Ed Zitron: Newsletter: Goldman Sachs has called BS on Generative AI, and I believe that it's time that everybody follows suit - generative AI is unreliable, unsustainable, requires an entire rebuild of America's power grid, and is most decidedly not the future. www.wheresyoured.at/pop-culture/

Newsletter: Goldman Sachs has called BS on Generative AI, and I believe that it's time that everybody follows suit - generative AI is unreliable, unsustainable, requires an entire rebuild of America's power grid, and is most decidedly not the future. www.wheresyoured.at/pop-culture/

Pop Culturewww.wheresyoured.at A week and a half ago, Goldman Sachs put out a 31-page-report (titled "Gen AI: Too Much Spend, Too Little Benefit?”) that includes some of the most damning literature on generative AI I've ever seen. ...

Goldman Sachs recently issued a report called "GenAI: Too Much Spend, Too Little Benefit?" savaging the industry, with GS' Jim Covello saying that there's limited upside, and that it's incapable of solving the complex problems that would make it worth it. wheresyoured.at/pop-culture/

Sachs' report is brutal, and finally asks some very blunt questions: what does generative AI actually do? And when we say it'll "get better," what does that actually mean, and how does that justify the massive costs and errors associated with LLMs? wheresyoured.at/pop-culture/

What do we get from all of this investment, stolen training data, and strain on the power grid? The answer is not much - Daron Acemoglu of MIT believes that GDP and productivity growth from Gen. AI is "much more limited than forecasters expect." wheresyoured.at/pop-culture/

The lone pro-Generative AI part of the GS report is a mess, making leaps of logic to suggest LLMs will lead to an AGI with human consciousness - something that researchers have repeatedly proven isn't possible. Gen. AI fanatics live in a fantasy land. wheresyoured.at/pop-culture/

Those replacing human beings with AI are doing so because they've no respect/understanding of labor, commoditizing outputs without seeing that a job is both the acquisition of labor and the outsourcing of risk. You don't just hire humans- you trust them. wheresyoured.at/pop-culture/

The most brutal part of the report was from Goldman Sachs' Head of Global Equity Research Jim Covello, one of the most technical and well-respected analysts in tech. He thinks Gen. AI is full of shit- and nothing like the early days of the internet at all. wheresyoured.at/pop-culture/

worth noting the people pushing for this don't have a concept of "trust" they understand only that their money buys labor, and labor to them is an interchangeable and fungible commodity; they have no concept of experience or expertise, only that things happen if you throw enough bodies at a problem

This is a nitpick, as in general I agree with you, but the part about workers getting jobs in other fields is a fairly standard part of economic growth—the assumption is that the new jobs will be higher-value than the ones that are replaced, in the same way that 98% of people aren’t farmers.

I say it’s a nitpick because I share the scepticism that AI can actually do that thing.

who could have predicted this ed

That invent and test bit (just after this) with materials synthesis hasn't worked out right either. The A-Lab thing doesn't understand compositional disorder - they were reporting perfectly ordered versions of existing imperfect materials. Nature published on the CompSci story, not the MatSci one.

New analysis raises doubts over autonomous lab’s materials ...www.chemistryworld.com Experimental and computational issues flagged as researchers conclude that a fully automated lab failed to make new materials

If you were looking to put AI to work in the physical world you'd need to let it discover its own rules and not just train it on human rules. The bigger problem is it takes a lot longer to validate a theory than to invent another crappy one. It is a built in DDOS on experimentalists, on knowledge.

How2AIgstaneff.medium.com Google and Microsoft are in the news recently for using AIs for new material synthesis. How could those efforts have been more effective?

📌

Meanwhile my profession is like “this is the future for the legal field!” Yikes.

You're comment about what people should have been asking pains me. But I don't think it would have mattered. I was in investor advisory meetings and I and other people with experience and history said "this is nice but here's the same old unsolved problems" and they still threw bags of cash at it

JFC I typed you're instead of your. I'm going to go take pictures of birds

Just FYI, Ed, nobody abbreviates Goldman Sachs as 'Sachs.' Goldman, yes, GS, sure, Sachs? Never.

piss off

Guy with Trekkie in bio bravely defending the honor of Goldman Sachs on the world's 543th most popular website

543th?

Not much what's 543th with you

*Mr* Sachs

Sachs of shite, seems like the exception to that rule

The other issue on "cost going down", fundamentally the training problem is O(n^3), so you make the problem 10 times bigger and it costs 1000 as much to do. Big models are just, frankly, a stupid design.

Where I work I remember someone saying about a dozen years ago “big data is the new oil!” (It wasn’t), and now I feel like I am hearing “gen AI is the new new oil!” (It isn’t)

Remember quantum computing

Ball bearings

New snake oil, perhaps.

heads up, palette*-cleanser should be palate. Paint thinner is a palette cleanser

bsky.app/profile/zitr... didn't ask

please note that if you send me a typo and haven't retweeted the article i have no interest in hearing from you. I also really don't if you want to point out typos either way! Go do something else

thats fine sorry for the info

It's just really annoying, especially at a post embedded in a long thread, knowing I can't fix it?

that is not even slightly a thing i know? seems like you can edit any web page you like any time you like

I think they can axe the question mark in the title.

GenAI : An obvious scam to bilk credulous buffoons.

The AI hype reminds me of the very very early hype for the Segway. Which has been erased from history because it was a catastrophic embarrassment for the media, but the South Park episode is a good refresher on just how gullible our media is.

We were told they’d build cities around it.

You mean "IT!"

That was so wild. People breathlessly saying "this will redefine human civilization". And then we all saw it and we were like 'well that is pretty neat, I guess.' The company is still around but they don't even make that version of it anymore

the owner of the company DIED rolling off a CLIFF on a SEGWAY

once your product outright kills its chief advocate, you've *traditionally* been pretty cooked

*Titan submersible has entered the chat*

Lol. Can you even imagine riding off-road on something where you have no brakes and must shift weight to slow down, something essentially impossible to do if you start picking up speed going downhill

The funny thing is people seemed ready to redesign cities around it but it wound up just being a worse bike We still have bikes we can still redesign our cities

Yes, exactly, very anti-bike thinking which is such a USA thing, 'bikes are for children not grownups so we need this other thing that an adult can signal that they are NOT a child or poor'

the segway worked exactly as intended it just had no compelling use case. still have a soft spot for it. that seems different from generative AI which has a number of well known issues that interfere with its proposed uses

It’s almost like the people that called it early on were entirely correct. But no we were told that we just didn’t understand.

Post