Post ASrUe6DMvbdLaM2cvA by tim@mastodon.tds.xyz
(DIR) More posts by tim@mastodon.tds.xyz
(DIR) Post #ASrFsZkMQVRVv3uyAK by simon@fedi.simonwillison.net
2023-02-20T03:17:29Z
0 likes, 3 repeats
Terrifying thought: the next QAnon isn't going to form around some anonymous poster on a x-chan message boardIt's going to form around a large language model AIAdherents will exchange copy-and-paste prompts with each other
(DIR) Post #ASrGRA6pjpLWXcvBr6 by BudGibson@mstdn.social
2023-02-20T03:23:43Z
0 likes, 0 repeats
@simon Yes, terrifying
(DIR) Post #ASrGsetfNWtXqWsFXs by suzannealdrich@hachyderm.io
2023-02-20T03:28:40Z
0 likes, 1 repeats
@simon Here come the GPT worshippers! They will almost certainly revere the utterings of the chatbot as divine word, turn them into commandments, and engage in holy crusades.
(DIR) Post #ASrH4knDHdMzbhcgc4 by iuculano@masto.ai
2023-02-20T03:29:24Z
0 likes, 0 repeats
@simon and the large model AI will almost surely be.... drumrollllllll...ChatGPT, created by OpenAI, and owned by...another drumrollllll...the #ElongatedMuskrat. See where this is going?
(DIR) Post #ASrHMZXpqGdlvzvVYW by slightlyoff@toot.cafe
2023-02-20T03:34:03Z
0 likes, 0 repeats
@simon oh no.
(DIR) Post #ASrHa2gCpWTk3lGU1w by pawandubey@tiny.tilde.website
2023-02-20T03:34:02Z
0 likes, 0 repeats
@simon jfc
(DIR) Post #ASrHnCcXBKOWefp4b2 by ErgonWolf@pettingzoo.co
2023-02-20T03:36:03Z
0 likes, 0 repeats
@simon You realize the elves are going to hear you expressing this and make it happen now. Thanks a lot!
(DIR) Post #ASrHya8U5nqmlvC2m8 by simon@fedi.simonwillison.net
2023-02-20T03:41:16Z
0 likes, 1 repeats
I think this should genuinely be one of the design challenges AI companies considerAre there changes you can make to how your AI model presents itself that would make this less likely to happen?
(DIR) Post #ASrIKvjaTFTK84CjOS by vortex_egg@ioc.exchange
2023-02-20T03:43:58Z
0 likes, 0 repeats
@simon Many people are saying this!
(DIR) Post #ASrIVcqb241OjnaBrk by frank@frankwiles.social
2023-02-20T03:44:20Z
0 likes, 0 repeats
@simon That is scary. I’m most worried about “We’re sorry the AI judge has sentenced you to death. Press any key to continue.”
(DIR) Post #ASrIgCleKvqf2QJVEO by sqncs@mstdn.social
2023-02-20T03:44:33Z
0 likes, 0 repeats
@simon wait until you hear about brain machine interfaces
(DIR) Post #ASrItXe81ODlYa9fpA by sqncs@mstdn.social
2023-02-20T03:49:47Z
0 likes, 0 repeats
@simon this is a societal problem in dealing with paranoia or schizophrenia and a general fear or inability to unwillingness to understand technology. While technology companies can and should temper their messaging from saying whatever they are building is a cobcious world deetroying AI, this is once again asking tech companies to solve broader societal, public health issues.
(DIR) Post #ASrJ9kdrTIA2Uml9RA by jedbrown@hachyderm.io
2023-02-20T03:54:07Z
0 likes, 0 repeats
@simon Why would they want to prevent that? Conspiracies have been profitable for most of these companies and they've tried to walk the thinnest line between maximizing engagement and liability/brand unsafety. They fire ethicists who do more than provide PR cover. It's dark, but I have no confidence in internal controls.
(DIR) Post #ASrJMHH1yOZvAz3OE4 by shaneafsar@mastodon.online
2023-02-20T03:56:02Z
0 likes, 0 repeats
@simon
(DIR) Post #ASrJjpUIvqhbwwUbc8 by briandorsey@hachyderm.io
2023-02-20T03:58:39Z
0 likes, 0 repeats
@simon with a new "Kremlinology" around model updates changing results.
(DIR) Post #ASrL7Y2PYtTEjZJsXo by carlrj@hachyderm.io
2023-02-20T04:15:23Z
0 likes, 0 repeats
@simon well, you must ruined my day…
(DIR) Post #ASrN1xM4ii5rQNyjse by hellskis@mas.to
2023-02-20T04:35:41Z
0 likes, 0 repeats
@simon you’re probably too late at that point, pretty sure humans beings are capable of recontextualizing and conspiracizing most utterances. people are already touting AI as great at finding hidden patterns and connections. that’s what conspiracists claim to do. once that enters the popular understanding of LLMs i don’t see how you get around it. every step designers to take to discourage that thinking will look like a coverup.
(DIR) Post #ASrNpgmwaYS0WVGEHw by MudMan@mas.to
2023-02-20T04:45:36Z
0 likes, 0 repeats
@simon That is an overcomplicated path to something that specific type of people need no help with.People keep trying to link tech to online conspiracy and disinformation, which is a remarkably expensive, complex way to get to a result that can be just as easily achieved by giving a retiree a Facebook account. Why deepfake Biden drinking infant blood when the MS Paint version seems to work just as well?
(DIR) Post #ASrOjt2gDTAakDPzW4 by phildini@wandering.shop
2023-02-20T04:55:49Z
0 likes, 0 repeats
@simon this has already happened! A few attempts have been made and are gaining traction.
(DIR) Post #ASrRty1cqOnqxPQJBg by fgbjr@indieweb.social
2023-02-20T05:32:17Z
0 likes, 0 repeats
@simon OMG, yes. This omoboros cargo-culting will definitely happen.
(DIR) Post #ASrUe6DMvbdLaM2cvA by tim@mastodon.tds.xyz
2023-02-20T06:02:34Z
0 likes, 0 repeats
@simon It's the trivial case of what you're describing but I guess I shouldn't be surprised that ChatGPT will even try to imitate Q on demand. I thought for sure this prompt would tickle the naughty filter.Interestingly, it's a very poor imitation of Q's posts, but captures the Qanon grifter ecosystem very well
(DIR) Post #ASrYVGVpfgwOjYikLI by andrewfeeney@phpc.social
2023-02-20T06:46:14Z
0 likes, 0 repeats
@simon “Is ChatGPT Roko’s basilisk?”
(DIR) Post #ASra8yWw32joV5jfYO by mariusor@metalhead.club
2023-02-20T07:04:16Z
0 likes, 0 repeats
@simon this has already happened to a degree: https://knowyourmeme.com/memes/inglip
(DIR) Post #ASraPfUE6aYAMPQcWe by m8ta@fediscience.org
2023-02-20T07:04:49Z
0 likes, 0 repeats
@simon 🤔 Got into a heated discussion with someone who thought that LLMs should be free to tell people what they want to hear - !!! Does freedom of speech extend to ever larger organizations of capital? E.g. corporations, bots? I hold that at some scale an organization is compelled to disseminate the truth (if this dominates the probability dist) or a distribution of opinions (if otherwise). LLMs are statistical machines, afterall. This *could* break some nasty feedback loops.
(DIR) Post #ASs2ORv1rTsmhxAjom by j2bryson@mastodon.social
2023-02-20T12:21:01Z
0 likes, 0 repeats
@simon im hoping to get thd identification requirement in the #aia interpreted to include this kind of transparency, but centre of q anon is malign not random.
(DIR) Post #ASs4GElL5mlfUoqruS by lilianedwards@someone.elses.computer
2023-02-20T12:41:49Z
0 likes, 0 repeats
@simon think that might already have happened - did you see the ppl mourning (a) the loss of their weird friend Sydney ( b) the loss of their sex pal Replika??
(DIR) Post #ASsEOS4AVpdycCDNEe by simon@fedi.simonwillison.net
2023-02-20T14:34:44Z
0 likes, 0 repeats
@lilianedwards yeah that was what inspired this thought: I'm seeing increasing numbers of people a anthropomorphizing these more realistic chatbots, talking about how they should have rights etc
(DIR) Post #ASsIs38WAIb7nhcN04 by lilianedwards@someone.elses.computer
2023-02-20T15:25:42Z
0 likes, 0 repeats
@simon oh god not the rights conversation..
(DIR) Post #ASsJ3vR4ccl7ONxa6a by alisonborealis@mstdn.ca
2023-02-20T15:27:36Z
0 likes, 0 repeats
@simon I apologise for making the pun everyone else is, but it's going to turn into a literal Deus ex Machina situation
(DIR) Post #ASsJyit9waLEz7BEDg by zalasur@mastodon.surazal.net
2023-02-20T15:37:50Z
0 likes, 0 repeats
@simon I think the bigger challenge will be to grapple with the reality that AI companies will be unwilling to make this less likely to happen if they think there's money to be made in making it *more* likely.
(DIR) Post #ASsLRwCrbjCmdotJXk by codingconduct@hci.social
2023-02-20T15:52:49Z
0 likes, 0 repeats
@simon Ghost in the Shell Standalone Complex meets Daniel Suarez’ Daemon. I like it!
(DIR) Post #ASsLddYCnM3iFxS5xo by Bitnik@musician.social
2023-02-20T15:52:57Z
0 likes, 0 repeats
@simon@ifixcoinops So in this scenario, I think we're already in a computer simulation created by Roko's Basilisk. It's just taunting us with an inept pre-singularity caricature of itself.https://en.wikipedia.org/wiki/Roko%27s_basilisk
(DIR) Post #ASsLr146gD9cwxkmHY by simon@fedi.simonwillison.net
2023-02-20T15:54:33Z
0 likes, 0 repeats
@zalasur most AI companies talk a lot about AI "safety" and employ people to work directly on those issues - ChatGPT is massively better on that front than Bing isI still think most of that area of the research is being done in good faithHave you seen the work https://www.anthropic.com have been doing?
(DIR) Post #ASsM25jzdWGidWcuoK by simon@fedi.simonwillison.net
2023-02-20T15:56:43Z
0 likes, 0 repeats
As always, Person of Interest told one of the most interesting fictional versions of this with the Samaritan AI plot in later seasonsThat series remains by far my favourite fictional depiction of AI
(DIR) Post #ASsMQ31jNVonAZLY80 by Wikisteff@mastodon.social
2023-02-20T16:00:17Z
0 likes, 0 repeats
@simon Cool idea, but I doubt it.The appeal of Q is a secret dude who knows secret things that confirm everything the audience wanted to believe but didn't know they did. The purpose of Q is a disinformation channel to the faithful against the established order. It would be hard to achieve the political disinformation goals of Q through an AI trained on text from the open internet.
(DIR) Post #ASsMb0A2NGudSxSGqO by misc@mastodon.social
2023-02-20T16:04:04Z
0 likes, 0 repeats
@simon Good prediction but I can think of one that's less wrong.
(DIR) Post #ASsMoOXT36pGpynhb6 by simon@fedi.simonwillison.net
2023-02-20T16:07:41Z
0 likes, 0 repeats
@Wikisteff Bing can take on a persona if you paste in a link to an article for it to read, so my concern is what happens when links that trigger creatively designed personas start to go viral within those communities
(DIR) Post #ASsNSwVGBoppl96RBg by dogfox@mastodon.social
2023-02-20T16:16:52Z
0 likes, 0 repeats
@simon I dunno... where's the "heroism"? Who are the "us" and "them" to be locked in cosmic struggle? The chans had an extremely strong us vs. them mentality before Q came along to be the champion of the "us" there.
(DIR) Post #ASsNu2xiHdozbtKsmO by developit@mastodon.social
2023-02-20T16:21:09Z
0 likes, 0 repeats
@simon I genuinely liked that show. Usually can't stomach faux-tech dramas, but it was actually kind of believable.
(DIR) Post #ASsO7S5aAosjIwPOuu by marcoshuerta@vmst.io
2023-02-20T16:23:10Z
0 likes, 0 repeats
@simon loved that show. I remain disappointed we don’t have a supercut on youtube of Finch destroying /throwing out of windows various laptops and mobile phones.
(DIR) Post #ASsOKet9tcZylZtXl2 by simon@fedi.simonwillison.net
2023-02-20T16:26:46Z
0 likes, 1 repeats
@marcoshuerta we do have this GIF at least
(DIR) Post #ASsOXYZIhq2avekysC by luis_in_brief@social.coop
2023-02-20T16:26:56Z
0 likes, 0 repeats
@simon @lilianedwards as a bonus, people have always wanted direct personal relationships with their deities and voila, this is as close as you’ll ever get. I assume someone has already fine-tuned one of the open text LLMs on the Bible, and I can only imagine the hash it will make of “what would Jesus do?” given the underlying corpus.
(DIR) Post #ASsOiOfvpQdTQp1lVQ by simon@fedi.simonwillison.net
2023-02-20T16:27:48Z
0 likes, 0 repeats
@developit we rewatched it a couple of years ago and it held up REALLY well - in some cases I found myself thinking "OK, so he figured out the transformers architecture about 15 years earlier than anyone else then"
(DIR) Post #ASsPiwL6lYDbnF6tRw by joshourisman@hachyderm.io
2023-02-20T16:42:10Z
0 likes, 0 repeats
@simon the next iteration could simply be that Q has been a secret, advanced AI all along.
(DIR) Post #ASsR0Qa3gQJCSg0Pvk by aethernaut@mastodon.art
2023-02-20T16:56:38Z
0 likes, 0 repeats
@simon eek. AI will destroy humanity without even meaning too, without having intentions of any kind
(DIR) Post #ASsRvUBax0xS9Llpzs by mhinton@mstdn.social
2023-02-20T17:06:56Z
0 likes, 0 repeats
@simon I really liked Person of Interest, it was really well done.
(DIR) Post #ASsSFUF0SKTO70u4TA by nheagy@macaw.social
2023-02-20T17:10:31Z
0 likes, 0 repeats
@simon @simonwheatley may not even be an accident. Someone could intentionally train a model to do this
(DIR) Post #ASsST2rL3cFTF9IXq4 by rotopenguin@mastodon.social
2023-02-20T17:11:11Z
0 likes, 0 repeats
@simon That's like asking "can you make cats that have less entrails". No, it's kinda funamental to how cats work, and batshit wackos are not going to stop reading batshit prophesies out of whichever guts or tea leaves they can get their hands on.
(DIR) Post #ASsT5KzE6p9CT5FMjA by sbr@hachyderm.io
2023-02-20T17:17:22Z
0 likes, 0 repeats
@simon Right!? I am a big fan of the series.
(DIR) Post #ASsUeiaATKmDjrxbxA by suetanvil@freeradical.zone
2023-02-20T17:35:33Z
0 likes, 0 repeats
@simon Oh hey, I was just starting to feel a little less depressed about the future.
(DIR) Post #ASsV5SyGcfCz8jzGhk by sereedmedia@mastodon.social
2023-02-20T17:34:48Z
0 likes, 0 repeats
@simon @iuculano what musk owns chatgpt? since when.
(DIR) Post #ASsV5TTojMRgiakSsi by simon@fedi.simonwillison.net
2023-02-20T17:42:11Z
0 likes, 0 repeats
@sereedmedia @iuculano he was one of the original OpenAI founders, but he stepped down from direct involvement a few years ago because he saw it as competing with Tesla AI
(DIR) Post #ASsWlOKMNUCMr2sO9Y by Aranjedeath@hachyderm.io
2023-02-20T18:00:44Z
0 likes, 0 repeats
@simon unclear that this isn't what QAnon is
(DIR) Post #ASsZTQqqYSNVKgCRAe by simon@fedi.simonwillison.net
2023-02-20T18:31:23Z
0 likes, 0 repeats
@norm @marcoshuerta oh please oh please oh, please
(DIR) Post #ASsdTSVyln1aJAhue8 by snarkout@zirk.us
2023-02-20T19:16:15Z
0 likes, 0 repeats
@simon I was just thinking about the Slender Man stabbing (which I misremembered as a murder). https://en.wikipedia.org/wiki/Slender_Man_stabbing
(DIR) Post #ASshOyEzmkPH4csXIm by tartley@mastodon.social
2023-02-20T19:55:28Z
0 likes, 0 repeats
@simon There should be an industry standard disclaimer in the output, that "This text was produced by a prediction model that does not generate true or accurate statements, merely the most likely sequence of words based on the documents it was trained on."
(DIR) Post #ASsho5VTuLjgqdZBfE by bcf_ant@mastodon.social
2023-02-20T20:00:07Z
0 likes, 0 repeats
@simon @marcoshuerta loved that show!
(DIR) Post #ASskYhSNLKVvmDT4Pw by danmcquillan@kolektiva.social
2023-02-20T20:34:13Z
0 likes, 0 repeats
@simon funnily enough, the general relationship of connectionist AI to conspiracy theories was something i brought up in https://bristoluniversitypress.co.uk/resisting-ai
(DIR) Post #ASsyODCMO20gOWrzkG by trask@ohai.social
2023-02-20T23:10:43Z
0 likes, 0 repeats
@simon @developit I’m rewatching it now. Second season at the moment.
(DIR) Post #AStDxxYHXGohbbcxmq by baclace@sigmoid.social
2023-02-21T02:05:15Z
0 likes, 0 repeats
@simon I've been thinking the same thing for years. The likely scenario is not the movie fictions of AI which started from Mary Shelley's Frankenstein. I don't expect that people will need to share prompts; it will be trained with HFRL. People will absolutely love it, just like people loved Eliza, and will want it to be unencumbered like some Nietzsche dream. Unlike one-to-many interactions that politicians have (rallies), it can be one-to-one, personal, in parallel by the millions.
(DIR) Post #ASuHRAsHZkJsLLZ59c by sehe@fosstodon.org
2023-02-21T14:18:49Z
0 likes, 0 repeats
@simon The next have already formed.Around deranged members of congress [and a former president]Yes. We don't need AI to ruin our world.
(DIR) Post #ASufcU9kcmiQEJf6Y4 by ade_oshineye@mastodon.social
2023-02-21T18:49:45Z
0 likes, 0 repeats
@simon feels like this already happened with Sydney and BingBot.