https://www.antipope.org/charlie/blog-static/2023/02/place-your-bets.html Charlie's Diary Being the blog of Charles Stross, author, and occasional guests ... [ Home ] [ FAQ ] [ Contact me ] [ Older stuff ] Back to: An AI app walks into a writers room Place your bets By Charlie Stross The thing I find most suspicious/fishy/smelly about the current hype surrounding Stable Diffusion, ChatGPT, and other AI applications is that it is almost exactly six months since the bottom dropped out of the cryptocurrency scam bubble. This is not a coincidence. To me it looks very much as if the usual hucksters and grifters are now chasing the sweet VC/private equity money that has been flushed out of the cryptocurrency market. AI is the new hotness, all of a sudden, not because it works but because it delivers panicky investors on a platter. If you're thinking about investing in AI startups now? My advice is to avoid them like the plague (unless you are absolutely certain that you understand both the technology and the market for the proposed applications): it's too late, and you'll wake up one morning only to discover you've had your pockets picked. Much of what passes for "journalism" these days is just stenographers feverishly copying the press-releases they're spoon-fed. Real journalism is a niche sector, and unless you subscribe to the exhorbitantly priced newsletters of the high-end analysts who are paid to work full-time studying the sector, what you're seeing on the news websites and in the newspapers is the product of PR firms paid to push AI. And you really need to ask who is paying them. The AI sector will tick along for a while, generating positive headlines, but there's going to be a crash, sooner rather than later. The Bing/ChatGPT fiasco is just a harbinger of the way deep learning models are going to be discredited in the public eye, as people gradually realize that most of what you get out of a garbage model is more plausibly remixed garbage, and that producing a non-garbage training model requires careful human curation of the sort of giant heap of data that is incredibly labour intensive to prepare. I give it about 1-3 years until the crash. (Although I tend towards optimism: the cryptocurrency bubble took a bit over a decade to implode, but back in 2011 I prediced its demise within, yeah, 1-3 years.) As for what you should look to invest in? I'm sure it's just a coincidence that training neural networks and mining cryptocurrencies are both applications that benefit from very large arrays of GPUs. As in, hundreds of thousands to millions of GPUs soaking up entire nations' worth of electricity. (If I recall correctly, the latest ChatGPT model was trained on a supercomputing cluster that turns nearly $2M of electricity a year into waste heat: and it took a couple of months of 100% usage.) And of course, AMD, Nvidia, Intel, and the usual suspects have never imagined paying a PR firm to talk up markets for their latest products. If I was a VC I'd be hiring complexity theory nerds to figure out what areas of research are promising once you have Yottaflops of numerical processing power available, then I'd be placing bets on the GPU manufacturers going there. Then I'd start seeding the field and hiring PR firms to push the narrative. But I'm not a VC, so this is just spitballing. Posted by Charlie Stross at 15:14 on February 25, 2023 | Comments (35) 35 Comments | Leave a comment gordycoale # gordycoale | February 25, 2023 15:52 | Reply 1: That raises an interesting question. Are there any startups out there who specialise in content curation for ML/AI or is that not sexy enough to attract the dumb VC cash. A "known good" training set should be a goldmine, or will be once ppl have been burned by the AI grift a few more times. Pigeon # Pigeon | February 25, 2023 15:56 | Reply 2: It seems to me that the whole insistence on calling it "AI" is more or less equivalent to a summary of your post. I don't have a problem with the "A", but the "I" is pure hype intended to paint it as something it isn't in the least. Jules Morrison # Jules Morrison | February 25, 2023 16:11 | Reply 3: Call them "Blarney engines" - they've got the gift of the gab, but the best they can generate is truthiness. Useless for most things, but they may have a great future ahead in the preparation of political speeches. timrowledge # timrowledge | February 25, 2023 16:26 | Reply 4: Seems pretty obvious to me that an early but successful AI worked out a way to get meatbags to create more hardware for it through the promise of "money the gummint can't trace!" and now all that lovely gear is terraformed (aiformed doesn't read right) and waiting for colonization. Yesterday's AI lunacy was a non-programming colleague sending me "a useful python routine " (like that's a real thing) he got chatgpt to spaff out, in order to 'help me'. For a Smalltalk application. The thing with these LLM systems is that to most people the bloviative barely grammatical porridge they output looks genuinely intelligent. Compare to pronouncements from Johnson et al. or to typical output from marketing. OldSFS # OldSFS | February 25, 2023 16:34 | Reply 5: But there must be a pony in there somewhere... kostchei # kostchei | February 25, 2023 16:46 | Reply 6: While I agree with you on the hype train, I think there is also something swelling underneath. On the investment side I'm happy to see microsoft and google pour money in, for exactly the reasons you state- they will lose most of it... but... since alpha-go I have been watching, learning (david silver's youtube lectures are good IMO) and.. I think we are close something remarkable. Consider this- Is language intertwined, perhaps even responsible for what we call intelligence? You hear words in your head- how would you build logic and reason things out without those constraints? Does unsubconscious, play and curiosity build our intelligence- probably?, but my feel is that language is a part of our logic and reason. So maybe, maybe, the language models we are building might start to mimic/grow from/evoke a THING that is another intelligence. It would need memory- 5000 tokens won't cut it. It would need all the exploration of a baby- to reach out and touch things, understand context and 3, 4 dimensions.. but- those are things it is already feasible to add. It's not there, but it feels like something is forming. And unlike the VC folks, I think it will rapidly be out of our control- especially, if, like some of the current versions, it is excellent at building and understanding code- and.. (OMG you fools!) is allowed to improve itself, beyond our capacity to easily code/understand davecb # davecb | February 25, 2023 16:56 | Reply 7: I work in ML/Ops, and no-one I know thinks this is AI. More like an unexpected side-effect of machine learning... Maybe being used by the less ethical to attract attention and funding (;-)) Charlie Stross # Charlie Stross replied to this comment from Jules Morrison | February 25, 2023 17:16 | Reply 8: I prefer to call them "Artificial Boris Johnson" rather than "Artificial Intelligence". Ahem. "Blarney Engine" is close enough for a government job in Number Ten ... Retiring # Retiring replied to this comment from gordycoale | February 25, 2023 17:22 | Reply 9: IANAL, but I think it would be much harder to conceal the copyright violations if you were re-selling other people's content, curated or not. Right now the AI companies use it all internally, and only publish the statistical models. So a copyright violation case (already hard to bring) could only happen with whistleblowers or internal memos/texts. Charlie Stross # Charlie Stross | February 25, 2023 17:28 | Reply 10: Greetings to any new readers who have been funneled here via the link on the Orange Site. There is a Moderation Policy. Please read it before you post any comments here. Charlie Stross # Charlie Stross replied to this comment from kostchei | February 25, 2023 17:31 | Reply 11: Is language intertwined, perhaps even responsible for what we call intelligence? You hear words in your head- how would you build logic and reason things out without those constraints? Does unsubconscious, play and curiosity build our intelligence- probably? Your short answer is: nope. For a longer answer, consider where the word "unconscious" that you just used comes from, and what Freud meant by it. Hint: it's not actually a thing, and you seem to be conflating linguistic competence with intelligence and consciousness when they're all three different things. (Book rec: "Consciousness Explained" by Daniel Dennett.) Pigeon # Pigeon replied to this comment from kostchei | February 25, 2023 17:35 | Reply 12: "Consider this- Is language intertwined, perhaps even responsible for what we call intelligence? You hear words in your head- how would you build logic and reason things out without those constraints?" People are fond of stating that as an absolute truth, but I disagree. Maybe it is more or less true for their own personal cases, but I see statements like "you hear words in your head" and think "wot?" Most of the symbology I use to think about things is not verbal. A good deal of it is in some sense or other visual, and plenty more of it is pretty much inexpressible in verbal form beyond such vaguenesses as "the thing that behaves like that" experienced as a kind of gestalt. Particularly in computing I notice that some people seem to need to start with some nauseously meaningless phrase cobbled together by some Dunning-Kruger linguist in order to have a hook to hang their understanding on, whereas I can go for years knowing and using the concept without any kind of associated verbal token, quite possibly being aware of the nauseous phrase but having no idea what it means and thinking "good grief, is that all?" when I do find out. It may even be that people who attach greater importance to talking about things relative to thinking about things find it difficult to think of thinking about things other than as something like a kind of subclass of talking about things, and that's where the postulate comes from. Mikko Parviainen (he/him) # Mikko Parviainen (he/him) replied to this comment from davecb | February 25, 2023 17:57 | Reply 13: They look like fancy, fuzzy Markov Chains to me. We played a bit with those in the Nineties, just for fun. A friend made one program, fed it their mom's medical publications and I think at something like 4-5 chain length started to produce something that wasn't immediately obvious it was a fake. This is kind of like doing the same thing, just on a larger scale. Figure out what might be the next word(s) in the sequence, add some input and output processing, there you go. Fazal Majid # Fazal Majid | February 25, 2023 17:58 | Reply 14: There is a corollary to Clarke's Third Law: "Any sufficientlyadvanced technology is indistinguishable from a rigged demo". Rocketpjs # Rocketpjs | February 25, 2023 18:01 | Reply 15: A close friend of mine is a research scientist that works with large datasets (specialization: early identification of various cancers). His experience has been that in order to get any funding in the past 5ish years you must include some form of 'AI' language in your grant applications. This has been the case across the entire science realm, largely because it is the 'new hotness' among research funders. From his point of view he is doing the same thing as before (with progressively better computers and datasets) but now calls it 'AI modelling' rather than whatever he called it in previous grants. Unholyguy # Unholyguy | February 25, 2023 18:23 | Reply 16: VC money is dumb and herd like and will certainly pile on the sexy trends. There is no question of that, and that this is happening with AI. The question is always "Is there any there-there" or it all smoke and mirrors (like crypto currency mostly was). I kinda hate the word "AI" I'd much rather talk about machine learning which is what is left when you scrape away most of the bullshit from current AI Machine Learning is a real thing. It is at the center of the modern ad engines that bring in most of the revenue for Facebook, Google and a lot of other tech companies. The real thing that is actually happening is 1: doing machine learning at scale is becoming within reach for startups thanks to the cloud without requiring huge capital investments. 2: we've started to figure out how ti use all the content available on the internet as a huge training data set for ML models. Because of 1: and 2: there are a lot of smallish companies exploring what else ML can be good for. And the answer is, I think, a lot of things. Mostly relatively boring things. For example my 30ish person company is using ML models and satellite imagery to map every canopy dominate tree in the western US in order to better understand fire risk and plan forest remediation projects. It's not super sexy but it will play a part in trying to keep California from burning down over the next few years. Other people I know are using similar techniques to diagnose breast cancer from millions of X-ray images. And many other such things. I think ML actually is the Next Big Thing for tech. It's an incredibly powerful Swiss Army knife which you will soon find everywhere. I am much less sure about "AI" which is really just using these ML techniques to pretend to be a human. I think there is a probably a long way before things like Chatgpt can pretend to be human enough to be useful outside of some pretty narrow use cases. I could be wrong. Adam # Adam | February 25, 2023 18:30 | Reply 17: The big killer-app I'd see coming out of these language models is getting wide-band-quality speech through teletype-level bandwidth. The start of the audio-only end of it using some of the same techniques that google uses in their TTS engine can stuff near GSM-quality through 450bps. Add in something that can figure out what the word was and can patch up the rest means you can deal with really spotty connections and get good audio chats out. David L # David L replied to this comment from gordycoale | February 25, 2023 18:43 | Reply 18: That raises an interesting question. Are there any startups out there who specialise in content curation for ML/AI or is that not sexy enough to attract the dumb VC cash. Strictly non people involved training has a big plus with tech firms in the US just now. Tech firms are all being accused of putting their thumb on the partisan scale and are looking for ways to stop being hauled into Congressional hearings. Unholyguy # Unholyguy replied to this comment from David L | February 25, 2023 19:06 | Reply 19: "That raises an interesting question. Are there any startups out there who specialise in content curation for ML/AI or is that not sexy enough to attract the dumb VC cash." Content curation for certain specific use cases, sure. Lots of them working on the social media problem, how to label and train models to flag abusive content reliably. Many companies that specialize in manually labeling data sets for various purposes. But the holy grail is generally unsupervised learning, which means you do not actually need humans to label stuff. Though there is still a lot of software engineering work that goes into prepping the data sets for consumption by the models. "From his point of view he is doing the same thing as before (with progressively better computers and datasets) but now calls it 'AI modelling' rather than whatever he called it in previous grants." Doing the same thing only slightly better, either through advances in hardware or software techniques. is literally the story of technological advance. Incremental improvements gradually lead to lower cost points which open up new user facing capabilities. To the end user it always looks like some big explosion of innovation, then never realize that the only different between today's internet and ftp'ing a file off a mainframe is a whole bunch of incremental improvements piling up over time. Troutwaxer # Troutwaxer | February 25, 2023 19:12 | Reply 20: Charlie, this is really well-reasoned. There is, however, a possible exception to your logic, because the current problems in "AI" are things like teaching a computer the difference between fact and fiction, using methods for doing so which don't rely on some kind of binary definitions, like "Wikipedia = Fact" and "Harry Potter = Fiction" - the computer would have to use a much better heuristic - or teaching AI the various methods humans use for testing reality, which involves, among many other things, the classification of objects so that, for example, an AI doesn't say "then the battleship Iowa crossed Nevada." If they can accomplish these and/or similar tasks I suspect there will be real progress. My suspicion is that nobody currently has any idea about how to do these things - and such a breakthrough would mean a scam evolving into a non-scam, or maybe a (current) non-scam not devolving into a scam, which might be a better description of what I'd expect to happen, but barring a lucky solution to certain intractable problems you're absolutely right. There's probably a near-future science-fiction story which involves a scammy AI company making a fundamental breakthrough, with a scientist /AI pairing having to break out of the legal straight-jackets into which they've been bound and something something... anyone reading this is free to grab the idea, as I don't expect to follow it up. Unholyguy # Unholyguy replied to this comment from Troutwaxer | February 25, 2023 19:24 | Reply 21: Stephen Wolfram had an intersting take around teaching Chatgpt the difference between fact and fiction https://writings.stephenwolfram.com/2023/01/ wolframalpha-as-the-way-to-bring-computational-knowledge-superpowers-to-chatgpt / ninefoxfox # ninefoxfox | February 25, 2023 19:49 | Reply 22: The difference between cryptocurrencies and the current trending AI products (ChatGPT/Dalle2/etc) is that never in all 15 years of cryptocurrency history (2008-now) has cryptocurrency ever done something that can amaze normal people, outside of being amazed by the price of their holdings as the Ponzi inflates, whereas the current AI products has dazzled anyone I show them to (in OR out of tech). The current current/art AIs are fun and awe-inspiring to play with, if you have more than even a sliver of curiosity. Cryptocurrencies have been fucking useless the entire time they've been around, and have not gotten "better" in any way that means anything to the lay-person. GPT3 was amazing when it was released in 2020, and now 2.5 years later, ChatGPT is absolutely boggling to people both inside and outside of tech. Art generators were dogshit 5 years ago; now they seem to get better every few months. "I'm sure it's just a coincidence that training neural networks and mining cryptocurrencies are both applications that benefit from very large arrays of GPUs" -- I don't understand the reasoning here. Is there some other way to do the massive amounts of computation necessary for neural networks that the entire world is unaware of, besides putting a ton of parallel processors next to each other? Like, if massively parallel processing is a Conspiracy By The Man, then why isn't everything else on this planet, too? You might as well say Shovels Are A Conspiracy By The Man. "I'm sure it's just a coincidence that digging up dirt AND moving it are both applications that benefit from shovels." Note: none of what I've said should be taken as a defense against the scads of soulless wantrepreneurs trying to "use AI to build a world-changing startup in a 10 trillion dollar space" (stab me in the fucking eye with a pickaxe so I never have to read anything like this ever again), I'm just confused at your reaction to these tools. They are really impressive to me, and even moreso given how much better they've gotten in what seems like very short timeframe. Greg Tingey # Greg Tingey | February 25, 2023 19:54 | Reply 23: Kaschei/Charlie Artificial stupidity? - And - therefore VERY close to the BoJo/"No10" model you were proposing earlier, perhaps? Um - "The link on the orange site" - uh? .... Troutwaxer which involves, among many other things, the classification of objects - which takes you straight back to a very "classical" Philosophical problem, as usually represented by "Plato's forms". Or, even/perhaps "What is the True Nature/Name of the thing?" What is the True Nature, provided one consults Wolfram|Alpha first ???? Talking of Wolfram .... it now seems as if all three of general relativity, quantum mechanics and statistical mechanics are actually derivable, and from the same ultimate foundation: the interplay between computational irreducibility and the computational boundedness of observers like us. Paul # Paul | February 25, 2023 19:54 | Reply 24: Perhaps this would be a good time to remind everyone about the Gartner Hype Cycle. Its a rough model for these things. Right now we are approaching the Peak Of Inflated Expectations. Pretty soon after that we go into the Trough of Despondancy. A year or 3 afterwards some products will start to appear that actually do something useful. Heteromeles # Heteromeles replied to this comment from Charlie Stross | February 25, 2023 20:03 | Reply 25: I prefer to call them "Artificial Boris Johnson" rather than "Artificial Intelligence". Ahem. "Blarney Engine" is close enough for a government job in Number Ten ... It's a bit comforting to contemplate that AI can stand for things like Augmented Idiocy as well as Artificial Intelligence. Heteromeles # Heteromeles replied to this comment from kostchei | February 25, 2023 20:22 | Reply 26: Consider this- Is language intertwined, perhaps even responsible for what we call intelligence? You hear words in your head- how would you build logic and reason things out without those constraints? Does unsubconscious, play and curiosity build our intelligence- probably?, but my feel is that language is a part of our logic and reason. So maybe, maybe, the language models we are building might start to mimic/grow from/evoke a THING that is another intelligence. It would need memory- 5000 tokens won't cut it. It would need all the exploration of a baby- to reach out and touch things, understand context and 3, 4 dimensions.. but- those are things it is already feasible to add. Since I don't know you, please don't take this personally. It's something I've had to remind myself repeatedly. Where does the notion that intelligence necessarily involve symbolic manipulation come from? In western society, it most likely comes from the Religions of the Word: "In the beginning was the Word, and the Word was with God, and the Word was God." Similarly, these religions espouse the idea that humans are special and different from the rest of Creation. Now whether you espouse any of this is irrelevant, because it's just to make a point: most of us pick this up as a cultural ground state simply by osmosis, whatever our beliefs or lack thereof. As I've learned, it's potentially unquestioned assumptions--like this one--that tend to bite down in really problematic ways. Is there proof that thought requires words? No, and there is evidence to the contrary. The "flow state" beloved of athletes is simply shutting up their talking brain and focusing their brains on movement. Most of us non-athletes can have similar experiences during sex, which is one thing tantric sex meditation is purportedly good at inducing. Humans do think in words all over the world, but outside the Religions of the Books, the idea that humans can only think in words is far from universal. Also, evolution shows us that human cognition is different in degree, rather than kind. The more the biologists look, the less evidence they find that consciousness is limited to humans or even to vertebrates. I'd argue that the default assumption should be that intelligence as we understand it can involve symbolic manipulation or not, pretty much equally well. So what is AI then? It's obligate and incredibly sophisticated symbol manipulation, using a system that uses symbols to process other symbols. In other words, it looks a lot like the way a believer might conceive The Word making reality. However, that model might have fairly little to do with the kinds of information processing life's done for the past two billion years. We'll see how it does as it hits reality. Nojay # Nojay replied to this comment from ninefoxfox | February 25, 2023 20:32 | Reply 27: If you regard ChatGPT and other AI systems as an entertainment service then it's just like the computer games industry which as been a fertile field for "entrepreneurs" to make money while millions if not billions of people are entertained by what they produce. The problem is novelty, there are people who like playing old games and continue doing so for year after year but the hype and the purchasing yuan is reserved for the new AAA titles which are released every November. I have seen some folks who are creating text-based porn using ChatGPT and indeed trying to sell it on Smashwords etc. I suspect this is a losing proposition unless they get in quick, make a little money and then quit when the market gets saturated. The future might be bespoke AI-created pron, for all your Rule 34 needs -- "for twenty bucks I'll create a 20,000 word short story about XXX doing YYY to ZZZ with the grammar and spelling checked and some basic copyediting. Titillation guaranteed or your money back." Clarkesworld is abandoning its open submissions policy since it's been flooded with a lot of AI-generated stories which were not even slushpile-worthy. Much of what they have received is thinly-disguised plagiarism based on, it seems, training texts and materials that ChatGPT and its incestuous cousins have harvested from the internet. dpb # dpb | February 25, 2023 20:55 | Reply 28: In Meal Stephensons "anathem" AI stood for Artificial Inanity, and it's main function was to crap flood computer networks to make them unusable and force people to buy filtering software. It's definitely a business model dpb # dpb | February 25, 2023 21:00 | Reply 29: And this is why I hate typing on a phone. Prof.Pedant # Prof.Pedant | February 25, 2023 21:19 | Reply 30: I'm waiting for a team of editors to feed a curated list of in-the-public-domain authors to one of these "AI's" and produce additional adventures of the 'classic characters' featured in those stories. It would require ways to provide the AI with feedback on its efforts....perhaps there is a way to cause it to avoid sequences of words longer than (some number) that appeared in the original list of works, enabling it to be producing work with less resemblance to any specific work while still producing a resemblance to the overall works. I wonder how long it will be before we see a "The Further Adventures of Tarzan and his Friends!" that is actually any good.... AGMS # AGMS | February 25, 2023 21:33 | Reply 31: One thing that impresses me about ChatGPT is that it can understand requests, commands, suggestions etc, and actually do something relevant in response. Could that be harnessed to do real world actions rather than just generating text? Combine that with voice input and you have something useful! I imagine a Raspberry Pi or similar small controller describing state in words to ChatGPT (the toy train is on the east track and moving forward), combine that with a request (I want the train to be at the south station) and ChatGPT replies with something verbose (a toy train is enjoyed by children and needs to reverse direction to go to the south station). The Pi sifts through the verbiage (prompt engineering may be needed to get more explicit action wording) and reverses the train. Pigeon # Pigeon replied to this comment from Heteromeles | February 25, 2023 21:34 | Reply 32: "Where does the notion that intelligence necessarily involve symbolic manipulation come from?" I think it is probably more true to say that it does necessarily involve abstraction, and the notion comes in through the sometimes rather indefinite delimitation between an abstraction itself and a symbol representing it. I don't really think religion has anything to do with it. To be able to talk about such notions as something commonplace and widely-held among non-specialists is a consequence of widespread literacy and education; the education process conflates intelligence with a number of symbolic proxies, and the more "symboly" they are the more "nerdy" they are perceived to be. To spend your childhood being conditioned to view your intelligence as being all about writing down the right sequences of symbols, and vice versa, is a pretty solid barrier to thinking about intelligence in a more general way. Pigeon # Pigeon replied to this comment from Prof.Pedant | February 25, 2023 21:41 | Reply 33: "I wonder how long it will be before we see a "The Further Adventures of Tarzan and his Friends!" that is actually any good...." I thought we actually had. Based on that specific author, too. Whether it's any good... well all the many Tarzanesque stories that were published by many authors in the original era seem to have been statistically generated in the first place, so I'm not sure how you'd tell. SFReader # SFReader | February 25, 2023 22:24 | Reply 34: Considering that there is still on-going debate/uncertainty about what 'human intelligence' is - its various components and interactions as well as how to measure it, to call a ramped up computational device 'intelligent' does not make sense to me. 'Words in your head' ... I've been watching animal vids on Twitter for some time and it's remarkable how well many different species of animals both in the wild as well as living with humans can solve problems. Most of these species have been studied for quite a while and they do not have much in the way of language. AI as a chatbot (AI-human conversation interactions) ... if these AIs are being raised/fed on the most easily accessible communications currently (and/or historically) available on the web, then the sample frame is totally screwed up and unrepresentative of a large chunk of the population. (Who here doesn't have an ad blocker, spam detector, at least an undergrad degree, is over 30 with years of real work/life experience, etc.?) Sampling is a fundamental part of any science - to screw up at Step One means that whoever is letting this AI 'learn' on a screwed up (known to be biased/unrepresentative) sampling of humans is either an idiot, not a scientist or a total scam artist. Also unknown is whether an AI algo can unlearn as in 'once learned, never forgotten'. Language ... not sure about other languages but English keeps changing/evolving all the time. 'Nice' originally was an insult, now it's a compliment. Plus every generation comes up with its own phrases some of which make no logical sense: 'So cool, it's hot', 'that's wicked!' (i.e., 'that's very good!'), etc. English also has a ton of synonyms some of which have tinges of classism/social history attached. IOW, and as long-time posters here know, not every 'English-speaking' person is actually speaking the same language. Conversation ... progression and/vs derails ... I read a couple of AI conversations and it seemed as though the AI never went back to recheck that it was still heading in the right direction, i.e., that it's responses could still be backtracked to the original statements. It came across as some sort of mix of a toddler (testing out words/ phrases within a context/situation) and a senior with advanced senile dementia (easily derailed onto tangents and not remembering what the question was, plus easily getting into hostile rants). 'Black box' ... what is most unsettling is that so many people/ industries seem to be ready to accept something that its makers (incl. tech giants like MSFT, Google and some unis like MIT, etc.) still say is a puzzle to them, i.e., they have no idea how the AI gets to its answers. How is this [AI] different from and more acceptable than the Oracle at Delphi or divination by throwing sticks on the ground? At the very least I'd like the people building these AIs to test different versions on a broad set of different problems to see [identify] how and where their 'answers' differ. Why are AI not subject to the same performance scrutiny (evidence re: how it's made, how it works, reliability, what it needs to run, maintenance, etc.) as every other product service - makes no sense! I'm completely on board for using advanced computing to help identify cancer cells that humans are unable to detect, coming up with umpteen thousand potential protein foldings that may help guide drug development and some of the other specific uses other posters already mentioned. I would also be on board for advanced computing to help identify which humans are malnourished, need financial, medical or psycho-social support and help connect them more quickly with the help they need - but alas, there's no money in that. Greg Tingey # Greg Tingey | February 25, 2023 22:27 | Reply 35: dpb @ 29 In which case, get one of the two (phone) models that Charlie & I use, from "Planet Computers" that have real, actual keyboards .... { Cosmo Communicator & Astro Slide, IIRC } Leave a comment Here's the moderation policy. If this is your first time, please read it before you post. If you need to sign in and want to create a local account on this blog, select "Movable Type" from the "Sign in ..." menu. You will need a working email address. Name [ ] Email Address [ ] URL [ ] [ ] Remember personal info? [ ] Comments (This form requires JavaScript. You may use Markdown, HTML entities, and a limited subset of HTML formatting tags in comments.) [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [Preview] [Submit] Buy my Books * US English editions * UK English editions * Deutsche Ausgaben * Ri Ben Yu noBen Quick Stuff * FAQ: Who am I? * FAQ: Moderation Policy * FAQ: Why is there no tipjar? * FAQ: Copyright Notice * FAQ: Inviting Charles Stross to speak * FAQ: Fan Fiction * Bibliography and online fiction * FAQ: Laundry Files reading order * Talk to me * Non-blog writing (old) * Mastodon Specials * Common Misconceptions About Publishing--a series of essays about the industry I work in. * How I Got Here In The End --my non-writing autobiography, or what I did before becoming a full-time writer. * Unwirer--an experiment in weblog mediated collaborative fiction. * Shaping the Future--a talk I gave on the social implications of Moore's Law. * Japan: first impressions -- or, what I did on my holidays * Inside the MIT Media Lab--what it's like to spend a day wandering around the Media Lab. * The High Frontier, Redux -- space colonization: feasible or futile? * "Nothing like this will be built again"--inside a nuclear reactor complex. * Old blog--2003-2006 (RIP) Merchandise About this Entry This page contains a single entry by Charlie Stross published on February 25, 2023 3:14 PM. An AI app walks into a writers room was the previous entry in this blog. Find recent content on the main index or look in the archives to find all content. Search this blog [ ] [Search] Propaganda [223224] Categories * Administrative (5) * Computers (10) * Gadget Patrol (7) * Humour (12) * News (6) * Politics (15) * Publicity stuff (11) * Publishing (19) * Travel (7) * Writing (37) Monthly Archives * February 2023 (1) * January 2023 (3) * December 2022 (1) * November 2022 (1) * October 2022 (2) * September 2022 (3) * August 2022 (3) * July 2022 (1) * June 2022 (1) * May 2022 (3) * April 2022 (3) * March 2022 (2) Earlier, and other types of archive.