I appreciate your efforts to convey your insights and the insights themselves. Thanks for that.
It is something that is becoming ever more rare here on Slashdot, especially when it comes to AI. There is a weird "organics are magic, machines are stupid" kind of crowd that I had never expected to be popular here, but here we are. So again, thanks for not being like that and pushing back against it with logic and reason.
Used to do that decades ago, but realised that if I ever want to re-watch something, a higher-quality version of it would probably be available and downloadable in minutes.
I wouldn't be too sure of that. I've seen torrents become more ephemeral, with people just not bothering to seed very old stuff (which makes sense) either in existing or newly uploaded torrents. Streaming/commercial services are also far from guaranteed to offer old (obscure) stuff.
I'd say that a better argument would be that there is so much new stuff to watch, that incidental rewatching of stuff has too little relative value to do the archiving work.
Have you tried high-res 180 degree stereoscopic porn (viewed on a proper VR-device)?
It is far from as convenient due to various technological and interface challenges, but once past that it 'hits different', as the kids say. POV porn gets a whole new dimension. Literally.
A sufficiently advanced AI can hack and/or infect the fuck out of 95% of our digital infrastructure and hold the entirety of humanity hostage through that. Which datacenters, corporate computers and home computers are you going to 'bomb out of existence', all of them? Talk about hand waving.
We can't even deal with underground terrorist organisations and state hackers properly. You think we stand a chance against a superhuman intelligence?
There is actually a test you can (have people) do:
Picture a 3x3 grid with three words written horizontally, then read out what is written in the columns. If you can visualize the grid, you can just read from the mental image. Otherwise you'll need to mentally reorder the letters, which is generally a slower process. It is definitely something where you know which of the two approaches you are using (there is a third approach, which is remembering what the answer was for a specific set of words, so use new words each time).
I suppose you think you're very cleverly not giving in. The reality is that you're acting like a child.
I clearly pointed out the flaw in your reasoning and you tried to escape the resulting cognitive dissonance by making up shit that supposedly invalidated what I said and now you're being willfully obtuse. Next step will be a full blown temper tantrum, probably.
OK, so are you claiming that LLMs like ChatGPT hardly output spelling mistakes because their output is fixed by first "running it through [a] common or garden spell-checker"?
I hate marketing and sales drivel as much as you do, but it's wholly the wrong thing to focus on. Ads gonna ad, marketeers gonna marketeer, you're not going to stop that without fundamentally changing our economic system. And what of it? What are you actually going to achieve by 'fighting' marketing assholes via comments in this community?
I am not exaggerating here: We are at the advent of the next step in the evolution of intelligent life. We always knew it would happen at some point in the future (we're pretty much naked apes, after all), but very few people expected it to come so soon (including me). I'm not saying it is tomorrow, but it might very well be within the next ten years. Within my (expected) lifetime for sure. Seeing this community of intelligent and (previously) well-reasoned people looking away from the thing that is obviously happening before our eyes makes me sad and angry. And afraid. Because the marketing drones and the general public sure as fuck aren't capable of or going to improve our chances of dealing well with what is coming.
Dealing well with it starts with seeing the state of technology for what it is and what it can become (soon). Mindlessly and jadedly calling it "useless chatbots", "overblown hype", "statistical parrots", "artificial stupidity", etc. (as 80% of the comments on every fucking Slashdot post on AI do) is just a bad psychological defense mechanism that creates a false sense of security.
Don't try to weasel your way out of it by pretending that that is why LLMs hardly output any spelling mistakes. Just admit that you were wrong.
Nonsense. The word 'useless' has a very clear meaning. Don't try to defend that jaded grumpy old man bullshit.
If you've missed the news, please look into how adding the transformer architecture to image and video generation AI has yet again blown everything before it completely out of the water: https://openai.com/research/vi...
If you want to call that progress 'overblown', 'truly next-level stupid' and 'useless', go ahead. I'll stick to being open minded, open eyed, and scientific. The transformer architecture and what is being done with it really is something amazing (and terrifying), even if it is not perfect yet.
Because, of course, if there are bugs in the training code, it's going to regurgitate them to the novices who will need it to complete an unfinished line of code.
That would be the same as saying that a similar system trained on natural language would 'regurgitate' misspellings because its training material contained a few of those. Clearly, the current AIs are far more capable at spelling than the average human even though they were very probably (also) trained on a shitton of terribly spelled English (Reddit, anyone?).
So your statement is incorrect and displays a lack of understanding of how the training process works.
We aren't anti-technology.
You may not be, but the person I was responding used these words:
"BINGO! Back to where we are, with useless chatbots."
Calling the "current round of AI" useless chatbots is myopic, retarded and anti-technology. Your head has to be pretty far up your own ass to ignore the incredible and clearly quite usable advancements of the past year in the field of AI.
Maybe some people have matured a bit more than "hey, new tech, cool!" and think a bit more about it critically. The AIs are hallucination machines, not knowledge machines, by DESIGN, so it's a bit hard to remedy.
This is exactly what I'm talking about. You don't understand well enough what they are (even AI researchers don't, so have some humility), but you do latch on to one of the few elements in which this new tech is imperfect, ignore the enormous impact it already has in its fairly infant form, and pooh-pooh it and everything slightly like it for the remainder of the existence of this universe.
Think about tasks like speech recognition, image recognition, OCR, translation and all the other tasks that these "by DESIGN hallucination machines" have blown previously existing tech completely out of the fucking water. Is that 'hallucination'? 'Useless chatbots'? I can now give one of them an image with a visual joke and it does a pretty good job of explaining the joke.
Critical thinking is not just being jaded and contrarian, you know.
The gobs and gobs of R&D money are up for grabs from opportunists everywhere, they're not always good investments.
No shit, Sherlock. That wasn't the fucking point. Again you latch on to some imperfection in the stated and ignore the main thrust of it. The dot com boom and the early days of the internet also saw fucktons of bad investments and grift. Yet the web grew like a motherfucker because of all the money and research that was well spent, now didn't it?
Also the ridiculously old and proven technique is expensive and doesn't scale, so now you have the costs of both humans and machines.
Oh, now whether it scales is a criterion? Way to move the fucking goalposts. Damn this new approach, it only increase productivity by 50% instead of exponentially.
As to how expensive it would be: you clearly have no experience with these AI systems, or you are being willfully ignorant. Getting AI to generate shit for you that is 80% OK and getting that to 100% of what it would have been if you had done it yourself is trivially cheaper if that saves you a fuckton of time. I have been doing this almost daily for the past couple of months with fairly complex matters. It is very easy to see how a bunch of Excel banker grunt work can easily be made much cheaper by doing the same. Even if each employee stays on and just checks their own AI-assisted work thoroughly.
How is this even anti-technology?
Calling the incredible advancements in AI "useless chatbots" is clear anti-technology speak.
What would a remedy actually look like?
AI that doesn't hallucinate. Demanding of me that I now personally come up with the solution for hallucination is unreasonable. Having said that, what might work is something like self-validation of the answers, using external systems or even just the AI asking "was the answer that I just gave truthful?" Pushing the AI to include more qualifiers as to the level of certainty it has in different elements of an answer would also go a long way.
For an internal check or review to do that, you're literally talking about having a review for basically every question that could be asked. Do you see the problem with that?
Don't be silly. Remember the inane process that was presented:
"Step 1. Chatbots says something
Step 2. Bank gets sued"
Only if you have each answer be a direct output to the outside world is this a thing. Otherwise we're talking about (FTFS) "tasks that take up almost three-quarters of the time bank employees now spend working" where the things that matter are the deliverables, the actual production of those tasks. You don't have to check/review all the (answers to the) questions, you just have to check/review the deliverables. You know, like we already do for human production where it matters.
God, Slashdot has become an awful place. When the fuck did it become such a cesspool of anti-technology assholes who can't see any further than their own nose?
With all the developments and the ridiculous speed of progress of the past year, do you really think the hallucination thing can't be remedied? Even if the gobs and gobs of R&D money and mindshare thrown at it don't succeed in fixing it on a technological level, the ridiculously old and proven technique of just having in internal check/review by somebody else fixes your retarded scenario.
It is easier to write an incorrect program than understand a correct one.
