 ππ΅πΈπΌπΌπͺπ»π Blockchain: a slow database Crypto: an expensive slow database NFT: an expensive slow database to store URLs AI: a way to write slow and inefficient algorithms LLM: a database that stores text in a slow and inefficient way Chat GPT: an expensive imprecise query language for slow and inefficient text databases that often returns wrong results 141 comments
@thomasfuchs Society glorifies the stories of successful salesmen, not the engineers whose innovations they sold.  @ganonmaster @thomasfuchs I remember thinking this as a child after reading the story of Steve Wozniak and Steve Jobs in a computer magazine!  @thomasfuchs My wife was volunteering at Lighthouse for the Blind and one of the other volunteers had a thing he wore on his head that was an interactive wayfinding tool that described everything he saw and read signs. It was mindboggling. (It's gotten better since.)  @glennf @thomasfuchs could you provide me more info on Lighthouse for the Blind? Is it this one? https://lighthouse-sf.org/   @DeltaWye @thomasfuchs Apparently the normal version is OpenCL and ROCm. Still, batch computing on the GPU when you aren't using it for graphics is getting a lot of power out of the hardware that we just didn't utilize before. @Raccoon @DeltaWye @thomasfuchs OpenCL was an early attempt at an open standard. The current-gen standand is Compute Shaders, which is integrated into Vulkan, Metal, and DX12. The wgpu (rust) and dawn (C++) libraries provide cross-platform access to this interface.  Unfortunately.. #SurveillanceCapitalism: An increasingly faster way to throw slow, inefficient technology into society to ensure insane capital flows to the already hyper wealthy and keeps them in power. @smallcircles @thomasfuchs A person can setup Lightning channels between their friends, colleagues, organic growers to do the fast stuff like fast payment, messaging etc. Yes, paired with SATs, Lightning eliminates spam so a proper back and forth convo costs **nothing** but bandwidth π @smallcircles @thomasfuchs There's symmetry here. We should not all jump on the bandwagon that something is bad because securing it is expensive. All good things require some energy to maintain, and the #bitcoin network will require half the energy of existing #banking. 1/2 @smallcircles @thomasfuchs NB: We call for #GlutPlug, a decentralised media-delivery network on fedi, based on media boosted/posted/faved. We think those action can result in participants serving such media to others over I2P. @dsfgs @smallcircles @thomasfuchs Even if at one time there might have been some value in bitcoin and blockchain, itβs been tainted forever by the culture. Itβs even bled into making decentralized sound dirtyβ¦Sorry but anything around it, even assuming best intentions, is a very hard sell. @dacig @smallcircles @thomasfuchs @dsfgs @smallcircles @thomasfuchs Sorry canβt keep feeding you: bitcoin and crypto culture BAD @dsfgs @smallcircles @thomasfuchs I want scams like bitcoin to progress slower than that. Like, to a full stop. You do you, as long as you mine with a pedal powered rig.  We're in a lull just now. I believe the New Hot Thing will arise from technologies such as Apple's Neural Engine. All this hoopla over LLMs - their limitations are coming into focus - they're very good for what they do. But there's a new neural network model, the Attention model: I started looking at this and my jaw dropped.  @tuban_muzuru @thomasfuchs The attention mechanism *is* what LLMs are built on. That paper you cited is what started LLM research. I'll be the first to say I don't know much. I worked with vision systems and QC/QA and I'm just now coming to terms with what it means. So I'm reading.  @thomasfuchs I think the retro computing world is where the interesting ideas are forming right now Cool tech stuff? Look at the database Larian Studio implemented to model the relationships between game mechanics for Baldur's Gate 3. That is cool vibes modern Prolog like genius @thomasfuchs You're on the cool stuff. Making tech work for people instead of billionaires bank accounts. And btw ... I riffed on your toot.   @thomasfuchs I don't think that it's ChatGPT's results that are wrong, but the person interpreting them. LLMs do a pretty good job of storing collocation probabilities! If only we could persuade people to see them that way... @libroraptor @thomasfuchs as a non-native English speaker, I actually use chatGPT a lot for collocation research. Other than that, I never trust anything it tells me, but rather use it to teach me relevant jargon or keywords on any given topic from which I'll start my real search. Pretty good tool IMO, if I don't overestimate it. I am not saying all LLM evangelists are thieves who are OK stealing artists work but I am saying that I dont care to find out 99 times out of 100  @thomasfuchs all of this work too if you replace database with βponzi schemeβ and still would be true.  @thomasfuchs while I understand the sarcasm, I find your take a bit too partial. LLMs may be lossy JPEGs for text, but JPEGs have their uses. Also, by your logic, mastodon is just a less efficient twitter and the fediverse is the 00's internet but harder to use. @muzzle @thomasfuchs LLMs are JPEG for all text in the world, and require a supercomputer to process  @muzzle @thomasfuchs An LLM isnβt a lossy JPEG for text. A lossy JPEG of a cat doesnβt show you a photo of a horseβs arse. And your other two deductions are non sequiturs.   @tessarakt @muzzle @thomasfuchs Usenet is topically organized, but you can cross post to the universe at least once. Itβs long form and asynchronous. Mastodon has visible photos, videos. Cats have their own day finally. Admins donβt post FAQs every month. You can chat in realtime with some of the people some of the time. @skry @muzzle @thomasfuchs Usenet articles are essentially message/rfc822, so images can be straightforwardly included in a multipart/related message. @muzzle @thomasfuchs problem is, people know that jpeg artifacts are a defect of the image not of reality, but hallucinations have already been taken seriously in important contexts. An attorney was found out giving them to their attorney to present to a judge as precedents - can't recall if it was giuliani or another of the minions of Tangerine the Foul and Terrible who did that. It's got its uses, but its misuses are many and dangerous. We haven't got an immune system for them yet. Mastodon lets me say things without Mark Zuckerberg smothering my face with a pillow. Energy efficiency for my voice is not a factor  @thomasfuchs crypto can also mean a secure way to communicate depending on what word comes after crypto   @thomasfuchs come to think of it LLMs are the βread equivalentβ of βEventual consistencyββ¦ which was a shitty way to store data anyway! @thomasfuchs They are slow because we don't have the necessary #technology to be faster :) Moreover, #LLM cannot be the only algorithm in the universe capable of some human-like #intelligence  @thomasfuchs ChatGPT can get back to me when they drop LLM for a Deep Understanding Model. I could add: Cloud: Someone else slower server you don't know where it is, to which you should entrust all of your business. Big Data: A huge and slow database of irrelevant data. You don't know their value. Amazon and Goole do. IoT: way to plug a WiFi board with a shitty firmware into anything pretending it is now smart and disrupting entire networks.  @thomasfuchs Don't forget Ethereum, which is a really slow and expensive way to both perform computations and store data. The "Ethereum World Computer" manages to combine 7000 full-sized computers all over the world so that their net computation power is less than that of a Raspberry Pi, and storing 1K in that system costs more than buying a 1TB disk.    ...  "human studies": years lost trying to grasp things, reading fictional books written by people who knew even less than the reader" "art": the only field when people can achieve a lot, under the condition to never specify in advance what they wanted to achieve ".  Heh, anyone that thinks this isn't troublingly accurate, doesn't actually understand what the tech does.
I explained NFTs to my normie friend's that were jumping on the hype train, and they were all "What. That's all it is" lol.  @thomasfuchs and we thought those 90s computer book store doorstop books full of websites were bad :blobcatsweats: @thomasfuchs @thomasfuchs None of this stuff is an actual databases. Databases are efficient and fast. @darmont @thomasfuchs I think you have a very limited definition of "database ". @WhyNotZoidberg @thomasfuchs You do have a wide one, I'd say. π I present to you: A database. Joke aside, "Efficient" and "fast" are not part of the definition of "database" at all. @WhyNotZoidberg You have an index? You are efficient! π @thomasfuchs  @thomasfuchs It annoys me how people assume "crypto" means "cryptocurrency" and not "cryptography" these days.  @thomasfuchs most of these things subsidized by piles of cheap money too... Just have to look at Uber to see what their future holds. And Uber doesn't have the massive upfront cost of some of these models  @thomasfuchs Glossary of Nvidia Marketing Team - very few actually useful applications for massive parallel computing  @thomasfuchs this ordering implies a hierarchy and the need for a horrifying crypto chatGPT hybrid @Lyle there's definitely people using LLMs to try to astrology their crypto buys and sells @thomasfuchs  @thomasfuchs I fear the definition of AI is a little too specific to the specifics of today, but maybe not.  @thomasfuchs Keep an eye out for the sneaky one: "Ledger", the one that means "someone designing a distributed system needed a canonical ordering of events, and they don't want anyone to describe what they made as a 'blockchain' because that would be super embarrassing"  @mcc @thomasfuchs you know what where are the likes of kpmg in all this block chain stuff. @thomasfuchs @thomasfuchs I love how LLM's are trying to solve complexity, that we've constructed - and could just as easily, drop. Of course not; We love to layer complexity, upon complexity - and if it turns out to be too complex, wrap it - wrap it good.  @thomasfuchs  Someone reposted on Facebook and it's making the rounds there: https://www.facebook.com/groups/1481260748650209/permalink/6810496815726549/?mibextid=Nif5oz  @thomasfuchs For years I was thinking about Software getting worse faster then hardware is getting better. When Nikolaus Wirth died, I found https://en.m.wikipedia.org/wiki/Wirth%27s_law and finally I felt understood.  @thomasfuchs only more like "AI: A buzzword, actual meaning is 'some magic', we don't have the slightest clue about how to create said magic. Currently used to indicate that something techy want's to be trendy" |
Gregory's Server
I really hope to see some actual cool tech stuff getting some time in the limelight after all the douchebro hype of the last 15 years.