Post ATSqIzzYvxCAGthYBs by Texan_Reverend@kind.social
(DIR) More posts by Texan_Reverend@kind.social
(DIR) Post #ATSqIzzYvxCAGthYBs by Texan_Reverend@kind.social
2023-03-10T04:51:02Z
2 likes, 7 repeats
Discord is getting in on the "AI" trend with testing of new features utilizing user data scraping, potentially for training a large language model. To opt out of this, go to the Privacy & Safety section of your Settings. From there, toggle off both "Use data to improve Discord" and "Use data to customize my Discord experience."
(DIR) Post #ATSvQ7DAzPJ6ScHXXc by tyil@fedi.tyil.nl
2023-03-10T07:28:59.188Z
1 likes, 0 repeats
@Texan_Reverend@kind.social to opt outStop using #Discord.It'll immediately opt you out of all their other bullshit, present and future!
(DIR) Post #ATVVrpNCtG8puWPIeG by halcek@merveilles.town
2023-03-10T11:28:29Z
0 likes, 0 repeats
@Texan_Reverend To opt out of this, leave Discord. We do have @matrix after all for God’s sake.
(DIR) Post #ATVVrps32aoNSApvii by Jennifer_Pinkley@zirk.us
2023-03-10T12:12:44Z
1 likes, 0 repeats
@halcek @Texan_Reverend @matrix i've never heard of matrix. I was just telling someone how I need to start using Discord more because it's a better alternative to Facebook and now I see this. I guess I was wrong!
(DIR) Post #ATVVrqNx7yKf37lPRw by Pineywoozle@masto.ai
2023-03-10T13:11:04Z
0 likes, 0 repeats
@Jennifer_Pinkley @halcek @Texan_Reverend @matrix I wonder if those of us w/ good moral compass shouldn’t allow the data scraping so that at least some of the training comes from people who aren’t also just w male centered. I get the desire to opt out & not be part but it’s going to be there, it’s going to be a part of our lives. Shouldn’t we, instead of opting out… step up? Step up & help make it at least partly reflect our values #AI
(DIR) Post #ATVVrr1Ilfo91AApma by sarky@strangeobject.space
2023-03-10T18:14:30Z
0 likes, 0 repeats
@Pineywoozle @Jennifer_Pinkley @halcek @Texan_Reverend @matrix they don't want to find out what you want or reflect your values though. They want to find more effective ways to manipulate your behaviour, so no.
(DIR) Post #ATVVrrXuoPtaeJQscK by Pineywoozle@masto.ai
2023-03-10T18:18:06Z
0 likes, 0 repeats
@sarky @Jennifer_Pinkley @halcek @Texan_Reverend @matrix They’re going to develop the AI anyway and I still think that if it’s going to be developed if it’s not just the input of ugly people who want to manipulate other people you have a better shot at having it have a less negative affect
(DIR) Post #ATVVrs0H6ya44GhWoy by csepp@merveilles.town
2023-03-10T21:28:57Z
0 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend If you want to step up for better AI models, step up for one that isn't developed by a VC funded firm with a very loose grasp of morality. Or just step up for laws that keep them accountable.Also you have zero guarantee that they will use the whole dataset equally. For all you know, they are developing language models for trolling and would use you to evaluate how effective they are at pissing decent people off.
(DIR) Post #ATVVrsVTEzXBd1IRRg by Pineywoozle@masto.ai
2023-03-10T21:43:17Z
0 likes, 0 repeats
@csepp @sarky @Jennifer_Pinkley @halcek @Texan_Reverend I’d be a bad model. I rarely get pissed off. I tend to just block & move on. They’re training the AI by letting it scrape all the data & I still think that if everybody decent opts out the only training that AI is going to get is shitty & who they point it at will have a worse experience. I dealt w/ 1’000s of bots on #Twitter Some #FarL some #FarR you name it
(DIR) Post #ATVVrsyXUump5Atekq by Pineywoozle@masto.ai
2023-03-10T21:45:15Z
0 likes, 0 repeats
@csepp @sarky @Jennifer_Pinkley @halcek @Texan_Reverend I’m all for stepping up and getting me some kind of legal oversight on these toads.
(DIR) Post #ATVVrtRbkq2SXKUs40 by csepp@merveilles.town
2023-03-10T23:15:17Z
0 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend Do you think Palantir's impact on the world would be positive (or at least less negative) if it had the data of more decent people?
(DIR) Post #ATVVrtr8DwSHoURFqa by Pineywoozle@masto.ai
2023-03-10T23:38:31Z
0 likes, 0 repeats
@csepp @sarky @Jennifer_Pinkley @halcek @Texan_Reverend Even something as evil as a company connected to the creep Theil, I think would be less evil if they had to work to exclude the decent morality the AI they employ scraped up. Energy spent pushing it to ignore or bypass the decent morality it absorbed is less energy to fuck with people.
(DIR) Post #ATVVruMgKdgzOLCS1Y by csepp@merveilles.town
2023-03-10T23:40:10Z
0 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend That is sadly not how training a machine learning model works. It doesn't "absorb morality".
(DIR) Post #ATVVrupkaYwcqUnfKi by Pineywoozle@masto.ai
2023-03-10T23:59:19Z
0 likes, 0 repeats
@csepp @sarky @Jennifer_Pinkley @halcek @Texan_Reverend It absorbs reactions to prompts and spits them back those reactions are based on the morality of the person who’s reactions it scraped.
(DIR) Post #ATVVrvH2x4mMD9ZSsa by csepp@merveilles.town
2023-03-11T00:03:03Z
0 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend What you're talking about is predictive text, which is not the only kind of machine learning model by far.And the model could also be used to impersonate a decent speaker and use kind language to promote negative ideas. You know how your brain can instantly spot most spam messages as scams because of the terrible grammar and jumps in logic? Imagine a language model that makes that no longer possible.
(DIR) Post #ATVVrvrCmdhc1IULEu by csepp@merveilles.town
2023-03-11T00:12:07Z
0 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend By the way, I don't know why you think groups like white supremacists are going to opt in to the AI training crap or stay on Discord. There are Matrix servers for them too.
(DIR) Post #ATVVrwI9ATFlMr5rEW by Pineywoozle@masto.ai
2023-03-11T01:24:53Z
0 likes, 0 repeats
@csepp @sarky @Jennifer_Pinkley @halcek @Texan_Reverend Every encounter I’ve ever had w/ them makes it apparent they want their ideas to prevail. But I wasn’t just refering to that level of extreme belief. There’s enough misogyny & racism etc even in the general population. As for spotting AI. I was the first person in 20 yrs to spotted the syntax issues of a prof who’d been writing his quizzes in Latin & translating
(DIR) Post #ATVVrwpp9GBx3Iqkj2 by csepp@merveilles.town
2023-03-11T01:29:07Z
0 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend The point was not whether you personally can spot algorithmically generated text or not. Feeding language models when there is zero guarantee they won't be used for deception is a terrible idea. Like, just because you personally can dodge bullets wouldn't make it okay for you to hand someone a minigun.
(DIR) Post #ATVVrxSooHNr0F5tVQ by Pineywoozle@masto.ai
2023-03-11T01:30:59Z
0 likes, 0 repeats
@csepp @sarky @Jennifer_Pinkley @halcek @Texan_Reverend They are going to get fed regardless. The only choice we have is wether any of that data they are scraping comes from decent moral people.
(DIR) Post #ATVVry3KcWagpUB3Q0 by csepp@merveilles.town
2023-03-11T01:32:22Z
0 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend That is patently false.
(DIR) Post #ATVVryRRAtsC2FSIzY by Pineywoozle@masto.ai
2023-03-11T01:36:41Z
0 likes, 0 repeats
@csepp @sarky @Jennifer_Pinkley @halcek @Texan_Reverend So what is the choice? Obviously we worked to write laws against it but untill then opting out just means that the sample is minus decent moral people.
(DIR) Post #ATVVryo7oY1NAc4QM4 by csepp@merveilles.town
2023-03-11T01:39:49Z
0 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend No, the sample would just be smaller. You are under the misapprehension that a bigger or more representative sample would be used for good.Two words:Cambridge Analytica.
(DIR) Post #ATVVrzCEMvIsNNLfvc by Pineywoozle@masto.ai
2023-03-11T13:07:02Z
0 likes, 0 repeats
@csepp @sarky @Jennifer_Pinkley @halcek @Texan_Reverend I am under no misamprehension that there are not bad actors out there. You keep singling out those type of examples and missing that a huge amount of AI use will be “benign” except if it learns from ugly people a trace of that misogny & racism etc will leak into their presence & subtly poison the well of human discourse.
(DIR) Post #ATVVrzYZ1tATUdnVjs by csepp@merveilles.town
2023-03-11T13:12:00Z
1 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend The same sample can be used to build either. Like with blockchains, right now there are not many actually useful use cases for large language models. You are correct that a model used for good would need a representative dataset, but currently the majority of LMMs have no guarantee of what they will be used for and by whom, whereas there are numerous malicious use cases.
(DIR) Post #ATVVs0IIHrkPmTC21A by csepp@merveilles.town
2023-03-11T01:42:26Z
0 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend A few further words though, just in case:Google is already ratting people off in the US when they look up abortion options.The FBI is purchasing location data.I'd rather these agencies did *not* have any data about decent people.
(DIR) Post #ATVVs1SbwzavOl1mjo by csepp@merveilles.town
2023-03-11T13:12:39Z
1 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend Add to that the massive carbon footprint and I see little reason for anyone to contribute to these datasets.
(DIR) Post #ATVVs2rSk54Fk7f972 by csepp@merveilles.town
2023-03-11T01:47:14Z
0 likes, 0 repeats
@Pineywoozle @sarky @Jennifer_Pinkley @halcek @Texan_Reverend And again: language models trained on the language good communities use would be most useful when targeting those communities.Think of a scammer wearing a disguise.In the scenario that you suggest, the scammer can disguise themself as an aid worker, whereas in the preferable scenario they can only disguise themself as a nazi.