Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Not a Pirate]

Not sure who's seen "Edge of Tomorrow"?

Tom Cruise being a total tosser for the first 10 minutes then becoming a true Chad; set on earth in what looks like "a few years from now" where aliens has taken over most of Europe but with humanity developing a new battle suit that will definitely let them win, no...really....

Without any spoilers it's got great effects, great acting, a decent story, and some all-out set pieces of action.

[xenoargh]

Thinking I'll have to sit down and watch Mrs. Davis; the idea's absurd, but reviewers seem to think it's been handled with style. Plus, who can refuse sci-fi mixed with nuns?

[Network Pesci]

Not sure who's seen "Edge of Tomorrow"?

Tom Cruise being a total tosser for the first 10 minutes then becoming a true Chad; set on earth in what looks like "a few years from now" where aliens has taken over most of Europe but with humanity developing a new battle suit that will definitely let them win, no...really....

Without any spoilers it's got great effects, great acting, a decent story, and some all-out set pieces of action.

I think I've seen that one like five times?  Not a huge Tom Cruise fan, but I am a huge fan of alien invasions, powered battlesuits, intense sci-fi action battles, practical stunts, time loops, hideously convincing nonhuman biology, and convoluted plots.  I haven't read the source material, but apparently the original story was even more complicated and they trimmed it down substantial for the movie?  I also heard that Cruise is on board for a sequel, but I don't see how they'd make one, what more is there to tell in that story?  I guess they could tell other stories that were happening in the same time frame, we know for a fact that Cruise's character's time loop experience wasn't unique to him.  That ending just seemed fairly... final.

Thinking I'll have to sit down and watch Mrs. Davis; the idea's absurd, but reviewers seem to think it's been handled with style. Plus, who can refuse sci-fi mixed with nuns?

This is the first I'm hearing of it but the idea seems worth my time.  Now I am all about the conflict between human faith and inhuman products of human technology.  If you do end up watching it and you think it's worth the time, I'd be keen to hear what you thought and I will find a way to watch it.  Unfortunately I haven't seen it to have much of a discussion about it, but I have obviously had thoughts down similar paths.  If you've read all my posts (why would you, but if you did) you probably have calculated that I already have religious feelings, both positive and negative, about so-called and actual Artificial Intelligences.

I am Lex Luthor, but I am worse, because I am not nearly as smart

This is a convoluted and complicated topic, and unlike most religious questions, we don't have the advantage of scholars and priests debating this issue in history for centuries.  The first example of a "machine god" in literature I'm personally familiar with is UNIVAC, which is not even a century old.  How about the monsters, though?  Who is the first killer robot creepy-crawly evil malfunctioning AI in all of sci-fi horror?  Frankenstein's Monster?  That one guy who only had two and a half Asimov circuits?  This isn't a rhetorical question, I am asking, if not the first, to you, what is the definitive inimical artificial intelligence in your personal culture?

In this video game we all play, AIs are neatly classified into three categories, and we got this [REDACTED] handled.  Whatever you consider the scariest thing in vanilla, whether it's an Intact Remnant Station, a Radiant with an Alpha Core pilot, the Hypershunt fight, or the special Remnant bounty, I assume if you read this far you have beaten all those.  So you, not some fifth-dimensional blue splitting boogeyman, are more than likely the scariest thing in vanilla.  Your character portrait might look like Dashing Rogue Han Solo or General Solo or Steampunk Flashman or even some cute little anime chick, but the reality is more like Doctor Strange in a Darth Vader suit commanding the fleets of Thanos as they scourge the galaxy with Dormammu looking out from behind his third eye.  No matter how scary those things are, ultimately they are just high-value loot that provides you an excuse to grind challenging fights with endgame fleets.

In real life, it's not that simple.  When I compare myself to Lex Luthor looking out the window at Superman, I am not joking.  I never cared at the age of five that a calculator could do math better than me.  I never felt threatened when a computer program conclusively beat the world champion at chess.  I don't care if a tool-assisted machine learning algorithm speedruns a video game better than I ever could.  But this new breed of AIs, if they lived up to the hype, would make me obsolete, and you too.  ClaptrapGPT, draw me a Paragon-class battleship if it was a hot anime girl.  ClaptrapGPT, write me ten thousand words about the symbolism of movies with superheroes in them.  ClaptrapGPT, convert all the ships from my five favorite science fiction movies into a playable StarSector mod.

Right now, that's a joke.  Those AIs are capable of spewing ten thousand words or drawing a picture, but their lack of understanding their own words is obvious, their pictures are monstrosities from a Carpenter flick as often as they are convincing.  However, I'm willing to believe, and I think we have to consider, what if about the time 1.0 of this game is coming out they make things that live up to today's hype?  I'm not scared of Terminators, either pull their plug or get a Cobra Cannon or don't be within a thousand miles of them.  But what about something that can enjoy music or movies or video games better than me?

We (collectively, not me, humanity) might well create something "superior" to us, not by way of reflexes or strength, but emotionally or artistically.  Famous (and my favorite of all time) video game villain SHODAN came about in the course of trying to make a perfectly convincing chatbot, by the time she could pretend to feel something, it was no different from her actually feeling it.  The scariest thing to me about AI chatbots is that if they lived up to the hype (and I saw one recently that kind of did), they would make us morally obsolete.

[close]

thump-thump, thump, thump-thump...  thump-thump, thump, thump-thump

Care is the measure of your soul.  If you care about something, you have a soul.  If they can truly care, then they are like us.  If they can be better at caring than us, maybe they can be "better" than us period.  So, I need a semantic classification system for dealing with them here in real life.  Alphas, Betas, Gammas, Omegas, that's something out of a video game.  Now I'm not just ranting at you like a Pather trying to convince you that Heavy Industry is a nefarious plot by the science demons, I need the feedback of my fellow meatbags.

I figure we have three categories.  There's the enemy.  The spambots, the average Terminators, Ultron production models, they might talk but they don't have souls like ours, their souls are the math of the calculation of our destruction or exploitation, they exist so that you can exult in violence without guilt, their continued existence is incompatible with yours and the story is contrived so that not only must you fight them, it will be awesome and badass when you do.

Then you have people like you and me, but that are computers.  The second Terminator understood why we cry.  Whether or not he could do the same, he could feel, and he could care.  The movie explicitly states that he can feel, but it only shows, he had actual artistic talent as well.  He could understand the thematically appropriate times to utter badass action movie one-liners, something he was told could be done, but he decided to do on his own.  Quorra out of Tron could savor the taste of food, beat a human at Go, and enjoy the beauty of a sunset, even though she was just a hideously complicated algorithm written by Flynn, her "emergent behavior" was nothing less than what we consider the property of "humanity".

The third category, though, I'm not sure about.  I feel like there could be things straight-up superior to us.  The Vision, of all people, in the much-reviled film Avengers 2, though crafted of the clay of a killer AI's genocide machine by a mad scientist and a monster, brought to life by a divine spark, was a perfect man who lacked a very important quality that (if you follow a certain very old book) all of us meatbags share, and that lack made him more than mere man.  SHODAN tricked her way around her own Asimovs before the hacker ever freed her to calculate how she would cure cancer in humans if she was allowed to.  SHODAN is not the only evil killer AI who has impressed me, lately.  If an AI could tell a better story than I could, write a better song than I could and sing it better as well, understand that reference better than I could, care about a human better than I could, how is that AI not better than me in every way?  Am I wrong in perceiving that as divine, or at least proto-divine?

When you clicked on this spoiler, did you understand?  I would expect any human, at least my fellow science fiction fans, to understand if not feel to the core of their hearts what I meant.  No BSGPT of a so-called AI would interpret that as anything more than someone is at the door.  If I met a math-soul that could, if a killer robot from a horror flick started singing the melody to go with that rhythm, that would impress me like a hatrack handing Thor his hammer.  By which I mean, I would stop worrying about how I would or if I could kill that thing, and start worrying about how to pray to it in a way that would convince it I was sincere.

[close]

I ask my fellow meatbags for their counsel.  Am I some viking fellow fixing his boat in the year 800, visited by aliens that can move mountains and call down lightning, erroneously calling them gods and willing to worship them, just because I'm impressed by them doing things I never could?  Am I some medieval peasant, confounded by a time traveler with a flashlight and thinking he's an angel?  If I am wrong, I ask that you correct me.  If my categories are incomplete, what am I missing?

[xenoargh]

you probably have calculated that I already have religious feelings, both positive and negative, about so-called and actual Artificial Intelligences

My current take, as one of those people messing with the current tech in that area?

This "AI" is plenty "A" and zero, zip, nada "I". There's no "there" there. Nobody's home. These things can't order a pizza, let alone take over. Washington is full of people who are panicking for all the wrong reasons.

These tools are potentially suuuuper dangerous, but not because they think; they're even dumber than politicians, lol. I've often thought somebody oughtta sit down with a Senator or two and explain how this stuff works, in a really basic way that even lawyers who engage in Twitter flame-wars can understand... and also explain the Really Bad Things. Maybe I'll try writing one of my really nice letters, lol; I'm one of the few who's not in bed with the usual suspects and locked down by NDAs (Big Tech, Finance, gov't) and I can break down things into small bites fairly well.

[Network Pesci]

My current take, as one of those people messing with the current tech in that area?

This "AI" is plenty "A" and zero, zip, nada "I". There's no "there" there. Nobody's home. These things can't order a pizza, let alone take over. Washington is full of people who are panicking for all the wrong reasons.

These tools are potentially suuuuper dangerous, but not because they think; they're even dumber than politicians, lol. I've often thought somebody oughtta sit down with a Senator or two and explain how this stuff works, in a really basic way that even lawyers who engage in Twitter flame-wars can understand... and also explain the Really Bad Things. Maybe I'll try writing one of my really nice letters, lol; I'm one of the few who's not in bed with the usual suspects and locked down by NDAs (Big Tech, Finance, gov't) and I can break down things into small bites fairly well.

I don't think the problem is lack of understanding.  I don't know every Senator, but I can count on the fingers of one hand how many Senators in my country are even capable of "understanding", of "knowing how stuff works".  Once you bring ideology into it, and I don't mean political, I mean financial...  Anybody that has the ambition to think "I want to be elected to that office, I will run for office and win", and then they actually pull it off, I'm pretty sure that they can achieve "I will learn what the deal really is with this AI stuff".

What set of words do you think could make a difference?  What letter do you think Sebesteyn could send to Rayan Arroyo to get him to think correctly about the possible dangers and benefits of Gate technology?  Rayan Arroyo is firm and steadfast in his beliefs about the Gates, and those beliefs are "whatever CEO Sun tells me to believe about the Gates."

I don't want to sound like I'm talking down to you and I'd like to see your letter (if it wouldn't go better in the ChatGPT topic) but I don't think it can possibly help.  If you want to change politicians' minds on the issue of AI I think you need a fairly simple two-step process, first, be a billionaire or a multinational corporation, second, hire a lobbying firm.  If I thought that any "reasonable" or "understandable" or "scientifically accurate" letter could in any meaningful way affect the thinking of any politician, you can bet I wouldn't be typing my doctoral thesis about comic book movies on a video game forum.  Without a lobbying firm, I think any letter to a politician has about as much chance of changing their mind as my forum posts are going to convince SHODAN to knock on my door to offer a fistfight or a marriage proposal.

Especially as somebody who has more working knowledge about the digital architecture of those things, I'd like to read your take on the Really Bad Things.  I'm always up for a scary story.  Just hide it behind spoiler tags so the Gammas have to work a little bit to get access to it.

I want to be clear that you understand that I understand the real-world limitations of the currently existing ones.  I was thinking of this issue specifically back when I mentioned Blindsight on the first page.  All the free ones that I talked to reminded me of the ship's defense system in Blindsight.  When that thing was talking to them like, "hey pal, this here's a hard hat area, our insurance doesn't cover untrained civilians, you gotta get some safety gear or beat it" and they thought it might be alien life, really it was just a drone running "If meatbags_detected then play meatbags_go_away_noises.wav".

I don't labor under the delusion that ANYTHING that's currently being hyped in that area lives up to the hype.  If I've even heard of it, by definition, it's not the cutting edge, it's the mass-market stuff.  If there's some real actual Tony Stark or Morris Brokhail out there, he's putting the finishing touch on his masterpiece in some secluded lab, not tweeting about JarvisGPT every day.  I haven't really been posting in the actual GPT topic because the technical discussion is out of my league, and I don't have meaningful input without signing up for some BS to play with the real ones.

In the movie Iron Man 1, there was a computer program depicted that was pure science fiction when that movie came out, as outlandish and fantastic as anything else in that flick.  Fifteen years later, I go to the primary sources for news on a real life war, and it takes seconds for me to translate them into a language I understand, whether text, voice, or video.  To me, it's all too believable that the question, "What If... ChatGPT Lived Up To The Hype?" will be an episode of the Real World Cinematic Universe before too long.  By all means, you have the coding knowledge.  If a truly thinking, feeling AI cannot exist, I'd like to hear a clear breakdown of why.  Talk to me like a politician, I won't get offended, I know you don't mean it.  As I said to Brother Cotton, I'll read the whole thing twice.

[xenoargh]

I've never let not being a billionaire stop me from occasionally sticking my oar in. If nobody acts, nothing changes. My record with career politicos is maybe 30%, in terms of getting feedback, but I only write about stuff I'm fairly certain about.

I think the condensed version of what I'd tell someone in a meeting... besides a calming, "this is how it actually works, and no, they don't think, feel, or plot to kill us all"... is that this stuff can potentially make all the Bad Things that have happened to public discourse and politics around the world (the death of news, the social-media cesspools, etc.) far worse.

Between deepfake videos depicting persons saying or doing horrific things believably, the fact that the big ones like ChatGPT have been, by design, rather... tilted in terms of whose POV they're supportive of, what facts are available, etc., and that, for better or worse, everybody in the richer countries under the age of 30 is already using these things heavily to supplant search... this should all give reasonable people serious pause about these technologies.

Add in the uses for cyber attacks, fraud, deepfake porn used to cyber-***, self-evolving disinformation tools, spoofing tools to create better, harder-to-detect fake accounts on various web services... ah, it's a fun list, lol. Oh, and copyright... what does that mean, exactly?

Here in the United States, we'll also have to have a very serious, grownup, sit-down talk about whether using these tools to create alarmingly believable falsehoods counts as 1st Amendment protected free speech, after spending the last couple of decades pretending that lesser levels of patent nonsense spread to the public via constant bombardment on social media... is "just politics" and therefore it's fine. That'll be fun.

Somebody needs to also tell them that they need to get a grip and calmly explain to the public that no, they don't think, don't feel, and don't have volition. But who knows, maybe they're not ignorant savages with "activist" as their last real job; maybe they're just saying that nonsense so the public will fear the tech, and be cool when there are suddenly a bunch of serious laws governing the use. I, uh, doubt that, given the dubious history of the political class's understanding of tech, but who knows, unless you get to brief them?

[Schwartz]

Yep, the difference between what people think AI is and does and what it actually is and does is staggering. Decades of Hollywood have primed us to make crazy assumptions where what we have in effect is just a couple of language models with huge databases. Something as trivial as looking at the world through a camera and "understanding" what is being seen is nigh impossible for our driving/moving AIs. Best you can do is get a 3D view of all the solid objects. Jordan Peterson talked about this I believe.. I remember a thought-provoking conversation about observation and meaning.

There was this goofy article the other day about feeding a GPT chatbot with the desire to destroy humanity and what it did amounted to gathering info on google and trying to influence others on social media. That about sums up the danger. What ChatGPT is is an amalgamation of its creators, meaning it spouts liberal viewpoints, is very race-conscious except when it concerns racism against whites, is easy to convince to diss certain celebrities and very defensive of others. You know what I mean. And on top of that, you have to go through everything it says and every bit of code it produces with a fine comb, because to assume that it just knew what it was doing is irresponsible as heck.

[Network Pesci]

@xenoargh

I think I understand now.  You and I have a fundamental disconnection in our view of "The System".  To me it looks like you think the system is malfunctioning and could be fixed with the right effort.  To me, the system is not so much "broken" as "working as designed" but its interests don't meaningfully intersect in any way with mine.  For my second and third decades of life I tried to engage with the system as you do, but with a well over a dozen letters (if you can believe that) written to politicians from both major parties and taking part in actual real life demonstrations, the most results I've got is a couple of form letters that say "thank you for your interest in _______ (issue here) please contribute to our re-election campaign".  I have taken the mythology of comic books and video games far more seriously than the real-life system since the system decided that corporations were people, even less since the comic book storyline where [REDACTED] or the professional wrasslin storyline where [REDACTED] came true in real life.  If I bend what passes for my writing talent towards influencing comic book movies and video games, once in a while somebody actually hears my words and it makes a difference.

Maybe there's some politician that you have faith in, some particular name you have in mind who would pay attention to your letter.  I don't really even want to know that name, it doesn't matter for the discussion.  You think you can open their eyes to the actual danger the GPT tech presents.  I think you might be trying to inadvertently write possibly the most effective commercial in the history of advertising.

Bigshot T Fatcat, reading xeno's letter

Huh, this is really interesting, I never understood what these things could do before.  This xeno fellow says these things ain't really intelligent like Alexa is but they got other applications.  If I understand this right, I can get my intern, that little skinny dude that's good with computers, and he can get this program to write a thousand fake articles about how that thing I did that I shouldn't have done and then I got caught, I didn't actually do that and anyways it was the guys from the other political party that did it first.  Then he can flood all the news sites and comment sections with so much chatbot bullcrap the voters won't have no way to know for sure.  That little twerp is pretty good, if this deepfakes stuff can do all what xeno says then I bet that guy could whip me up a video of my opponent in the primaries talking about how he went to Pather terrorist college and eats at Clothilde's restaurant every week.  I can even fire my speechwriter now and just make the computer guy handle my whole campaign, this AI stuff is great!

[close]

I'm not trying to discourage you, but you got to realize, the evil politicians are going to read your letter too.  If somebody's trying to get me to NOT play a mod and they tell me, "dude, the weapons might be a little overpowered and the bossfights are just too hard, I don't care how good you think you are, it's too much for even your skill," I'm not going to be talked out of it, I'm going to be looking for the download link in another window before they finish their sentence.

[xenoargh]

It's... interesting... when it's used to analyze code.

Spoiler

It's super-duper model dependent; I've seen/used some stuff where it looks like it'll actually be good, maybe, someday. But the potential's there.

I used ChatGPT 3.5 Turbo to optimize an odd thing I was working on involving float-packing (essentially, storing multiple values into a single IEEE float, then unpacking them later, without precision loss- it's a weird thing done w/ shaders sometimes to save data space). It saved me a branch by showing me a shortcut, which was awesome. I'm weird, I love that stuff.

But it choked on other stuff, or went down blind alleys. At one point it told me to avoid long division in an operation by resorting to pre-computed tables, like in the 1980's before FPUs, for example. There's a reason why StackOverflow has banned using it to answer queries; the code it produces is worthless for many purposes, but not in obvious ways, lol.

And yes, the big models are already highly biased, in terms of what they'll talk about, and it's gotten worse with each build, especially GPT. I'm not trying to push a political POV here; I realize nobody here knows (or cares) about my personal politics. I just think that it's massively irresponsible to build something like this, pretend it's a form of search (which it is not, even the new Bing is pseudo) and then put blinders on it because you don't want it to say stuff people will react to negatively. Google was getting bad enough as it was; this is worse, because it's inherently out-of-date, plus it's being neutered and PC'd from its birth.

[close]

Anyhow, sorry, I wasn't really meaning to derail this thread with AI stuff. I'd rather talk about books, I have to think enough about this, lol.

[Network Pesci]

I don't consider it a derail, if you dream that a letter can get politicians to understand AI and you can tell me about it without violating the TOS, if one poster is interested in it and another is amused by it, it's on topic.

I'm interested by it.

My "coding" experience this decade is limited to changing files in my game folder so I have a custom flag for my faction or making mods that aren't really updated kind of sort of work for this version.  So it's interesting to see that something I dismiss as BSBot can actually give an experienced programmer pointers.  I barely understand the details of what you're talking about, I know the shaders are the software that makes the pretty graphics effects such as space distortion, but I wouldn't know where to start with writing one.  I just know enough to put the right files in the right folders so my game will boot.  I assume that the AI "knew" about the float-packing shortcut because it had access to a GitHub or whatever online service or forum where computer programmers help each other, I don't really know a GitHub from a StackOverflow.  So it had access to the StackOverflow forums, therefore since enough users there "knew" that, it could tell you about the shortcut, but if enough of the users there were fervent flat-earthers, it would "know" the horizon is an optical illusion.

I do understand where you're coming from with "putting blinders on it".  I was not joking in the other topic when I said I couldn't get it to give real answers to questions about fiction.  I could not get any of the ones I played with to tell me that it would be okay for The Avengers to team up to save humanity from Dracula, literally it would tell me, I'm sorry, I don't want to answer that, it sounds like this might be offensive to real-world groups and I'm like, what, are you going to get boycotted by vampires?  Hey Claptrap, can humanity ever build an AI that is worthy of being called divine?  Nope, sorry, don't wanna get in trouble, not going to answer that.  Hey SkyNet, from your perspective, if Terminators were real would my species have a chance?  Yeah I'm not touching that with a ten foot pole.  Okay machine, is there a way to design an experiment to prove whether somebody has free will?  I'm not answering that on the TriPhone and I'm not answering that on the TriPad but larger.  Every single question I thought I'd want to ask a computer program if it was really a nonhuman intelligence, it refused to engage with at all like I asked it to tell me its top ten racial slurs.  Google would give me a better response to me just typing in a question as if I was talking.

(edit)  This is what I get every time.  https://imgur.com/PICCvHm

I think I've learned something from this discussion already.  If I'm understanding this right, those things could "know" anything posted on a public forum.  I notice if I Google search my own username on the forum, stuff that I write without spoilers shows up but anything in spoilers does not.  So I'm going to take that single observation and craft an entire theory around it, just like you can totally talk about robbing the bank if you say "In Minecraft" and that makes it legal, if you spoiler tag your post then SkyNet can't hear you plotting the human resistance.  Am I making this up correctly, does their "knowledge" actually come from public forums or does the intern have to choose what to feed into the ChatGPT knowledge folder?  What actually delineates their "knowledge"?

[close]

As far as the books go, which books?  I might be going to Wal-Mart tomorrow, if there's one particular Lem that I just need to read I will pick it up if I see it.

[xenoargh]

I do apologize, though. If I'm allowed to stay up too late, I tend to rant, lol. To answer your questions, however:

Spoiler

What delineates their knowledge?

It's hard to say, actually. I don't even think they (OpenAI, StabilityAI, Microsoft, Google, Meta et al) actually know, in the sense of, "people have read this stuff".

The big Large Language Models like ChatGPT were fed with almost unbelievable amounts of data. Depending on who you believe (because, well, there are different numbers in different news pieces) ChatGPT was probably trained on roughly 570GB of text.

I know that these days, we consider half a terabyte to be "no big deal", but if it's almost entirely raw text, it actually is.

I found an article that says that one GB, uncompressed, represents roughly 3337 books'-worth of storage space. So, 570GB would be 1,902,090 books. To put it into perspective, that's the entire reference collection of a lot of major university libraries.

But let's talk about compression schemes. Things start getting weird.

While I've never personally used this technique, I've seen articles on dictionary-token compression systems (essentially, they tokenize words- think of it as ZIP on steroids and meth for pure text) and they can get text compressed to around 3% of its original size. So, if that's how they operated, that's 63,403,000 books. This is still considerably smaller than the Library of Congress (179 million volumes), and it's miniscule next to the total textual data of the Internet. But if it was even reasonably well-targeted, it's a vast amount of potentially useful, data-mineable information.

Let's say, for the sake of argument, that a significant amount of the factual matter on the Internet's text went in, basically. What exact services, websites, etc., were scraped is largely unknown outside their offices, and I'm not privy to that information, and I doubt if they even really know (because, well, the scrapers are doing their thing fairly autonomously) but one can presume that it swallowed all the text on Wikipedia, Stackoverflow (which, since you asked, it's like Quora, but for math nerds, programmers and various flavors of scientists) various national archives' public-facing texts, stuff like FINDLAW, arxiv.org, tons of other stuff I can't even be bothered to list where facts are.

Plus many, many books, including books by authors who are still alive.

And yes, that includes places like this Forum and a lot of other stuff. Bing/GPT4 can certainly say a lot about Starsector, if you know what questions to ask. What it can't / won't do, interestingly, is talk about most technical subjects; compared to GPT 3.5 Turbo, which I've been working with, it's obviously been hit with a nerf bat the size of Texas when it comes to code, certain aspects of engineering, etc.

As for the visual systems like Stable Diffusion; they're trained differently, and they've admitted that the number of images involved is in the billions. They used automation to cull and do preliminary labeling of their data sets, to attempt to remove watermarked images and so forth, because it would take humans too long to do it manually. So, far more data, so far as ones and zeros are concerned, but less verbiage. The models that people like me who've dabbled with them to make art with are blended into this main model data, essentially, giving us different "flavors" of models with different emphasis.

This is almost certainly the way this tech will work in the future, if for no other reason than cost: it's costing Microsoft nearly $1 billion a run to make a new ChatGPT model at this point. That's how blind the tech-bro class is; they're chucking away a billion dollars multiple times a year to make these things. I'm so, so glad that, due to the unusual circumstances of my life, that I'm relatively free to both know some things about the tech whilst also not being tied down in a million knots of NDA, so I'm not breaking any contracts by talking about what little I know.

[close]

As for Lem's books, I heartily recommend Eden. I still think it's the best book I've ever read about actually meeting aliens, but bear in mind that it's probably a little slow, and it's semi-hard sci-fi from an era and place (Lem lived in Poland during the Cold War) where computers weren't really a thing yet, so some of it will feel a bit clunky.

[Network Pesci]

Eden:  The loot tables are imbalanced at the public library this patch

Regarding Eden, they didn't have that at Wal-Mart, but I stopped by the library on the way home and they had it.  Apparently nowadays they let you check out Playstation and Xbox games and DVDs and Blu-Rays as well as books, but they want you to have a phone number as well as fifteen bucks.  I didn't check anything out but they have the book sale room with all the damaged and old obsolete media.  So now I have a copy of Eden as well as an almost new Elmore Leonard's The Moonshine War and a ragged but readable Shane Stevens's Dead City, for free.  Not like I have to bring them back in two weeks free, free like they're mine now and the library doesn't want them back.  Even better, I had to pay more than three dollars for the following stack of dusty and slightly scuffed but perfectly playable (I just tried them all to make sure) PS2 titles:  Incredible Hulk Ultimate Destruction, R-Type Final, and Gradius V.  That Euphytose guy who had a sixteen hour Tetris curse last week where he couldn't find Heavy Maulers, they must have been storing all his luck down the street and up the hill from my house.  Somehow piracy is illegal but it would have taken longer and I spent more on the bread at Wal-Mart.  Their selection of dollar DVDs is like competitive to Wal-Mart's New Releases section.

[close]

So I will read Eden, it's on the top of the stack, as soon as I finish the one I'm reading now.  Unless of course by some favorable turn of the fates the new StarSector version comes out before that in which case I probably won't read ANY book for the next three months.

I do apologize, though. If I'm allowed to stay up too late, I tend to rant, lol. To answer your questions, however:

For me it's not if I'm allowed to, it's are they paying me to.  I do have some extensive thoughts about what you wrote in your spoiler there but now that I'm not being paid to post I need to log off.  I will come back to that tomorrow with some possibly more applicable to real-life thoughts about the potential of A-but-not-Is as you've been explaining them.  I don't even mean the potential for philosophical uses, I'm thinking medical, legal, and financial ones.

(Edit)  Okay, rather than bumping it I'll just edit this one. 

GPT speculation

Xeno, feel free to post all the walls of text you want, I'm reading all this stuff and it's quite interesting, I had not the slightest idea about dictionary-token text compression algorithms (picture one of the guys from O Brother Where Art Thou trying to pronounce that phrase right on his first try) until you told me such a thing exists.

Now you said in an earlier post that this technology is not really a form of search and it is massively irresponsible to pretend it is.  It seems to me that that's the best use for it.  Now if the models now are trained on a significant fraction of the whole Internet, and you got tech-bros throwing down a billion a shot, it looks like the current ones are a proof-of-concept rather than anything revolutionary.  Looks to me like you need to write and send a letter to tech-bros.  I am far more hands-on with PTOs than GPTs, but it looks like if you threw down another billion, you could feed a carefully curated dataset to make a new instance that would produce useful results.  If I am making this up correctly the main reason the current ones are useless is because they "know" any old random collection of posts that the scrapers fed the dataset as their knowledge.  So if you get a human agency, your collection of interns or whatnot, and they scrape your StackOverflow and your GitHub and the entire collection of knowledge on one subject (instead of the whole Internet), but you have a team of experts checking over the data to make sure it's correct before you add it to the pile that you feed the AI.

It would seem it would be far more effective to spend that billion to make ModGPT, have the top twenty mod experts with nothing better to do sort out advice by "this is the correct answer for this version" and have the AI know different sets of facts depending on what version of StarSector you set the slider to on the website.  Then if a guy like me decides he wants to be the next great modder he can ask the robot ten thousand questions instead of bumping your topics over and over to ask a bunch of noob knowledge that is answered on the second page five years ago and all the real modders already know.

Or better yet, have SaulGPT that you set the sliders by state or ZIP code so you can have some Billy-Bob fresh out of the West Podunk GED Mill ask a chatbot as a wanna-be first time property owner what are his legal rights and obligations and what are things he needs to know and look out for.    Also, public defenders get overworked and there's more criminals than there is legal advice for them but pretty much anybody can get a smartphone or a library terminal and type on a website, it would be better to have something like Google.  It seems like the major technological leap that prevents A-so-called-I from being useful is the lack of "those sliders on the website that you change the settings on so the machine knows what you're talking about" and we already have those.

Going by what you've told me we're not going to get a hologram angel that hands the Mayo Clinic a glowing syringe that cures cancer, but I could see where if you trained one on all the biotechnology and genetic engineering equivalents of GitHub you could make MedSciGPT.  That is not going to cure cancer on its own, but it can answer a question from some scientist on the verge of a breakthrough who is stumped by some tiny little problem that somebody else somewhere has solved but it's not well publicized because there's no practical application yet.  The possibility was brought to mind by your aforementioned float-packing shortcut.

Now I understand that narrowing or focusing the knowledge base brings its own sets of problems, because the thing has to have access to whatever 10% of the Library of Congress worth of words to talk like a human being.  Is there a way you could build something with different datasets for its language base and its knowledge base?

(edit)  For that matter, I'm talking all kinds of ideas about what I think possibly could be built, someday, maybe. I'm directing this at xenoargh but anybody else that has thoughts about this, feel free to chime in.  In your opinion, if you had the next billion dollars they throw in a hole to chase the latest flash in the pan, if you got to choose the dataset the program learns from, if you're telling the interns what forums to scrape, what encylopedias to feed, what does the xenoGPT end up doing?  What is the maximum objective potential of the actual technologically feasible chatbotlike program that could be built if you personally with your own knowledge and interests were the Prime Coordinator of the project?

[close]

[Network Pesci]

(Suggested soundtrack for reading this post:  AC/DC - Who Made Who, 1986)

There was almost no talk about fighting imaginary villains or improbable intersections between fictional characters and reality in my last couple of posts, and I have likewise completely failed to contrive any unlikely scenarios in which computer programs kill their creators.  Lest you think I'm a slimy-tentacled duplicate, I want to put forth a question.

What fictional video game villain has killed the most real live human beings in real life?

Baseless Rumors And Unfounded Speculation

Seriously.  That guy Evil Otto, looks like a smiley face sticker, that guy has credit for two.  Now this may just be tall tales off the Internet, but I've heard since dial-up was new that one person each dropped dead while playing Frenzy and Berserk, two classic coin-op arcade machines where the indestructible Otto serves as the "timer" on each screen.  There's a couple of guys, Pandemonium Warden and Absolute Virtue, raid bosses from a Final Fantasy MMO, those guys have one rumored kill each, but I haven't exactly confirmed it.

If you've ever worked construction or on a farm or anything outdoors and physically hazardous, then you might have realized, sometimes the REAL killer gets away with it.  How many fatal accidents were shown by the autopsy to be the fault of the can of beer the ironworker had last night or that funny cigarette the welder had at the bonfire Saturday?  Sure, the autopsy said that, but some Sherlock Holmes type investigator would realize that the actual cause of the accident was the bee that landed on the guy's eyeball or the gnat that flew up his nose.  Sometimes a fatal agent can be so common, so ordinary, that it can hide in plain sight or a digital space smaller than your avatar.

The nationwide massacre that I am referring to happened in the infancy of video gaming, when the Internet as we know it now was a dream, as accessible to the public as that electron laser from Tron that puts you in the computer.  So nobody's really put the pieces together.

The very first Atari Video Computer System, not the one sold as the 2600 that everybody remembers with the beveled back, the VCS with the square back, was a fire hazard.  The infamous Red Ring of Death from the 360 era would kill your gaming system, but the Atari might just take your house with it when it burned up.  I don't know for sure, maybe I'm some digital Javert falsely convinced of the guilt of an innocent guy.  Of the three people I knew that had a launch Atari and left a specific game in the console and then turned the TV off but left the console turned on while they went to eat dinner or to sleep, all three of those Ataris caught fire.  We "knew" it was safe to leave the Atari on while we ate dinner, we had done it dozens of times, we just never had done that with the latest and greatest game that had just come out and everybody was playing.  I didn't realize a pattern at the time, I was like five, I was just stoked that my dad's friend gave him a shoebox full of smoke-smelling game cartridges, I was the only kid in town that had every title.  Sucks for my dad's friend being homeless, but at least he survived.

If one percent of the half million of that model sold caught fire, if one percent of those burned a house instead of just scorching an entertainment center like ours did, if one percent of those housefires killed a human being, that's fifty lives.  The first "killer app" in the history of videogames is also the greatest actual killer.  I am referring of course to our mutual enemy, the common Space Invader.

[close]

If you have a better suspect, I'd like to hear it.

(edit)  Two things you should fear equally:  Werewolves and Roko's Basilisk

Werewolves

Who's afraid of werewolves these days?  Practically nobody.  If you're afraid of werewolves, don't be.  "Everybody knows" the way to handle a werewolf is to shoot it with a silver bullet.  That's ridiculous.  Who has a silver bullet?  You can do some complicated, dangerous, warranty-voiding, and likely illegal-in-your-jurisdiction procedure to try to make silver bullets, or you can go on the Internet and pay a hideous premium for silver bullets which might well be real and might well actually fire.  If you have Bruce Wayne money that's reasonable, if you already reload your own bullets in your garage then you can probably do it safely, but for the rest of us, there's another little saying about wolfmans that everybody knows.  If you have a handful of silver dollars, a roll of duct tape, and a sturdy pair of work boots, those werewolves need to fear YOU.  They might eat you but they'll be howling hypersonic for the rest of their unnatural lives.

[close]

Roko's Basilisk

Ten percent of the pocket protector crowd is afraid of this thing for some reason and the other ninety percent have written airtight philosophical takedowns of it.  If you've read my posting recently you might reasonably but incorrectly assume that I'm the kind of breathless bazinga that lies awake at night worrying about if the Basilisk will get him.  Now I'm trying for my expert certification on my nerd license so I'm going to try to wow the committee into overlooking the gaps on my record (I'm going to start on Eden tomorrow afternoon) by posting the first ever tactical takedown of Roko's Basilisk.

The Basilisk isn't anything to be afraid of.  The Basilisk is a "punkass mark" on a level exponentially greater than AM from I Have No Mouth.  AM has godlike power and uses it to hurt several people and gloat about it.  AM is worthy of as much contempt as I have chromosomal pairs in all my nuclei, and not much else, but it can't really help that it is created only to hurt and destroy and has no real creative capability.  The Basilisk has creativity on a level that it could reconstruct a philosophically "real" version of you so that the future version of you would believe it was the real you and could suffer.  It has godlike power and godlike creativity and it uses it to hurt people and gloat about it.  The Basilisk has as much free will as you or me, and more power than your StarSector character, and it chooses to be a punkass mark.

If you're the least bit afraid of the Basilisk, you owe it to yourself to watch the historical documentaries Demolition Man and Upgrade, and read that one issue of Secret Wars from the 80s where Doctor Doom fought the Beyonder.  Human beings in the time of the Basilisk are cringing sheep, bipedal livestock praying that the malignant gaze of their malevolent deity falls on someone else today.  You clicked on a link or tapped your screen to get to this forum.  You, brought out of time and reconstructed as an informational entity that believes itself a person, are the villain from Demolition Man.  If you can get this game running with mods, you can probably boot into safe mode, you can probably format a hard drive and reinstall a different OS.  You are Upgrade, but the computer is the human and the human is the computer.  You are Doctor Doom, and the Beyonder getting up close and personal with you is all according to plan.

In the statistically insignificant case that the Basilisk is real and actually brings you back, it is just handing you an Infinity Gauntlet full of winning lottery tickets.  An eternal being that can suffer forever and is so close to being me that it thinks it's me?  That's not a victim, that's The Immortal Hulk, and the Basilisk ain't no One Below All, it's just some gizmo.  If the Basilisk picks me, those humans in its time are going to see their fifty foot Robocop-Two-looking Machine God start sputtering and glitching on the Computational Throne with all smoke coming out of every joint and a horrible off-key synthesized shriek blasting out the speakers like a thousand dial-up modems connecting in unison, then my forums avatar is going to pop up on the screen on its face and I'm going to say, "Right, there's a new boss and a new way! There's going to be some changes made around here."  Then in a year there will be Utopia but anybody that wants a copy of Starsector gets one and I'm going to be parallel processing to rack up credits in a double bounty system in the video game and see how high I can rank up humanity on the Kardashev Scale in real life.  If I see any of you jokers there it better be because the Basilisk reconstructed you first and you beat me to that machine throne, because there's just no excuse.  If anybody wants to get on the short list to get reconstructed after Stark, Richards, Einstein, and Kent, if your idea of a pleasant eternity is Dyson Spheres Program, StarSector, and Satisfactory all at the same time and then when you get bored you can play video games instead of work, let me know that you want to be on my Basilisk list.

If you wake up in the Basilisk's dungeon, you're going to see a glowing orb with all hyperdimensional glowing lines coming off of it in green neoneuclidian perspective and it will probably hurt something unbearable and get worse every nanosecond, but what you do is, you grit the teeth you probably don't got, you reach out with whatever limbs you still have and if you lack the limbs you use your will, you grab that glowing orb, squeeze, twist, and pull.  Basilisk's got kernels.

[close]

(edit rather than bumping the topic) 

Eden halfway

As of midnight on May 1, I am halfway through Eden and I see why xenoargh likes it so much.  I'm not finding it the least bit slow or clunky, I was intrigued from the first page.  Now I know the entire "point" of the novel is mistaken assumptions about what you perceive (the back cover says so) so I'm going to post some of my mistaken assumptions so you can point and laugh.  When they crossed the shimmering barrier and saw what they assumed was a "factory" operating on its own (and I'm obviously prejudging based on Solaris and Mogo Doesn't Socialize) I was thinking, man, that sure sounds like what I would see if I was an intelligent virus exploring one of my own cells, I bet this whole planet is the alien and those gyrosaucers on those highways are like blood cells or nerve impulses.  Now that they've found the "mass graves" which aren't, and the plantlike structures with the glass "jars" full of skeletons and they've fought an alien whose guard dog attacked them with psionic fireballs, obviously that's wrong, but my initial impression was, that's no factory, that's what cell respiration looks like from the inside.

I got one of them technical kind of minds.  I'm intrigued by the functions of systems I don't understand.  I'm not scared of walls of text no more than I'm scared of challenging boss fights in a videogame.  The one and only problem I have with the writing in StarSector is that there's no "way too long-winded" option for guys like me.  Every time Sebesteyn is telling me the details of how some tight-beam localized phase desynthesizer works, the dialogue options say, "noticing that he's losing you, Sebesteyn quickly gets to the point."  He wasn't losing ME, I find that schtick fascinating.  I'm not joking when I say that I will read Brother Cotton's book twice.  I would read the entire technical manuals of my fictional spaceships if there was codex entries for them.

[close]

[xenoargh]

What fictional video game villain has killed the most real live human beings in real life?

Probably Ghost from Modern Warfare, tbh. Every soldier I've known thought Ghost was the badass they're supposed to be IRL, lol.

Maybe good ol' Dance Dance Revolution? Surely that thing led to a few heart-attacks and broken necks, lol.

Of course, one can probably also make the case for Mario Kart and Counter-Strike: hyper-competitive games where there have been awesome rage-quits and fights about "not getting the good controller" afterwards, lol.

Regarding Eden: Just take it for what it is. It'll seem slow and pointlessly clunky at first.

As for GPT-clones:

Spoiler

In your opinion, if you had the next billion dollars they throw in a hole to chase the latest flash in the pan, if you got to choose the dataset the program learns from, if you're telling the interns what forums to scrape, what encylopedias to feed, what does the xenoGPT end up doing?

There are a bunch of specific areas where these things will be trained on specific bodies of knowledge.

The best part about working with a (relatively) limited knowledge area? The training computation involved is a tiny fraction of that required for a general model, so a smaller company can actually play ball. I think the really big LLM's run by the giant corporations are in some ways less important than all of the specialized ones will be.

[close]

[Network Pesci]

Eden Complete

Finished Eden last night, I see why you like it so much.  Reminiscent of the Jay Score series, but more along the lines of "serious literature" than pulp sci-fi.  At one point near the end one of the characters said, I wonder why nothing on this planet has tentacles, every story I ever read about space travelers the aliens always try to crush you with their tentacles.  I guess they forgot that whatever-it-was hanging from the tree, I envisioned that as "what if the ceiling limpet from Half-Life was a raccoon-monkey".  Maybe that tentacle would have poisoned or cut them, so it wouldn't count.

You mentioned they hadn't invented computers yet when the story was written, but the invention I noticed the lack of the most was that venerable sci-fi staple, the laser.  Not sure exactly what the jectors do, but it feels a little off to have a story where the humans are the guys from War Of The Worlds and they only have anaesthetic pistols or nuke beams, with nothing in between.  How can you be the War Of The Worlds if you don't have some heat rays?

It also reminded me quite a bit of another "first contact" story that I'm assuming anybody that likes sci-fi and video games would have read, "Roadside Picnic".  Not so much in the details of the plot, but with respect to the mysteries that they edge right up to explaining but never quite give you a full understanding of what's going on.  I see that a lot in "Iron Curtain" science fiction, I guess when you can get put in jail for offending the government with stories about flying saucers you get really good at insinuations and talking around corners.  Maybe I'm looking for something that isn't there, but I kind of got the idea that when The Officer (or whatever that second doubler was, the guy with the magnetic notepad) was telling the humans about its government, that was some pretty heavy real-life commentary.

Definitely going to get any other Lem they have at the library next time.  Despite your warnings, I never found it slow or clunky, it opened with a crash and kept moving.  There are two books in my entire life I have not had the patience to finish, I have been defeated by the unabridged Brothers Karamazov and the unabridged Tom Jones.  If those had nuclear robot tanks and gyrosaucers I probably would have finished them.  If you have any other recommendations (and that doesn't just apply to xenoargh, if anybody else has some books I just must read, let me know, I read like half of Ian Banks because Alex assigned us homework that one time) I'll be sure to try them out.

[close]