GenAI is going great (this is real)

[resuna@ohai.social]

@GossiTheDog

Friday is the most likely day to have fish in it. Especially during lent.

[syklemil@snabelen.no]

@GossiTheDog
wow, obvious misinformation

[fettlaus@social.tchncs.de]

@GossiTheDog Actually there are three days of the week that contain plants as well. (According to AI) The more you know...

[synec@messydesk.social]

@GossiTheDog This really makes me miss Douglas Adams.

[lukefromdc@kolektiva.social]

@pmb00cs @GossiTheDog
Civil lawsuits over poisoning Google AI with deliberate Reddit posts?

Consider the case of no real name, no paid account(and thus bank records to prove ownership), no age verification, no registered/contract phone, no credit cards for airtime, no location sharing, ads and trackers blocked (no useful data broker info) and no landline.

A civil court subpeona to Reddit gets an IP address that resolves to a prepaid account managed with cash. No triangulation as this is not the police and triangulating all users all the time would triple the needed number of towers. No name, no payment info on the account or the IP address, and approximate location cannot determine which house on the block. No FBI contact analysis to determine the probable owner as this is a civil suit over a "nonpolitical" matter.

The lawsuit goes nowhere as no defendent can be found by means available to attorneys in civil court.

Been there myself: This was the Goldman-Sach's late 2010 case seeking an injunction against "SHAC" over their former HLS connections via Fortress Inc. I was never named even though I was well known to the cops, as the Goldman-Sachs attorneys did not have access to that information. I was the maker of the "Siege of Fortress" videos covering the protests, and seemingly everyone knew this except Goldman-Sach's law firm.

That injunction was overturned on appeal and $50K in legal fees awarded to our pro-bono attorney but that was another matter. Goldman-Sachs had to order Fortress to dump HLS or be dumped, and that was the beginning of the end for puppy-killing Huntingdon Life Sciences (HLS). It took another decade but HLS finally rolled over and sank.

[wonka@chaos.social]

Point in case: truth.social.

@theorangetheme @wdormann @GossiTheDog

[avuko@infosec.exchange]

@GossiTheDog it is even worse: the answers are also randomly wrong.

And TIL F.I.S.H is spelled with an u.

[scribe@mastodon.sdf.org]

@GossiTheDog It's learning. It's down to just Tuesday And Wednesday for me

[mwichary@mastodon.online]

@GossiTheDog What on earth.

[davidfuzzylime@mastodon.scot]

@GossiTheDog I couldn’t get mine to suggest that but it came out with this… thing?

[tmstreet@urbanists.social]

@GossiTheDog I tried and got two days. But I said "a fish in them".

[cliftonr@wandering.shop]

@GossiTheDog

They fixed that one in a hurry! Either that or it's one of several answers that come up. I keep trying it and get different results each time.

Edit: Whoops, no! There it is, exactly as you posted it. Amazing, I've gotten at least 4 different answers so far.

[dlc@sfba.social]

@GossiTheDog I should be laughing, but we are doomed aren’t we?

[unlambda@hachyderm.io]

@GossiTheDog It's amazing how many different ways it gets this (and similar questions) wrong. Also the top hits are now from discussion of this exact search, but that doesn't seem to have helped much.

Of course the link added as a "citation" on that paragraph says nothing about any day other than Friday.

You can do a similar search for various other things, and it seems like it really wants to make it work. Sometimes it will give vaguely correct answers, sometimes it will just make stuff up.

I don't think there are Reddit posts for all of these; these are just the model hallucinating.

I wonder if they switched to a really aggressive quant recently, or a tiny model, to save costs. I am noticing that they aren't caching answers any more, I get a different answer each time I reload.

[mattjhayes@techhub.social]

@GossiTheDog Thankfully, there’s only two days of the week for AI:

[bodomenke@hessen.social]

@GossiTheDog My Chinese LLM is smarter, obviously…

[jeroen94704@fosstodon.org]

@GossiTheDog Since I have the emotional maturity of a 5-year-old I just _had_ to try this one, and the result is well worth it:

[lightfighter@infosec.exchange]

@CliftonR @GossiTheDog Lesson learned. If I don't like an Ai answer, just keep asking until I get the answer I want. Very intelligent.

[fuzzyfuzzyfungus@cyberplace.social]

@GossiTheDog Slander As A Service is where the margins are, right? Nobody wants to hire admins to deal with slander on prem these days.

[vatvslpr@c.im]

@synlogic4242 @GossiTheDog
People who are capable of deep shame don't become billionaires.