For those of us who operate on site, we have to add back network latency, which negates this win entirely and makes a proprietary cloud solution like this a nonstarter.
Often not a dealbreaker, actually! We can spin up new tpuf regions and procure dedicated interconnects to minimize latency to the on-prem network on request (and we have done this).
When you're operating at the 100B scale, you're pushing beyond the capacity that most on-prem setups can handle. Most orgs have no choice but to put a 100B workload into the nearest public cloud. (For smaller workloads, considerations are different, for sure.)
Try implementing something that is too hard for you. Usually that'll involve implementing math in a high performance language or with parallelization. Then try going back to "writing by hand".
Nothing that is possible is "too hard" if you're willing to put in some effort. The only question is whether you will learn to do it, or press a button, hope the LLM did it well, and let it forever remain "too hard."
Honestly, without judgment, I think this is just a fundamental difference in how people approach their craft. You either want to be capable yourself or you just want the results.
if it’s too hard for you to write, it’s too hard for you to understand and comprehend. how are you going to take responsibility for that code and maintain it if needed?
> Try implementing something that is too hard for you.
This is almost the only thing I'm against when it comes to LLMs. You have no ability to figure out if it is right, and you will be overly impressed by garbage because you aren't qualified to judge. Has anybody come up with a pithy way to describe Dunning-Kruger for evaluating the output of LLMs, or are people too busy still denying that Dunning and Kruger noticed anything?
When it comes to implementing math, the main problem is that the tiniest difference can make the entire thing wrong, often to the degree of inverting it. I wouldn't be in any way comfortable in shipping something I didn't even understand. The LLM certainly didn't understand it; somebody should.
It’s like announcing you’re going to sell corn starting a year from now when your competition owns all the land corn is grown on, started selling their corn 6 years ago, and has gotten really good at making it cheaper and producing more, based on real world experience.
“Despite the welcome relief, climate change is expected to intensify weather swings from heavy rainfall to extreme dryness in a cycle that can fuel catastrophic wildfires.”
…but we’re still fucked and don’t you dare forget it!
The dollar has lost 13% vs the South African Rand in the past year. Thats an interesting metric for me given how South Africa has decoupled from the US in favor of China and Russia, with a strong anti-Israel stance. And also taking into consideration how the country is still struggling with crime, corruption and uncertainty.
It’s doing to be interesting to see how this plays out as the producers and makers of the world unite, and the EU turns away from the US.
>given how South Africa has decoupled from the US in favor of China and Russia
That's the party line & there have been some eye catching military exercises and statements to this effect, but daily life is still very much western aligned. It exports a ton of ore to china and imports a bunch of things from there certainly...but that's kinda just every country these days isn't it.
But is the exchange rate really the right metric for the "weakening" due to de-dollarization? Especially since world states decoupling from the US would try avoiding both transactions which gain them USD and which require paying USD?
For context foreign total holdings are around 9T and domestic around 30T. Uk and japan are the largest foreign holders at around 1.7T between them. This is a 100M divestment.
It’s as if our country has trained millions of our citizens in the art of harming others and destroying their communities and societies. And then gamified it for our kids and created entertainment that glorifies and celebrates it. We’re so fucked.
I’m fascinated at how out of touch that NYT article is. It’s as if it was written by someone who just spent 3 years on the moon. “The next big thing will be agents: The models will fill digital shopping baskets and take care of online bills. They will act for you.”
Golly Mr two times Pulitzer nominee, do tell us more!
There is a kind of liberal arts elite that seems to not be using AI very much and not be buying any of their services. Contrast that with those of us in tech who are handing over money as fast as we can and can’t get enough of gpt 5.2 codex on xhigh and similar products that are game changing enablers.
Makes me wonder if we’re seeing a fracture in society beginning to form where the doomsdayers, naysayers, cynics and skeptics will realize their error too late.
My view on AI is that this is the world’s first Unbubble: where the majority view is that we are over invested, but where history will show we actually underestimated future revenue and profitability.
The conditions for the Unbubble are perfect. We have a once in a species level innovation with an economic system where all value accrues to the creators and financiers. And we have just emerged from the housing bubble and the dot com bubble in the last 3 decades, freshly scorched.
We thought connecting everyone would create new value far faster than it did. But really it took a long time to run all that fiber and make it fast, and it was just laying the plumbing for this moment.
Training big foundational models may seem slow, but it’s happening way faster than circling continents and the globe with fiber and developing terabit switching fabric.
I spend 12 hours a day using codex CLI to write extremely fast Rust and cuda code with advanced math that does things I didn’t think were possible. My focus is on creating value from the second and third order effects of AI. These enabling effects are in few conversations. As weirdly innovative products emerge from small shops, they will begin to be discussed.
I have no idea why so many people think that an argument that AI works is the same thing as an argument that AI will be profitable.
The better AI gets, the better the training techniques get, and the better the algorithms get will result in fewer processors needed to run something of use. All of the advances will end up in the public domain if not immediately after or before they are even implemented, soon after. There will not be many economies of scale between having 100M customers or 10K customers, so no way to keep out competitors. They will all compete on price. If the models get really, really good, a "good enough" model will end up running on your old laptop and you won't have to pay for anything.
Saying that AI will be productive - which is yet to be seen, I don't know how much polishing or complete rethinking your code will have to go through before it can ship as an actual product that you have to stand behind and support - is not the same as saying that AI will be profitable.
We actually don't even need that many computer programs. Hypothetically, a ton of excess LLM coding supply might allow us to take out a few layers of expensive abstraction from our current stacks, and make more code even less necessary. They kept telling us that all of that abstraction was a result of trying to save developer labor costs, right? If AI is productive and rentiers can't manage to extract that productivity due to competition, that equation changes.
In the end, we say that the dot com bubble resulted in a huge amount of productive capacity that we were later able to put to use. But that doesn't mean that putting a quarter of a billion 90s dollars into DrKoop.com was a good idea; nope, still dumb.
> I have no idea why so many people think that an argument that AI works is the same thing as an argument that AI will be profitable.
The fact that it works well for expensive categories of output (like software engineering, legal strategy, etc...) makes it difficult to imagine that it won't be profitable. You could still make an argument that the investments being made today are disproportionate, or that intense competition will stifle margins, but it's creating enough value to capture plenty of money.
What did you think about the point in the article suggesting millions of people dying on South America caused the earth to cool down due to reforestation absorbing more green house gas. I find this hard to believe.
There was a noticable (and well recorded) change in the climate for awhile when Ghenghis Khan went on his rampage. And he was a lot less effective at killing people than novel old world diseases.
reply