[drchickensalad]

> The comparison suggests that the thing doesn't do what it's marketed to do

In my opinion you should interpret the usage of "AI" here to mean "the entire business/management/financial/bubble ecosystem surrounding LLMs". The snake oil is much more how LLMs are being weaponized and utilized rather than a specific technical assessment (although that often is an issue too)

[api]

The same thing happened around PCs, gaming, the Internet, the web, and cryptocurrency. It's a hit driven industry that loves hype.

[indymike]

> The snake oil is much more how LLMs are being weaponized and utilized

Calling something "snake oil" implies that the product only, at best, has a placebo effect (amplified, no doubt, by the sales presentation). LLMs seem to be much more useful than a placebo.

[datsci_est_2015]

I think you just simply repeated the mistake that the comment responding to you was pointing out, i.e. conflating “LLM” the technology with “AI” the hysterical product hype bubble phenomenon.

[patates]

I think then the AI-hype should be disregarded as snake-oil, not the tech itself. I know it sounds pretentious, but I need to say that words do matter.

[fasterik]

My prediction is that it will go the same way as the dot com bubble. The hypesters and fraudsters will eventually collide with objective reality, but the technology will persist and society at large will benefit from the infrastructure and the increased access to knowledge.

[Shellban]

Assuming that we recover from the damage being done now. As one example, a friend of mine has remarked that large corporations will benefit from the current AI-induced reality of no one being able to afford their own hardware, and keep prices that way to enforce a renters model on computers.

[thewebguyd]

That's my fear, and unfortunately I think its likely to happen. I feel like we will settle down a little bit, but into a new higher "normal" baseline that still largely makes it unaffordable for most.

There's also still the risk of the creation of a new economic underclass, if both a) hardware remains too expensive for local inference and b) subscription or pay-per-token based inference also remains expensive or increases in price, then individuals will largely be locked out of the benefits that having access to AI could bring, leaving it purely in the hands of larger companies. People will only get to use and experience these tools through their employer, for the benefit of their employer.

[teo_zero]

> People will only get to use and experience these tools through their employer

Like in the era of mainframes, before the personal computer was a thing. I wonder if universities will play the same role, though.

[abc123abc123]

This is not correct. Prices will adjuts when the crash comes. At worst, the market and prices will just open peoples eyes up to the reality that 99% of the daily software we use, runs perfectly fine on a 5 years old computer.

The idea that you _must_ have 128 GB RAM and 1 TB ssd in your computer is juts absurd.

Remember... we reached the moon with the compute power of a pocket calculator, and there is no eternal law that says that everything has to be written in javascript.

[thewebguyd]

> The idea that you _must_ have 128 GB RAM and 1 TB ssd in your computer is juts absurd.

I don't think anyone thinks that? Unless you are trying to do local inference with bigger models, which isn't everyone's cup of tea but I do think is an important capability.

Outside of that use case though, prices are still ridiculous. 32GB of DDR5 will run you close to $500 right now vs. $80 before the AI build out, and 32GB is what I'd consider comfortable so long as everything keeps being electron and web wrappers (assuming Windows & macOS here for the general population, obviously you can get by with less on a Linux desktop for the most part).

Personally, I'd love to see more software do more with less, and go back to performant native apps but in reality I just don't see that happening, except maybe over here in Apple/macOS land which has always had a decent culture of native apps. For now, the incentives aren't there (for commercial software).

I hope you are right on prices, but I don't share the optimism. We've seen time and time again that once prices go up, they don't always go back down even when the supply crunch is eliminated. Manufacturers realize people adapt to whatever "new normal" prices are at and refuse to lower them. Same thing happened post-COVID supply shocks. Companies have zero incentive to start a price war, even after supply shocks are over.

I think we are stuck with ~$500 RAM for a long time.

[protocolture]

>large corporations will benefit from the current AI-induced reality of no one being able to afford their own hardware, and keep prices that way to enforce a renters model on computers.

Depends how long the RAM correction takes. It is interesting how RAM prices have stifled the creation of cheap laptops capable of running big models. But at the same time, this seems like a second order effect and not the intention.

[api]

It’s a demand spike.

[echelon]

> As one example, a friend of mine has remarked that large corporations will benefit from the current AI-induced reality of no one being able to afford their own hardware, and keep prices that way to enforce a renters model on computers.

Choose one:

- You spend 30 hours writing a program to manage data for your hobby. You write it on your personal computer.

- You spend one hour generating a program to manage data for your hobby. You have to lease an H200 behind an API to do it.

Which one will you choose?

I know which one I'm choosing.

[anon7725]

I choose A.

I know that many others choose A as well.

A wonderful service known as the web has connected people who choose A with others who choose A and of course with a great many who don’t need to make a choice and benefit from the work of others.

I mourn a world in which few will choose A, because for many to choose B seems to lock us all, tragedy of the commons style, into a worse world.

[account42]

And let's not forget that choosing B is only possible at all because many people chose A before you.

[dgellow]

I’m choosing the first option. I’m not sure which one you chose?

[bigstrat2003]

Option A is fun, whereas option B is miserable. Option A is also cheaper. It's a pretty clear win for option A here.

[sdenton4]

Of course, you can use Option B to write the program, and then run it on your own machine...

[Hugsbox]

You're making a really strong case for option A, whereas B sounds really depressing

[gyomu]

> society at large will benefit from the infrastructure

Data centers as infrastructure are very different from DSL rollout though. Much, much more expensive to maintain, with a much much shorter timespan.

If the bubble pops and data centers get shut down because there’s no one to pay the bills, there won’t be much left 5-10 years later in terms of infrastructure.

[m0llusk]

Maybe we could repurpose old processors to power toaster ovens.

[bigbuppo]

I mean, a 125kW server stuffed into 5U will definitely work nicely for toast.