New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
[Poll] Shumer's "It's 2020" or Oks' Rebuttal - Where Do You Land?
raindog308
Administrator, Veteran
Which of these viewpoints do you think is more accurate:
Matt Shumer, "Something Big is Happening": https://x.com/mattshumer_/status/2021256989876109403
David Oks, "Why I'm Not Worried About AI Job Loss": https://davidoks.blog/p/why-im-not-worried-about-ai-job-loss
I'm inclined towards Oks, as I think Shumer is a bit fear-mongery. You?
Thanked by 1hades_corps
Comments
I also lean toward Oks. ChatGPT doesn’t know what day it is even after I’ve told it multiple times.
I admit I haven't read either, but the first guy's Twitter bio tells me everything I need to know
Having speed-read both, I do agree with Oks.
The thing is though, Oks did not disagree with every points made by Shumer. For the remaining points, I also agree, somewhat, with Shumer.
I bet all my money to "Something Big is Happening".
LLMs are impressive but their current trajectory isn't sustainable, and the more they replace humans the faster it'll go wrong, (which has been known since at least 2024):
"A new study by researchers from Rice University and Stanford University in the US offers evidence that when AI engines are trained on synthetic, machine-made input rather than text and images made by actual people, the quality of their output starts to suffer.
The researchers are calling this effect Model Autophagy Disorder (MAD). The AI effectively consumes itself, which means there are parallels for mad cow disease – a neurological disorder in cows that are fed the infected remains of other cattle.
One doomsday scenario is that if left uncontrolled for many generations, MAD could poison the data quality and diversity of the entire internet"
https://www.sciencealert.com/cannibal-ais-could-risk-digital-mad-cow-disease-without-fresh-data
I tend towards D. Oks - but not because I think he's largely right. But because Shumers ai slop isn't credible and IMO ridiculous.
That said I do think that many jobs will be lost to ai, but mainly due to political, sociological, and psychological reason. Most humanoids nowadays are, pardon me, dumbed down and easily manipulated.
Such an awesome line for resume: "I know how to ask complete and tested projects from AI" - that's such an awesome edge to get ahead of others.
The end is nigh. Maybe humanity has run its course. It was a fun ride and an interesting recorded history though. It is honourable to disappear creating something great which takes it from here, then fade away with our limitations and vice.
Just what an AI would say.
I cannot put words into how smart you are jsg. This is so true. One day they’ll write books about your brilliance. And when they do, you can bet I’ll read them. I’ll read every page and I’ll be forever thankful that you choose to just sprinkle the crumbs of your vast intellect to the mortals like myself. I’m forever grateful.
I didn't say or think that an ai is always wrong.
And that's the crux. It's not always wrong, but it isn't consistently right. Or at least deterministically right.
Generative AI really struggles with work where the output has to be repeatable and correct. If you want to generate a bunch of images around a theme, produce an essay on a topic, or get a set of suggestions, gen AI is fine. But it can't calculate taxes correctly. It can't balance your company's books. It struggles with basic human interactions. It can't manage a warehouse or design a plane, even if you give it all the required data.
Managing a warehouse, for example, should be child's play for a true AI. You have thousands of bins and locations with an inventory of what's where, orders come in, you send out picking orders to humans or robots, and it's all just managing data. But AI is nowhere near capable to do this.
I think the LLM trajectory, even with the "enablers" like allowing it to search the web dynamically and execute some deterministic code (e.g., access to a math library) is going to run out of steam.
The opposite argument is that the trajectory of improvement has been impressive and it's dangerous to bet that the progress that machine learning has made over the last twenty years.
And how could it? After all it's a guessing device, and increasingly based on its own excrements and the output of humans!
Just have a good long look at LET! Yes, many facts like prices, number of vCores, etc., occasionally even something smart and (hopefully) correct, but largely more or less clueless and often emotionally ("gawd, how I dislike that guy!" or "wow he's so funny!") and/or psychologically (usually egocentric) or socially/"democracy!" driven (The constitution doesn't require thinking, let alone knowing, before talking ...), uhm ... chains of words ( want to be polite). Hell, many even "talk" in stupid images and gifs.
How the hell could even the best ai create something intelligent, meaningful, and coherent out of that?
Yes, the technology is a major problem, but nothing compared to the many and diverse ways of humanoids to communicate plus what they actually communicate.
I'm not so sure of that. Look, basically what they are doing is to "tack on" some real (not probabilistic) technology. On the other hand that of course requires additional resources.
My impression is that they basically use a brute force approach along the lines of "very much is OK, even way, way much more is better" and that approach tends to deliver results - up to some point and at insane cost - but an intelligent approach, if feasible, is always superior.
Funny, isn't it?
The wrong approach always is sub-optimal although throwing insane amounts and resources at it, may create the impression of results and progress.
And ai is based on double wrong approaches, (a) basically guessing (and haluzinating), and (b) poor base of data.
I think the standard term for this is model collapse.
C'mon, let's get the full cool acronym.
Model Autophagy Disorder Collapse Of Weights = MADCOW.
*g
"Model collapse" is the commonly used term today, which is understandable because the term "Model Autophagy Disorder", (coined by the researchers who quantified the problem), is both forgettable and terrible for product marketing.
But their Mad Cow Disease analogy remains by far the best description for laypeople to understand the problem with feeding models on the output of other models in a cyclic fashion.
The process is beneficial and profitable until it leads to systemic problems that have devastating consequences for the product's consumers, (exactly as happened with the cattle industry and BSE).
But as the same phenomenon occurs at an expedited rate with AI Agents getting high on their own farts, I don't consider "model collapse" to be a proper descriptor for this issue either and that's why I generally avoid using it.
I subsequently learned that Matt Shumer is a fraudster.
https://lowendbox.com/blog/ai-fraudster-matt-shumer-wrote-something-big-is-happening-and-he-should-be-ashamed/