This is why IMO blitz scaling is dumb when your service is a commodity. I’m not any more loyal to Uber than Skip. If more investor money goes into making a cheaper meal or ride on Skip I use that. Consumers are mercenaries about that stuff.

The “blitz” part of blitz scaling assumes your customers can’t move.

You’re right if we’re talking about the entire population of Earth. With these local models though, other people have already done all the hard stuff. Anyone with an RTX card and just a minimum level of patience can get going.

There is a middle ground. Crypto farmers have transitioned into running AI workloads for money. There are things sort of like folding@home but you can let people use your GPU and you earn tokens which are used to buy compute or sold to people who want to buy compute on the network. So you can setup a bigass open source model for private on demand use it’s still not cheap but a lot closer to reality for a lot of people than a 15k initial purchase.

If you were paying the real price it would be 2 grand a year though. And in 5 years that 15k graphics card will be $200 and sip on electricity by comparison.

An older co-worker seems to ask AI for help during work, we are blue collar. But the Owner of the company does not seem to use it whatsoever.

I ask Claude on occasion, to see if it will say something smart (it was mostly useless as fuck).

Would be nice to see that used hardware for sale rather than it being junked as a writeoff.

I find that hard to believe, I recently had to uninstall co-pilot after it weaseled its way into my search bar. Its not an exageration to say that my PC literally ran cyberpunk 2077 with pathtracting better than it ran the fucking windows search bar with co-pilot.

It is sorta. Training is orders of magnitudes more intensive than inference, but we infer billions of times within a model generation.

That’s definitely costing them more than running it on their own hardware, but it doesn’t mean AI is costing them more than the AI startups. Anthropic for example is already paying SpaceX 1.25 Billion a month for compute, and has agreed to pay Google 200Billion oflcer the next 5 years for access to Google’s compute and TPU chips.

Google’s deal with xAI specifically lets them terminate the deal with 90 days notice after the end of the year. Google is also investing heavily in building new data centers with their hardware. I’m assuming this deal means they’ve eclipsed their current TPU capacity, and are just looking for a short term bandaid until they can catch up with their new constructions.

Anthropic is doing the same too. SpaceX over here providing the shovels and pans for the modern day gold rush, sheesh.

They have big plans to build more data centers for themselves, so they definitely want more compute than the have access to right now. But even if they’re paying more to rent xai compute, they’re still paying less overall for hardware/access than their direct AI competition.

Claude just kills the other models, it’s not close. Microsoft could ban claude extension from VSC tonight and ill start using command line Claude tomorrow. There’s just no comparison right now. Itd be like Microsoft trying to ban NVidia gpus from Windows, they’ll just lose.

OpenAI and Anthropic aren’t less evil than Google.

They’re all evil, so we just have to exploit the ones that offer us some value. If Google is cheaper, and has the ability to damage the others, then Google it is.

Google is shaping up to fare better than the others, but I dont think that means success. They, too, are spending more than its making, just at a less drunken rate than some competitors.

Exactly, these companies will keep leveraging more and more because they know the govt will step in and print whatever number of trillions of dollars needed to fix the accounting. Then they’ll tell us “core” inflation is only 2.8%.

Of course they don’t, that’s why they’re building bunkers. Thinking it’ll slow us down, as we’ll open their bunkers like cans of tuna. A bunker only works for so long, then the survivors start hunting for them like delicious shipwrecks.

And that’s why they’re trying underhanded tactics to inflate earnings and IPO directly into the index funds, so every American’s 401K will legally have to rebalance and invest in them. They’re racing to fleece retirement funds before the bubble bursts.

Not financial advice, of course :p but people should really consider getting their stuff out and into self-directed funds or whatever it is US people do to not depend on auto-allocated funds.

Money printer go brrrrrrr

That’s simple, because the people making laws and overseeing the adherence to those laws are great buddies with those same CEOs.

So, corruption.

Though i do agree with you, there is no such thing as too big to fail. Government shouldn’t have any handouts to corporations.

Both Uber and Spotify (and AWS too) had economics of scale going for them - the more users they have, the more the infrastructure could be leveraged. This does NOT work for LLMs. More users means using more compute, more advanced tasks (like coding) uses exponential amounts of compute. A single user running a complex task can make 8 Blackwell GPUs run full tilt, and you don’t even have any guarantee that the output will be useable.

There are a few narrow areas where LLMs might be successful, like scanning for security vulnerabilities or searching large amounts of documents. The massive amount of money invested will never be recouped with these usage scenarios.

Although, most people aren’t talking about Alphafold when they’re talking about AI. They’re usually specifically referring to the generative transformer models that are currently all the rage.

I doubt anyone would care too much about a linear regression model, or multi-layer peceptron , for example.

solving something like the Erdős unit distance conjecture

Tell me you listen to media news cycle without understanding what that actually mean without telling me that.

That’s not exactly what happened, isn’t it.

Not to bring up what’s also been accomplished in cyber security

Multiple new vectors of attacks, automation of attack pipelines…

I love that nobody watched anything on Yahoo! Screen except for that one season of Community

Imagine having a streaming service so bad it fails twice

Late to streaming? Netflix was the first big time streaming service that I ever heard of. The main reason their streaming service was able to take off like it did is that nobody else of significance thought that streaming was worth pursuing. What other companies were offering streaming services at anything approaching scale before Netflix?

late to streaming, but practically the first subscription based system to watch movies/tv online.

First years of Netflix were the best, the product began degrading quite early on. but that was mostly companies realizing that instead of licensing their content on Netflix, they can make their own platforms.