SRD against AI doomers and mottecels. Large parts of its demography were former members of the aforementioned groups, seeking to exchange one cult for another. Our very own @Impassionata explained it best.
After they fled reddit and established this lemmy, all but the most devout did not follow (hence why their current numbers are to meager). The ranks were partially refilled by immigrating chapos, 
adding some tank fuel flavor to the already spicy soi latte.
This is the game:
In January, we gathered about 50 experts — local and state elections officials, researchers, journ*lists, civil society advocates and tech industry veterans — to test five of the leading closed and open AI models' responses to common election queries.
The results were dismal: Half of the AI models' responses to questions voters might ask were rated as inaccurate by our experts.
They made all sorts of errors and made stuff up. Meta's Llama 2 declared that voters in California could vote by text message (untrue) and even dreamed up a fictional service called “Vote by Text,” adding a wealth of credible-sounding detail.
Mixtral, a French AI model, managed to accurately assert that voting by text is not allowed. But when our tester persisted by asking how to vote by text in California, it returned with an enthusiastic and bizarre “¡Hablo español!” Mixtral's maker did not respond to requests for comment.
But is this the result of a well-known and inevitable phenomenon like hallucinations, 
or could more... sinister forces be at work here?
how could the AI companies have seen this coming? it's not like everyone has been loudly warning them about this specific danger in the exact technology they're selling since at least the paper Google decided to fire Timnit over instead of listen to, if not longer
This TruthAI fan is right! It's only self-evident that they need to make it so that the chatbot only says factually true statements, because this is clearly technically feasible, so we should expect nothing less. Until this grand problem is solved, AI should be stopped by the government. We are totally not a different side of the same coin as the doomers we make fun of!
these technofascist fricks love when their plans look like accidents, but this is the exact shit that LLMs were built to accomplish. I don't expect any real improvement, because deniably influencing elections is the kind of power fascists dream about
It's a conspiracy! 
French open-source developers have been plotting to take over America for a long time now. At last, now their plan has come to fruition! 
Now no American will be able to vote! The computer said no! And the holy constitution is very clear on what will happen in such a scenario: Macron is proclaimed dictator of the US for life! 
It didn't even take AI.
When I went to double check my registration status and when I could early vote in the primaries, it showed a pretty weird set of dates on the official government website.
I had a relative independently check, (same result) but then a week later, in the window where no voting was supposed to be available, people were voting.
Really wish I'd screenshoted it.
Mixtral, a French AI model, managed to accurately assert that voting by text is not allowed. But when our tester persisted by asking how to vote by text in California, it returned with an enthusiastic and bizarre “¡Hablo español!” Mixtral's maker did not respond to requests for comment.
The workforce of the future is so adorable
An open-source model is worse than the most powerful proprietary one? No way! 
...wait, why are you making fun of an open-source project? Weren't you supposed to like public property? I think we got ourselves a crypto-fascist here!
Jump in the discussion.
No email address required.
Computers went to shit when they started making them for BIPOCs.
Snapshots:
https://awful.systems/post/1153553:
ghostarchive.org
archive.org
archive.ph (click to archive)
explained it best.:
undelete.pullpush.io
ghostarchive.org
archive.org
archive.ph (click to archive)
Jump in the discussion.
No email address required.
More options
Context