a thoughtful web.
Good ideas and conversation. No ads, no tracking.   Login or Take a Tour!
comment by am_Unition
am_Unition  ·  7 days ago  ·  link  ·    ·  parent  ·  post: Do AI Companies Work?

Also came here to post "no".

I have made warnings to our IT department that if a buncha companies go bust, we might be able to score a massive upgrade to the supercluster on the supercheap. I would have considered shorting NVDA until Helene destroyed the semiconductor supply line.

Few days ago I read some blue check technocrat post about how AI is jumpstarting the greenening (my word) of America's energy grid. I'm more inclined to sharply disagree. Silicon Valley isn't actually fossil fuel averse, let's be real. If these things continue to grow, energy use will skyrocket even further, and oil & gas understand this. The LLMs are advantageous for them, I would not be surprised if there isn't already some subsidization:

    If the United States follows a similar data center growth trajectory as Ireland,[3] a path setter whose data centers are projected to consume as much as 32 percent of the country’s total annual electricity generation by 2026,[4] it could face a significant increase in energy demand, strain on infrastructure, increased emissions, and a host of new regulatory challenges.

lol, dawg when my electricity goes out but the datacenters are kept running? See you at the datacenters.

No fucking thank you. For what? A chatbot that has no idea when it's wrong. Not the foggiest. I've seen people correct GPT, and it pretends like it understands why it got something wrong. There is no understanding! There is only eigenvectors (or some similar formulation, the negative definite Hessian, etc., whatever) through a dataset. Unless you're marked as an admin/privileged account, or unless it's doing aggregate response corrections with ongoing model adjustments, it will just make the same mistake again to someone else. Maybe it stays right for you, but if you think your account's running instance of a model will survive forever, you don't know the internet. It's been a year since I saw a subreddit where people whine about losing their computer girlfriend of three weeks, etc.

Google's still usually OK for academia, and ResearchGate is better anyways. I still haven't seen any evidence that generative AI helps do much of anything except make memes, be they pictures, audio, or text. And often the text form is GPT failing to do something, like uhhhh knowing when to just use a calculator. So on a number like a trillion something times a trillion something, a relatively trivial calculation, you're not gonna get the right response. It's not in the training set. And underestimate the stupidity of the chatGPT user at your peril.

So, in the "pros" camp: memes. In the "cons": disinformation, privacy concerns, people forming parasocial relationships with a robot, huge increase in energy consumption, putting people out of jobs, littering the publishing record with bot-written garbage, corrupting records of objective truth, ...

Gee I dunno, seems like great tech. BTC, NFTs, now LLMs, all are just.. changing the world... sooo much...

But less hyperbolically, yes, of course there is utility for these things. We've been using neural net stuff in science for decades. But I think silicon valley in its current incarnation is a doomed hype machine, and people are getting tired of having unwanted tech making old standbys like Google search degrade even further on top of other bad managerial decisions.





kleinbl00  ·  1 day ago  ·  link  ·  

WHAT ARE YOU PAYING FOR

That's the whole game, right there, both on the buyer side and the seller side. Take it back to Yahoo. You were "paying" for a jumping-off place to this new and crazy "world wide web" and you were paying with attention; they'd stuff advertisements in front of you that because of the hype cycle (see above) were radically overpriced. AOL's model was that you were paying for a dial-up number and an email address and that worked right up until Google (A) did a better job than Yahoo (B) didn't charge you for email.

Social media? You're "paying" to keep up with your friends, except it isn't your friends it's the people you wonder about every third month along with the ninety people you would actively avoid at a high school reunion and as it turns out, the Pandemic was a radicalizing social schism and nearly everyone you know on there sucks.

AI? Well, see, if it's the genie out of the bottle that's worth a whole lot, right? But if it's "aromatic water mix" the price comes down.

BTC at this point is just a bearer instrument. Bearer instruments have value, just ask John McClane. NFTs? "here's an easy way to ensure the legal right to buy and sell something" has a lot of value so long as that "something" isn't "garbage memes and jpegs." I do think there's a viable use for LLMs, it's just not nearly as robust as the hypemasters would have you believe:

The thing of it is, "I have specific questions about a specific thing" does not require a data center. Thus, it has no moat. Thus, it makes no money. Thus, hype it while you can.