Many people are exclaiming that the title is baity, but I disagree. It seems like a perfectly fine title in the context of this blog, which is about a specific product. It's unlikely they wrote the blog with a HN submission in mind. They're not a news publication, either.
Things AI is already better at than (many/most) humans: Customer service (chat, phone), writing software, writing docs about software, computer graphics (animation, images), driving cars.
There are plenty of uses for AI. Right now, the industry is heavily spending on training new models, improving performance of existing software and hardware, and trying to create niche products.
Power usage for inference will drop dramatically over the next decade, and more models are going to run on-device rather than in the cloud. AI is only going to become more ubiquitous, there's 0% chance it 'fails' and we return to 2020.
Only because companies have been cutting costs for decades here. This is not a good argument for AI.
> writing software
If you mean typing characters quickly, yes. Otherwise, there’s still a lot of employed devs, with many AI companies hiring.
> writing docs about software
The most useful docs are there because they contain info you cannot determine from the code. AI is not able to do this.
> computer graphics (animation, images)
If you are producing slop, yes.
> driving cars
True, but only because of its improved physical awareness. ie it’s a mechanical gain (better eyes, ears, etc) not an intellectual one (interpreting that information). Self driving cars aren’t LLMs and not really applicable here. Entirely different field.
> AI is only going to become more ubiquitous, there's 0% chance it 'fails' and we return to 2020
Absolutely true. But not for the reasons you think.
I think hallucination is grossly overstated as a problem at this point, most models will actively search the web and reason about the results. You're much more likely to get the incorrect solution browsing stack overflow than you are asking AI.
Gemini hallucinated a method name in a rust crate then spent several minutes googling the method name + 'rust example' trying to find documentation about the method it made up. Unsurprisingly it didn't find any, and then it just gave up and commented out the entire function and called it done.
I think the story is less about the GPUs themselves, and more about the interconnects for building massive GPU clusters. Nvidia just announced a massive switch for linking GPUs inside a rack. So the next couple of generations of GPU clusters will be capable of things that were previously impossible or impractical.
This doesn't mean much for inference, but for training, it is going to be huge.
> Been working on a derviative which hooks the VFS to allow dynamically remapping file paths on a per process basis so I can force badly behaved apps to load custom TLS certificates (looking at you Bazil builds in nixpkgs).
Well he said nix so it's probably hardcoded to load from the store. Tampering with the store itself might have unintended consequences if anything else references the same certificate package.
I actually use chatgpt for creating recipes from time to time. I wouldn't be too offended if there's an 'add to amazon' cart button or similar type of add.
What I'm not okay with is being served adds using codex cli, or codex cli gather data outside of my context to send to advertisers. So as long as they're not doing that, I won't complain.
If they start doing that, I'll complain, and I'll need to more heavily sandbox it.
Based on conversations I've had with some people managing GPU's at scale in the datacenters, inference is an after thought. There is a gold rush for training right now, and that's where these massive clusters are being used.
LLM's are probably a small fraction of the overall GPU compute in use right now. I suspect in the next 5 years we'll have full Hollywood movies being completely generated (at least the specialfx) entirely by AI.
Hollywood studios are breathing their last gasps now. Anyone will be able to use AI to create blockbuster type movies, Hollywood's moat around that is rapidly draining.
The problem with all of these, even the most recent one, is that they have the "AI look". People have tired of this look already, even for short adverts; if they don't want five minutes of it, they really won't like two hours of it. There is no doubt the quality has vastly improved over time, but I see no sign of progress in removing the "AI look" from these things.
My feeling is the definition of the "AI look" has evolved as these models progressed.
It used to mean psychedelic weird things worthy of the strangest dreams or an acid trip.
Then it meant strangely blurry with warped alien script and fifteen fingers, including one coming out of another’s second phalanx
Now it means something odd, off, somewhat both hard to place and obvious, like the CGI "transparent" car (is it that the 3D model is too simple, looks like a bad glass sculpture, and refracts light in squares?) and ice cliffs (I think the the lighting is completely off, and the colours are wrong) in Die Another Day.
And if that’s the case, then these models have covered far more in far less time then it took computer graphics and CGI.
There are lots of very good relatively recent novels on the shelf at the bookstore. Certainly orders of magnitude more than there are movies.
The other thing to compare is the narrative quality. I find even middling books to be of much higher quality than blockbuster movies on average. Or rather I'm constantly appalled at what passes for a decent script. I assume that's due to needing to appeal to a broad swath of the population because production is so expensive, but understanding the (likely) reason behind it doesn't do anything to improve the end result.
So if "all" we get out of this is a 1000x reduction in production budgets which leads to a 100x increase in the amount of media available I expect it will be a huge win for the consumer.
Wages and energy have not increased. Tariffs on food are basically non existent for most items.
It's 100% purely supply side pricing, propped up by government spending and credit (which is largely backstopped by the government as well).
I listened to a podcast recently that some 'homeowners' have not made a mortgage payment in years, have no ability to pay, but here are essentially unlimited 'no doc' mortgage modifications available since the Corona time period.
> listened to a podcast recently that some 'homeowners' have not made a mortgage payment in years, have no ability to pay,
How? Somebody is holding the bag here on the mortgage - a bank, probably. And they are fine with not receiving payments? Or is somebody else making payments on the homeowner's behalf?
So, the basic process is 1) Borrow stops making payments. 2) Borrow goes into forbearance for 12 months just prior to foreclosure start. 3) Forbearance ends, borrower cannot make current. 4) FHA steps in to do loan modification. Essentially, they roll the forebeared balance into the loan, payoff the existing mortgage, and issue a new FHA-backed loan, without any income or payment ability qualifications. 5) Repeat the process again.
In judicial foreclosure states, the process can take 12 to 24 months, and longer if contested or if other periods apply. In nonjudicial states, timelines are shorter but still typically 4 to 9 months from default to sale.
Lenders incur legal fees, court costs, property taxes, insurance, maintenance, HOA dues, servicing advances, and loss of interest during the process. Industry estimates often put foreclosure costs in the tens of thousands of dollars per loan, excluding the loss from selling the property below the outstanding balance.
“Writing off the mortgage” is not the realistic alternative. Lenders generally compare foreclosure against loan modification, repayment plans, short sales, or deeds-in-lieu, because charge-offs are accounting outcomes after losses are realized, not an operational substitute for foreclosure.
Yeah, those are some of the programs I was referring to. The 'loophole' aspect that was mentioned on the podcast is that when the FHA does the 'loss mitigation' (aka, refi's the loan), there is not any kind of qualification as to whether the buyer will ever be able to make a payment on the new loan. It's just approved anyway, and the cycle can happen unlimited times.
I think they're looking at adding a means test, but I'm unsure.
