Post AdamqRN0fujfj0OSHI by mistersql@mastodon.social
(DIR) More posts by mistersql@mastodon.social
(DIR) Post #AdaZDqjR9FJBgX9SeO by simon@fedi.simonwillison.net
2024-01-07T02:37:10Z
0 likes, 1 repeats
I'm interested in exploring the intersection between people who have mostly avoided using LLMs and people who have embraced them, compared to the group who think they should be be labeled "AI" vs those who think it's OK to label them thatWhich of these do you fit into?
(DIR) Post #AdaZeKWmzQq5cnrXu4 by cube_drone@mastodon.social
2024-01-07T02:40:10Z
0 likes, 0 repeats
@simon I also get very angry when people call burgers "food" when they're very clearly burgers. It's me, the strawman who doesn't understand when one category is very clearly a subset of another category.
(DIR) Post #Adaa2w84q9nJC42MFM by purserj@aus.social
2024-01-07T02:42:09Z
0 likes, 0 repeats
@simon we've probably lost the naming battle but for myind this isn't AI I'm any meaningful sense.It's advanced machine learning and pattern matching but no actual sign of intelligence
(DIR) Post #AdaaQcru6SiAPIlgUC by argenis@mastodon.online
2024-01-07T02:43:58Z
0 likes, 0 repeats
@simon I often call it Stochastic Parrot, per the great paper of said name.
(DIR) Post #AdaaojwBTLRDHVGiDQ by John@socks.masto.host
2024-01-07T02:44:16Z
0 likes, 0 repeats
@simon Since you ask, here's my framework. I was introduced to AI in the early 80s. I read Marvin Minsky, and even had one online interaction with him. I was very aware of his firm calls on what was and what wasn't AI. I actually knew a guy who worked on perceptrons and held the grudge that Minski had bigfooted and killed the field.The reason? Perceptrons did not demonstrate general learning, particularly the child like learning the early AI researchers desired.We are kind of there again.
(DIR) Post #AdaaomkF2CfRzKhiW8 by John@socks.masto.host
2024-01-07T02:47:30Z
0 likes, 0 repeats
@simon Early AI researchers clearly expected knowledge representation and systems that understood correctness and truth.I mean that was kind of the problem, they had a highly cognitive view of what brains did, and missed that many creatures on earth with brains are not doing highly esoteric estimations before making choices.Maybe that results in a diffuse goal that confuses people. Are we looking for an electronic brain like that of a fly, or a chicken, or the expert in a field?
(DIR) Post #AdaaoqINpilD05hWeO by John@socks.masto.host
2024-01-07T02:55:10Z
0 likes, 0 repeats
@simon So anyway, for those reasons I'm not going to call something a "real" AI unless it has a concept of truth, and even better yet, an estimation of correctness for any output.If you ask an AI if orange juice is poison, it should answer not only with yes or no, but also a confidence in that answer.Ideally a confidence better than 50%.
(DIR) Post #AdabrRyinfKOPBJOAy by faassen@fosstodon.org
2024-01-07T02:46:59Z
0 likes, 0 repeats
@simonThere's an old saying about AI: when it solves a problem, the solution stops being AI and is just computer science.That's part of what is at work in this debate. I think this era of massive neural networks trained on massive amounts of human data has crossed a threshold and may in fact be different. Computers are now responding to natural language much more like a human would. I think the term Artificial Intelligence applies.
(DIR) Post #AdabrVXvXEGtTEo2y0 by faassen@fosstodon.org
2024-01-07T02:51:44Z
0 likes, 0 repeats
@simonIt's intelligent in humanlike ways even though the way it behaves and learns is only vaguely similar to what humans do; it's artificial.I don't understand why only a general intelligence or only a human intelligence could be deserving of that term. I have been surrounded by superhuman calculating intelligence all my life, and super human chess intelligence for much of it.But I understand the cultural bias from science fiction, and how these systems appear like more than they are.
(DIR) Post #AdabrXO4gpZxBGDCt6 by faassen@fosstodon.org
2024-01-07T02:54:18Z
0 likes, 0 repeats
@simonI don't think we are at the end of this development, and these systems will be increasingly more difficult to distinguish from humans in their capabilities. I don't think a super-intelligence that uses its intelligence to do science magic is likely though. Real technological breakthroughs need a lab, and time.
(DIR) Post #Adac46LAYp7h3pqO0m by cube_drone@mastodon.social
2024-01-07T02:45:06Z
0 likes, 0 repeats
@simon I think the people who object to calling LLMs AI think that what people mean when they say "AI" is "AGI", THE MACHINE IS HUMAN INTELLIGENT, but actually AI has served as a kind of CS topic salad since the 60s.Using A* search to play chess? AI. Using Prolog to do logic programming? AI. Using constraint satisfaction algorithms to do logistics? AI. Using bayesian filtering to detect spam? AI. Using LLMs to predict text? That's right, it's AI.
(DIR) Post #Adac4Fx4pFnEs0eHqa by simon@fedi.simonwillison.net
2024-01-07T02:47:39Z
0 likes, 0 repeats
@cube_drone yes! I should have mentioned AGI in my post, I'll add that
(DIR) Post #AdacSNMGwNFYn4w1B2 by danil@mastodon.gamedev.place
2024-01-07T02:50:04Z
0 likes, 0 repeats
@simon Google engineers who invent "transformers" say - "AI is machine that can communicate with people and world same as human and using human-language"
(DIR) Post #AdacfVG2shvx3UHXU0 by simon@fedi.simonwillison.net
2024-01-07T02:50:38Z
0 likes, 0 repeats
@John right - one of the amusing things about that 1956 conference is the bit that says "We think that a significant advance can be made in one or more of these problems if a carefully selected group of scientists work on it together for a summer." https://en.m.wikipedia.org/wiki/Dartmouth_workshop
(DIR) Post #Adacrfkxfog35iXsLw by irenes@mastodon.social
2024-01-07T02:52:49Z
0 likes, 0 repeats
@simon mm. we haven't used them much but we actively study them. we avoid the term "AI" although it's not exactly that we think it's not okay to call them that, it's just always been a useless term, throughout its history.so we voted as the first category.
(DIR) Post #Adad3B6ZSccBVD8hWa by brouhaha@mastodon.social
2024-01-07T02:53:32Z
0 likes, 0 repeats
@simon I think calling LLMs "AI" is a crock. I've used Dall-E a little bit for amusement, no neither "avoided" nor "used a lot" fit.
(DIR) Post #AdadFmTyJmOEtx5uvg by jeff@soapbox.hackdefendr.com
2024-01-07T02:54:52Z
0 likes, 0 repeats
@simon I have spent entirely too much time learning and using LLMs. Call it Artificial, but do not call it Intelligent....yet.
(DIR) Post #AdaecoHVtYxF5zoPRo by John@socks.masto.host
2024-01-07T03:22:59Z
0 likes, 0 repeats
@simon Yup, and Cyc has now been out there for 40 years trying to build a classic reasoning machine.https://en.m.wikipedia.org/wiki/CycTo paraphrase Rodney Brooks, not only do we not have a machine that can tell us whether orange juice is poison (via reasoning), we still don't know how to *begin* to build one.
(DIR) Post #AdafFjcBBCFPur2UAS by irenes@mastodon.social
2024-01-07T02:54:35Z
0 likes, 0 repeats
@cube_drone @simon it's just, "AI" has been a marketing term, not a technical one, throughout that history. it has always led people to incorrect conclusions. more than once in the last 50 years, all funding for "AI" has dried up for a decade or so as a direct consequence of unrealistic expectations which were deliberately seeded, in part through the use of that term.
(DIR) Post #AdafFkQABMEKPsQP4q by irenes@mastodon.social
2024-01-07T02:55:22Z
0 likes, 0 repeats
@cube_drone @simon even the proponents of these technologies should be wanting to call them by names which help people arrive at accurate mental models of what they do.
(DIR) Post #AdafFlE9BWDEutoJzE by simon@fedi.simonwillison.net
2024-01-07T03:38:22Z
0 likes, 0 repeats
@irenes @cube_drone I agree with that in principle, but I've not seen language that definitely, clearly helps as opposed to potentially confusing people
(DIR) Post #AdafFmbw2YppCxwphg by irenes@mastodon.social
2024-01-07T02:56:23Z
0 likes, 0 repeats
@cube_drone @simon anyway, sorry to be confrontational about it, just, it's work-relevant for us so we're always kinda talking about it.
(DIR) Post #Adah1J5pBrF7Vw9Zyq by hboon@mastodon.cloud
2024-01-07T03:45:25Z
0 likes, 0 repeats
@simon the results of your poll so far is quite surprising and I suppose this is already biased towards people using LLMs due to your audience.
(DIR) Post #AdahOfA9Il4ucfJTkm by irenes@mastodon.social
2024-01-07T03:45:57Z
0 likes, 0 repeats
@simon @cube_drone that's pretty fair. we usually say ML for the current hot stuff
(DIR) Post #Adaifoc9Vu8EoV9dAG by PaulM@infosec.exchange
2024-01-07T03:50:42Z
0 likes, 0 repeats
@irenes @simon @cube_drone I work on it, and call it an LLM, but I suspect I've lost the disagreement and AI is what it's called now, similar to "cyber"
(DIR) Post #AdaifpUOGFW7WiWwhk by simon@fedi.simonwillison.net
2024-01-07T03:52:20Z
0 likes, 0 repeats
@PaulM @irenes @cube_drone I've also compared it to how crypto really should mean cryptography!
(DIR) Post #Adaj6toyGN1AmT44kS by irenes@mastodon.social
2024-01-07T03:56:15Z
0 likes, 0 repeats
@PaulM @simon @cube_drone we do have a tendency to keep fighting these things long after everyone else has gone homeattached, please find our proudest achievement on twitter (note the trending category)
(DIR) Post #Adaj6ue1CZqpKmwqJc by simon@fedi.simonwillison.net
2024-01-07T03:58:16Z
0 likes, 0 repeats
@irenes @PaulM @cube_drone that is excellent!
(DIR) Post #AdajsG1DYDKlEfNJSK by abesamma@toolsforthought.social
2024-01-07T04:03:20Z
0 likes, 0 repeats
@simon the results so far imply a very skewed sample IMHO.
(DIR) Post #Adak3Q1qfp0rT0zaVs by mcc@mastodon.social
2024-01-07T04:04:07Z
0 likes, 0 repeats
@simon I mean, one complicating factor is that as far as I'm concerned 2020s "AI" is a marketing innovation and not a technical innovation. For this reason, association with the word "AI" is useful information because what we're communicating is the company has chosen to lie to us, which strongly implies various things about what other lies they might be telling or cheats they might be performing. I often use "AI" with scare quotes if I am trying to capture this phenomenon?
(DIR) Post #Adak3S2HDCXlgvCxSS by mcc@mastodon.social
2024-01-07T04:05:37Z
0 likes, 0 repeats
@simon I would not use a machine learning product which is marketed as "AI", but I could imagine under some circumstances using (or writing) a machine learning product which is not marketed as "AI", because even though the two products are probably based on the same technology the first company has given me an active reason to mistrust it and the second has not yet
(DIR) Post #AdakIIzH8UNvyJt49A by Arkana@freespeechextremist.com
2024-01-07T04:43:04.153329Z
1 likes, 0 repeats
@mcc @simon This technology has been in development and use for a while. Remember that the internet overall was a military project too1678756292531103.webm
(DIR) Post #AdakeyqiAuxIAuh504 by rebecca_meadows@mastodonapp.uk
2024-01-07T04:10:22Z
0 likes, 0 repeats
@simon The definition of 'AI' is very important here.
(DIR) Post #Adal6BNHiKnoqAttqa by cypherfox@mas.to
2024-01-07T04:24:04Z
0 likes, 0 repeats
@simon I am deep in the weeds with LLMs, and have been for a year now. I avoid the term AI not because of the marketing, but rather because imprecision in language _encourages_ misunderstanding.Humans have, for all our history, had tales of demons and devils and dark bargains with near-human beings. I appreciate your willingness to trust in the wisdom of your readers, but you are flying in the face of _centuries_ of inculcation.I use ML with general folks and use LLMs with folks in the know.
(DIR) Post #Adam1V07PrzqDVOa9o by simon@fedi.simonwillison.net
2024-01-07T04:28:04Z
0 likes, 0 repeats
@abesamma it's an interesting illustration of a hunch I have, which is that Mastodon users tend to have stronger ethical values which can correspond to avoiding LLMs!
(DIR) Post #AdamCYCBsXfTJJJOZk by irenes@mastodon.social
2024-01-07T04:28:39Z
0 likes, 0 repeats
@simon @PaulM @cube_drone thanks! hehe
(DIR) Post #AdamqRN0fujfj0OSHI by mistersql@mastodon.social
2024-01-07T04:30:55Z
0 likes, 0 repeats
@simon This 2 dimensional four square is missing the squares for people who have non-science communication reasons for wanting to call it one thing or another: polemicists who'd prefer that the target of their ire be named by only derogatory words. I think that guy is trying to say he was upset because you were not using derogatory words and implying that the non-derogatory name "AI" meant LLMs and their users/advocates are not worthy of contempt.
(DIR) Post #Adan2CE0oHhIECn3xo by SpaceLifeForm@infosec.exchange
2024-01-07T04:36:11Z
0 likes, 0 repeats
@simon Note that the avoidance crowd is over 50% of the votes.
(DIR) Post #Adanwv8d0k89jyJvM0 by Lobrien@hachyderm.io
2024-01-07T05:05:16Z
0 likes, 0 repeats
@simon Well, with a little more context from scrolling my feed, I _prefer_ ML or even "statistical ML," but you are simply correct on the fact that "AI" has been used continuously since the 50s. I don't understand why that statement would cause pushback, much less vitriol.
(DIR) Post #AdarGtfoSyNyqs3k00 by knitcode@infosec.exchange
2024-01-07T05:59:15Z
0 likes, 0 repeats
@simon clarifying my vote .. I haven't avoided LLMs because I don't like them, they just aren't well suited to my field (DNS threat Intel).. we do use language models and anything that solves our problems....if I used them, I still wouldn't use the term AI... Probably the mathematician in me.
(DIR) Post #AdarcMfTrNycrH8llo by NoiseColor@geekdom.social
2024-01-07T06:01:06Z
0 likes, 0 repeats
@simonIn some jobs there is no other way than to use them extensively.It doesn't matter what you what to call them, the public is gonna call them AI
(DIR) Post #AdatUV2p24lIQfR4lM by darkuncle@infosec.exchange
2024-01-07T06:15:01Z
0 likes, 0 repeats
@simon @PaulM @irenes @cube_drone I have a sticker from @zackwhittaker that says precisely that :)
(DIR) Post #AdatkmMuOFMKn9arei by PaulM@infosec.exchange
2024-01-07T06:21:29Z
0 likes, 0 repeats
@simon @irenes @cube_drone I'm not willing to concede that one yet!
(DIR) Post #AdauaOfJlLT8Kuokca by Divorytaur@mastodon.social
2024-01-07T06:34:29Z
0 likes, 0 repeats
@simon I only call it AI in the sense of: ‘Don’t feed the AI!’ 😅
(DIR) Post #AdavnR9HwNZzl0cNCC by ramblingpolak@infosec.exchange
2024-01-07T06:47:05Z
0 likes, 0 repeats
@simon @adamshostack
(DIR) Post #AdaxS2EBbK5mIEpbSS by v@federate.social
2024-01-07T07:00:50Z
0 likes, 0 repeats
@simon I think all intelligence is the same. I think the intelligence in humans is the same intelligence as in llms it's just a different configuration of it. now maybe "Artificial consciousness" or something would probably be more apt but thats largely a philosophical concept and why these ai chads are culty
(DIR) Post #Adb1Eyuh0oxK48kDyq by HaraldKi@nrw.social
2024-01-07T07:40:35Z
0 likes, 0 repeats
@simon The level of ability to pass the Turing Test of these Artificial Idiots is absolutely amazing and an astounding achievement.
(DIR) Post #Adb1bgl8ftvGQF6TZI by icing@chaos.social
2024-01-07T07:45:04Z
0 likes, 0 repeats
@simon Not a user, but one being abused. On curl‘s security team, we see the fruits of LLMs regularly.Scammers, looking for bug bounties, try to exploit the „sounds intelligent“ aspect of the models.Not a fan.
(DIR) Post #Adb3hGNrqmyCGa0Kh6 by sjaakkeuvelaar@mastodon.nl
2024-01-07T08:17:26Z
0 likes, 0 repeats
@simon To me, LLM still would be short for Master of Laws. I think that’s an answer.
(DIR) Post #Adb51zLOV1YNSBIHQG by alastairmrushworth@fosstodon.org
2024-01-07T08:33:42Z
0 likes, 0 repeats
@simon AI is fine imo. But popularly AI gets conflated with artificial consciousness and artificial free-agency which are not usually meant (or even required for intelligence). So possibly we need some rebranding.
(DIR) Post #Adb5PrIO4iXRRRGmDA by spinningthoughts@pkm.social
2024-01-07T08:34:31Z
0 likes, 0 repeats
@simon I prefer the term "LLM" because it keeps in mind the learning objective and framing of these systems. And I‘ve been cautiously poking at them - however, I have become increasingly sceptical of what they can achieve technically especially in terms of reasoning. And the economics of the space look bad. At this point I am seriously worried about the social impacts and that‘s a big reason for me to avoid usage of them.
(DIR) Post #Adb8B4UT2CRrspYoGu by gummibando@mastodon.social
2024-01-07T09:08:42Z
0 likes, 0 repeats
@simonFor me, most of the current stuff falls under the ML umbrella.I wonder who this Al guy is everybody’s talking about.https://youtu.be/PS-sE9xCb-g?si=CsrbyoEmvbc-bYST
(DIR) Post #AdbBuzpD0pn2b4quPI by RolfBly@mastodon.social
2024-01-07T09:50:46Z
0 likes, 0 repeats
@simon "haven't found any use for them", rather than "avoided".
(DIR) Post #AdbIvTJk3khvkHJ9I8 by basil@sarcasm.stream
2024-01-07T11:08:45Z
0 likes, 0 repeats
@simon I'm in between on both.I've used them a little bit but not a lot.I don't like that they are called AI, because I hate stuff misnamed like that (especially when it leads to misunderstandings with Joe and Jane Public, which this does)But as you say, the ship has sailed, and whilst I might decide to call the LLMs, there's zero chance of me attempting to convince anyone else of doing so.As you observed, the battle was lost long before I was born 😅
(DIR) Post #AdbLwZ8qBsVLxdQEme by gwire@mastodon.social
2024-01-07T11:43:09Z
0 likes, 0 repeats
@simon I've tended to use “Artificial Intelligence” to refer to research, ie things not yet usable, if it works it gets its own name. (Tonally similar to “if it works we don’t call it Alternative Medicine”.)But I’ve ended up using AI to refer to basically anything that uses machine learning, because that's how the media uses it now.
(DIR) Post #AdbR2XYjUDFcYsCWum by box464@mastodon.social
2024-01-07T12:40:05Z
0 likes, 0 repeats
@simon I support people at work that demand to use them, so I’ve had to learn the ins and outs myself. Let me tell you, the general public interest in them far outweighs the developer interest, which I find quite interesting. I’ve followed you because I’m interested in your use of locally hosted LLMs for specific cases. Really cool, thanks for sharing your work.
(DIR) Post #AdbgdsbA2t3D3AH2xs by enigma@norden.social
2024-01-07T15:31:24Z
0 likes, 0 repeats
@simonin german tv 20 years ago there was a funny commercial of ( very ) fast food of squared shaped & pressed toast of meat in egg coating . A voice warned all with " don't call it schnitzel!" 🤣Seriously every survey forget to clarify to who you talk to about the term AI. However every simple mind ( none expert) has a common sense about AI e.g. by Spielberg's movie A.I. . Me , as an IT expert even didnot know that GPT is synonym of LLM. I even not knew what LLM really exactly means 😪😓🤯
(DIR) Post #AdbignAesgTOzhOUe8 by Biggles@qoto.org
2024-01-07T15:57:50Z
0 likes, 0 repeats
@simon Until I meet a LLM that is capable of responding "I don't know" when asked about something missing in its training instead of "hallucinating". The word "intelligence" has certain foundational implications that are missing here. This isn't to say it can't be made into a useful tool, just that it looks so much like intelligence that *calling* it intelligence seems intentionally misleading. I dip my foot back in every time a new model comes along - basic factual questions - and have yet to see an "I don't know".
(DIR) Post #AdbjseAeGkE8Vfbdr6 by simon@fedi.simonwillison.net
2024-01-07T16:09:19Z
0 likes, 0 repeats
@Biggles GPT-4 is pretty good at saying when it doesn't know something these days - hallucinations can still slip through but they are a lot less common https://chat.openai.com/share/f75295f2-3252-4841-a532-dcd6b62418b6
(DIR) Post #AddNdxypj4cMLVDKcK by andrewt@mathstodon.xyz
2024-01-08T10:46:01Z
0 likes, 0 repeats
@simon i'm one of the 13% and honestly kind of surprised not to see more people in the "we might as well call this 'AI' since 'real AI' is never going to happen" camp.
(DIR) Post #AddWuTvZc4n5eLdgpc by x_cli@infosec.exchange
2024-01-08T12:55:20Z
0 likes, 0 repeats
@simon Now that the poll is over, can you clarify if you were comparing AI vs ML labels or if it was something else entirely? ;)