|
pumpinglemma posted:Something doesn't add up here, ChatGPT is already capable of doing grade-school maths correctly most of the time and everyone involved knows it. The only way I can see this making sense as an advance is if it had perfect or near-perfect consistency, as a first step to removing hallucinations. Removing hallucinations would be a genuinely big deal. ChatGPT cannot remove hallucinations without being a fundamentally different thing.
|
#
?
Nov 24, 2023 18:01
|
|
|
|
| # ? Jun 6, 2024 22:47 |
|
|
Yeah, I'm leaning towards this mostly being a marketing ploy as well. "Why did your organization have a huge leadership kerfluffle recently? Uuuuh because our product is just too good, you see".
|
|
#
?
Nov 25, 2023 09:44
|
|
|
Perestroika posted:Yeah, I'm leaning towards this mostly being a marketing ploy as well. "Why did your organization have a huge leadership kerfluffle recently? Uuuuh because our product is just too good, you see".
|
|
#
?
Nov 25, 2023 14:43
|
|
|
"Don't Create the Torment Nexus" was one of the most ingenious marketing ploys in history
|
|
#
?
Nov 25, 2023 15:48
|
|
|
pumpinglemma posted:Something doesn't add up here, ChatGPT is already capable of doing grade-school maths correctly most of the time and everyone involved knows it. The only way I can see this making sense as an advance is if it had perfect or near-perfect consistency, as a first step to removing hallucinations. Removing hallucinations would be a genuinely big deal - not Skynet but still quite scary from the standpoint of e.g. someone like me who has to worry about students using it to write their coursework for them. The grade school math is still just fancy autocomplete. You train on enough text that says 2+2=4 and it will spit out 4 if you ask it what 2+2 is, just based on sheer probability. It's not the same as intrinsically understanding what a quantity is, or addition. I'll remain skeptical until proven otherwise that they have some solution to this.
|
|
#
?
Nov 25, 2023 16:47
|
|
|
Perestroika posted:Yeah, I'm leaning towards this mostly being a marketing ploy as well. "Why did your organization have a huge leadership kerfluffle recently? Uuuuh because our product is just too good, you see". the people who are appointed to the board of OpenAI, a nonprofit designed around a fear of imminent artificial general intelligence, tend to be the sort to jump at rumors that they've created AGI regardless of the underlying facts
|
|
#
?
Nov 25, 2023 17:47
|
|
|
shoeberto posted:The grade school math is still just fancy autocomplete. You train on enough text that says 2+2=4 and it will spit out 4 if you ask it what 2+2 is, just based on sheer probability. It's not the same as intrinsically understanding what a quantity is, or addition. I'll remain skeptical until proven otherwise that they have some solution to this. Same. You already have programming languages like Prolog that uses inference to give yes/no like responses. For example: animal(cat); animal(dog); likes(dog, food); likes(cat, food); And you ask it: Do all animals like food? The answer would be: Yes. HOLY poo poo A COMPUTER CONCLUDED ANIMALS LIKE FOOD WITH NO PROMPTING! ITS loving STEPHEN HAWKING! With 2+2=4, it would have to come up with the concepts of units, that you can combine them, and created its own version of math logic like every ancient culture did.
|
|
#
?
Nov 25, 2023 18:58
|
|
|
Vegetable posted:I recall the Wolfram Alpha guy saying that OpenAI should just staple Wolfram Alpha to ChatGPT to solve math problems. I wonder why they don’t just do that. This can be done right now via a plugin pumpinglemma posted:Something doesn't add up here, ChatGPT is already capable of doing grade-school maths correctly most of the time and everyone involved knows it. The only way I can see this making sense as an advance is if it had perfect or near-perfect consistency, as a first step to removing hallucinations. Removing hallucinations would be a genuinely big deal - not Skynet but still quite scary from the standpoint of e.g. someone like me who has to worry about students using it to write their coursework for them. It's exactly this, before it could score well on grade-school maths but had trouble with consistency, the new version can get a perfect score every time.
|
|
#
?
Nov 25, 2023 21:04
|
|
|
going by youtube comments on math youtuber videos, many adults dont know what math or numbers "really are" either. oh also some of the mathematicians that make those videos too.
|
|
#
?
Nov 25, 2023 23:58
|
|
|
It amuses me when people like engineers can't do basic addition without a calculator
|
|
#
?
Nov 26, 2023 00:14
|
|
|
Star Man posted:It amuses me when people like engineers can't do basic addition without a calculator It's a advantage in 'deal' meetings to have the ability to do math in your head.
|
|
#
?
Nov 26, 2023 05:49
|
|
|
Arguing with a young guy at work who's going to school for... computers I guess? He talks about AI as if it's some inevitability that it becomes the AI from science fiction (or as if it was already). Sent me a "source" that was, no joke, audio of some tiktok guy saying they almost have Jarvis from Iron man. He literally asked what I would do about alien invasions if I don't support AI development. Meanwhile the other coworker who's willing to discuss this is older. He says we've had AI for years, starts bringing up aliens and Isaac Asimov and you know how could we invent the microwave and satellites on our own? It's pretty frustrating. The tech bro hype is impenetrable. AI will improve exponentially until the techno singularity, it may as well have already happened. I just don't watch enough YouTube or tiktok to understand, fool that I am.
|
|
#
?
Nov 26, 2023 16:07
|
|
|
Contrary to what a lot of people outside the industry often think. You do not have to be smart to work in tech. We have a few guys at my place who are all in on the AI hype train, and would you believe it they were the guys into crypto before this. Mega Comrade fucked around with this message at 16:15 on Nov 26, 2023 |
|
#
?
Nov 26, 2023 16:13
|
|
|
BRJurgis posted:Arguing with a young guy at work who's going to school for... computers I guess? The only answer to an alien invasion is surrender. They traveled the cosmos to get here, we don't stand a chance. I work with "smart people" that think the moon landing was fake, but also that all the secret tech is real, but the lizard people control it all. Woo is prevalent in all places right now, and it increases as people lose touch with reality in order to shelter themselves in comfortable fiction.
|
|
#
?
Nov 26, 2023 17:07
|
|
|
VideoGameVet posted:It's a advantage in 'deal' meetings to have the ability to do math in your head. Kerry Packer, a since passed Aussie media tycoon/paper magnate, famous for hilariously huge bets (one story is that while in Vegas, a Texan businessman at the table boasted that he was worth 60 million to which Packer pulled out a coin and replied "I'll toss you for it") and a very direct talking style https://www.youtube.com/watch?v=EVIOmU3l0Zo. Despite being a sickly child and a poor student, evidently Packer was a demon at doing money math on the fly and would belittle finance people that couldn't do (for example) double digit multiplications in their head quickly.
|
|
#
?
Nov 26, 2023 17:30
|
|
|
BRJurgis posted:Arguing with a young guy at work who's going to school for... computers I guess? I mean the immediate comparison for that in my mind is Destiny and Rasputin got owned immediately then poo poo its pants, shut itself down and hunkered in, so I feel like I could handle that on my own.
|
|
#
?
Nov 26, 2023 19:25
|
|
|
BRJurgis posted:
Mega Comrade posted:Contrary to what a lot of people outside the industry often think. You do not have to be smart to work in tech. It's that absurd level of narcissism, like that guy doesn't personally know what he would do about alien invasions, therefore to him it's physically impossible that any humans on earth could possibly know what to do.
|
|
#
?
Nov 26, 2023 20:22
|
|
|
My friend worked with a guy who claimed autism was invested in a lab to give the West an edge over China. Effectively arguing he (as he had mild autism) was a kind of created superior mind.
Mega Comrade fucked around with this message at 20:59 on Nov 26, 2023 |
|
#
?
Nov 26, 2023 20:54
|
|
|
The problem with that theory is the depth of historical records about people who were almost certainly autistic long before we could engineer such a thing (if it's even possible today).
|
|
#
?
Nov 26, 2023 22:19
|
|
|
It also seems to assume that China has no history with autism.
|
|
#
?
Nov 26, 2023 22:23
|
|
|
If Hong Xiuquan were autistic, he wouldn’t have failed those exams.
|
|
#
?
Nov 26, 2023 22:26
|
|
|
BRJurgis posted:Arguing with a young guy at work who's going to school for... computers I guess? As a dead site created by a terrible person once said: the internet makes you stupid.
|
|
#
?
Nov 27, 2023 01:14
|
|
|
Jokes on them, I was already stupid
|
|
#
?
Nov 27, 2023 01:17
|
|
|
Agents are GO! posted:Jokes on them, I was already full of stairs in my house Ftfy
|
|
#
?
Nov 27, 2023 05:28
|
|
|
Evil Fluffy posted:As a terrible site created by a dead person once said: the internet makes you stupid.
|
|
#
?
Nov 27, 2023 05:55
|
|
|
Vegetable posted:I recall the Wolfram Alpha guy saying that OpenAI should just staple Wolfram Alpha to ChatGPT to solve math problems. I wonder why they don’t just do that. I think in their head, they're trying to build the general AI, and stapling a calculator to the thing's head wouldn't achieve that goal.
|
|
#
?
Nov 27, 2023 17:39
|
|
|
Neito posted:I think in their head, they're trying to build the general AI, and stapling a calculator to the thing's head wouldn't achieve that goal. Many of the big breakthroughs in deep learning have had to do with ideas of invariance (or equivariance) respecting various symmetries - although this may be in disguise. For example, convolutional neural networks attempt to take into account the fact that image features should be invariant to translation and/or rotation (a cat on the left side should be the same as a cat on the right side). This kind of thing works so well because it can narrow down a huge solution space into a much smaller one that respects the invariants/concepts that it should. This helps accuracy and efficiency. The issue is how to encode such symmetries. In general, this involves mathematical groups. Even such abstract concepts as abstract mathematical knowledge could be reflected within groups. The hard part is doing so. Not only such ideas need to be represented as groups and symmetries, but those themselves have to be translated into a GPU-friendly representation. If Stephen Wolfram's ability is a hundredth the size of his ego, he should be able to do it no problem. But I'm sure he just doesn't want to.
|
|
#
?
Nov 27, 2023 23:02
|
|
|
I'm sure he just wants to get paid and the bill is higher than AI researchers, who are used to getting things for free, are willing to pay.
|
|
#
?
Nov 28, 2023 04:08
|
|
|
In today's ChatGPT experiment: Prompt: The following is a multi-slide TiKZ diagram for use with Beamer. It depicts a specific heap memory allocation algorithm for a fictional system with heap memory addresses 0x800--0x3fff. On each slide from the second onwards, the most recent function call is depicted at the top left. Can you explain the algorithm depicted by the slides? \begin{tikzpicture} \draw [thick] (0,0) rectangle (10,1); \node [anchor=south west] at (-0.1,1) {\rotatebox{30}{RAM[0x800]}}; \draw [ultra thick] (0.3,0) -- (0.3,1); \draw [thick] (0.6,0) -- (0.6,1); \node at (0.45,0.5) {\rotatebox{90}{\tiny \only<-2>{0x37FE}\only<3->{0xDFF}}}; [Continues for 200 lines of my lovely code littered with \onlys and \onslides that make things appear and disappear on specific slides] ChatGPT response: This TikZ diagram represents a memory allocation algorithm with heap memory addresses ranging from 0x800 to 0x3FFF. The algorithm involves allocating and deallocating memory segments. Let's break down the diagram based on the function calls and actions: Slide 1: The initial state of the memory is shown, with the entire heap represented as a rectangle. The thick line at 0.3 separates the heap into two halves. The rectangle at the rightmost end represents a screen or reserved area. The labels on the top indicate the memory addresses of the first and last cells. Slide 2: The first function call is alloc(0xDFF), attempting to allocate a memory segment of size 0xDFF. The allocated memory segment is shown as a shaded rectangle from 0.6 to 2.5 with a label indicating the range (0x800 to 0xDFF). The arrow indicates the allocation direction. [Continues in this vein for 20 more slides.] Like, it doesn't actually get the algorithm right (the "allocation direction" is nonsense), even with the additional prompt that the arrows on the diagram represent pointers. But it is able to accurately parse the tikz diagram into shapes, explain how those shapes change with successive slides, and make reasonably-accurate guesses as to how those shapes correspond to the memory allocation scenario. That's much better than I was expecting it to do!
|
|
#
?
Nov 29, 2023 17:09
|
|
|
being able to deal with TiKZ does imply that ChatGPT has superhuman intelligence
|
|
#
?
Nov 29, 2023 17:31
|
|
|
I know video game recommendations generally don't fit with this thread, but I really have to pitch Going Under to the thread. It's a little action Roguelike send-up of tech startup, where you play as an unpaid intern who's tasked with clearing out the basement dungeons, which consist of the remains of the previous tech startups that occupied the building. The best thing is the art style, which manages to capture the corporate Clipart feel. Lots of fun. Very humorous too.
|
|
#
?
Nov 29, 2023 18:27
|
|
|
I think all the heads need to be smaller for that new pinhead look everything has now, but that is an amazing concept. It looks like it would even run on the potato I shoved into my video card slot!
|
|
#
?
Nov 29, 2023 22:51
|
|
|
https://www.tesla.com/cybertruck Looks like the nightmare is coming in 21 hours. Or well, a nightmare.
|
|
#
?
Nov 29, 2023 23:30
|
|
|
What you call a nightmare I call great content
|
|
#
?
Nov 29, 2023 23:40
|
|
|
Agents are GO! posted:I know video game recommendations generally don't fit with this thread, but I really have to pitch Going Under to the thread. It's a little action Roguelike send-up of tech startup, where you play as an unpaid intern who's tasked with clearing out the basement dungeons, which consist of the remains of the previous tech startups that occupied the building. The best thing is the art style, which manages to capture the corporate Clipart feel. Lots of fun. Very humorous too. Indie games? On the 2k page? 
|
|
#
?
Nov 29, 2023 23:53
|
|
|
Indie is an aesthetic. They list the publisher right there
|
|
#
?
Nov 30, 2023 01:22
|
|
|
Got this from google and as a parent this sounds like a tech nightmarequote:Google Play allows your child to make in-app purchases of digital content through alternative billing if offered by the developer. Learn more. e: I don't have any methods for payment on their phones, so I dunno how they could trick my kids into buying stuff without some way to transfer money, but sure feels nice to hear google say they will just have these nice backdoors from now on. His Divine Shadow fucked around with this message at 05:25 on Nov 30, 2023 |
|
#
?
Nov 30, 2023 05:17
|
|
|
surely there's a toggle e: yes the parent would have to set up the alternative billing situation. the terms of service are very strange but afaict it's something wonky to do with Indian and South Korean app laws
|
|
#
?
Nov 30, 2023 05:22
|
|
|
Ruffian Price posted:Indie is an aesthetic. They list the publisher right there You’ll get the joke when page 2560 comes around.
|
|
#
?
Nov 30, 2023 05:32
|
|
|
|
| # ? Jun 6, 2024 22:47 |
|
|
Google Jeb Bush posted:surely there's a toggle It's from the EU Digital Markets Act, I think. That is an antitrust law that, among other things, requires Google/Apple to allow 3rd party payment systems. If some app is using their own in-house payment mechanism where you've given them credit cards or whatever separately, it makes sense that permissions settings from the separate google-specific one don't apply to that and purchases through it won't show up when you look at the google payment system's history.
|
|
#
?
Nov 30, 2023 07:02
|
|
