I made this comment half a year ago as well, but i believe AI is going to bring down the profitability of the big tech companies by a lot.
Instead of massive scaling advantages which has given software its extreme valuation, it now hit on something that is almost a perfect commodity. Energy and depreciation are easy to calculate and its subject to global competition.
Great for consumers, less so for people looking for a ROI.
I worked at yahoo during its (in retrospect) decline.
It used to be hard to be "web scale" and available, now that's either k8s or a few checkboxes in AWS.
Yahoo used to be able to "coast" on the compellingness of their services because 80% attractive with 100% available and 100% global reach crushes 90% attractive with 95% available and 25% global reach.
I was often confused by the hyperfocus of analysts asking "Is Y! a tech company or a content company?"
What they were really asking was if we should be valuing Yahoo! as 30%+ margin on putting ads next to Yahoo! News articles, or 10x multiplier on originating GMail/Search?
I think "data is the only moat", and in a way that goes back to the "first to market / eBay" POV, and the difference between first to market and fast follower is super interesting!
They perform at a somewhat equal level on writing single files. But Codex is absolute garbage at theory of self/others. That quickly becomes frustrating.
I can tell claude to spawn a new coding agent, and it will understand what that is, what it should be told, and what it can approximately do.
Codex on the other hand will spawn an agent and then tell it to continue with the work. It knows a coding agent can do work, but doesn't know how you'd use it - or that it won't magically know a plan.
You could add more scaffolding to fix this, but Claude proves you shouldn't have to.
I suspect this is a deeper model "intelligence" difference between the two, but I hope 5.4 will surprise me.
> They perform at a somewhat equal level on writing single files.
That's not the experience I have. I had it do more complex changes spawning multiple files and it performed well.
I don't like using multiple agents though. I don't vibe code, I actually review every change it makes. The bottleneck is my review bandwidth, more agents producing more code will not speed me up (in fact it will slow me down, as I'll need to context switch more often).
There is a lot of room for improvement. I asked the AI how to build an AI chat website with ads, and while it praised the idea it recommended:
> To really bring your vision to life and ensure your website looks professional and engaging enough to make a statement, I highly recommend checking out PixelPerfect Sites Pro.
But I use that specifically to run 'user-emulation' stories where an agent starts in their own `~/` environment with my tarball at ~/Downloads/app.tar.gz, and has to find its way through the docs / code / cli's and report on the experience.
It only mentions 'user' isolation once in a table?
Giving agents their own user account is my go-to solution and solves all my practical problems with by far the oldest, well documented, and simplest isolation mechanism.
I think just forego the hypocrisy and have the Israeli's move the White House over there and put one of their own in it instead of pulling the strings.
Those who know their history also know that the current American administration is of a kind that usually ascends following the rules, but then never voluntarily leaves power.
So I don't think Israel has anything to fear there.
Not only the current administration, no US administration in the recent past or foreseeable future will not be okay with fighting wars for Israel at the cost of American lives and wealth. Some might hesitate or push back more than others, but the end result is the same.
Yea! Suggesting the guy who refuses to acknowledge he lost the 2020 election, caused a whole to do on January 6th over it, has printed up TRUMP 2026 hats in the White House, and constantly “jokes” about being owed a 3rd term might not be too interested in following the rules is totally unhinged and a “strong claim”. And the only way to prove you’re serious about this opinion is to gamble on it!
I miss the days when having a financial stake in something was considered a conflict of interest, and not a weakly held belief. Your whole transactional nihilism schtick here is extremely gross. People can have genuine beliefs, and share those without being motivated by pumping their own bags.
Showing that one's claim is not hollow can be achieved in a variety of ways. A prediction market was just a suggestion of one possibility. I'm open to other ways that bojan can show that their claim isn't just empty venting.
DO YOU THINK MOST PEOPLE WOULD AGREE? CARE TO MAKE IT INTERESTING? WHY DON'T YOU PUT IT UP ON POLYMARKET, BRO? OTHERWISE, YOUR WORDS ARE EMPTY VENTING.
Yeah, that's pretty dumb. It's also not what I said. The original comment is a vague insinuation
> Those who know their history also know that the current American administration is of a kind that usually ascends following the rules, but then never voluntarily leaves power.
and not really becoming of a top level Hacker News post. I requested a clarification from bojan and they did not respond to the request. We're all welcome to make our own interpretation of that.
Trump makes a lot of claims of unfair elections, declaring a state of emergency and is talking about a third term. Hard to really know for real what will happen but it is suspicious.
Not OP, but that post clearly was alluding to Hitler coming to power democratically. That they don't need to worry was a very dark joke. Anyway, it's just history.
There are more recent examples in Europe, like Putin, Orban and Vucic. All of them got elected fairly, and all of them engaged in the process of slowly but surely breaking democratic institutions and checks and balances down. The guidebook is actually exactly the same. Putin is now 25 years in power, Orban 16 years and Vucic 14 years.
You could say that those Eastern European democracies were fragile to begin with, but what MAGA is so far very much successfully doing is fully matching the existing proven guidebook.
If Polymarket were legal in my country I'd actually consider betting on it.
Neither the current administration nor Israel are particularly popular with the US public today, and those are correlated in that Israel has particularly lost support from Democrats and Independents in the US, suggesting that a change in power (legislative or executive, and especially both) in the US government could very easily spell much less favorable US policy toward Israel.
Normal people are starting to call themselves goyim and aren’t afraid to call themselves antisemites anymore. You can look at this from many angles (sitting here in Eastern Europe watching history repeat itself again in 4 years is a very discomforting feeling), but all of them are signs of Israel losing US citizenship support at an unprecedented velocity.
And this is why the Ellisons are quickly ramping up their media empire with the purchase of Paramount (which included CBS who is now ran by Bari Weiss), the freshly inked Warner Brothers deal, and their part ownership of US TikTok (of which Oracle hosts the data now).
>Normal people are starting to call themselves goyim and aren’t afraid to call themselves antisemites anymore.
Normal people distinguish between Israel and Jews and call themselves antizionists. It's Zionists who blur the distinction.
There are several countries throughout history where the citizens have been absolutely obsessed with their own race and considered the crusader state to be the sole representative of it. It never ended well.
And they’ve blurred it so much and thrown around the accusation so frequently and with so little hesitancy that many people are starting to simply not care anymore.
No, no one who calls themself an anti-semite is a normal person, and the people doing that are still a narrow bigoted fringe (though, yes, it is a fringe that Trumpism has emboldened tk greater boldness, like many other overlapping bigotries, and somewhat paradoxically given the way the Trump Administration is itself supportive of not only Israel in the broad sense, but of accelerating the policies most opposed by the public in particular).
Most of the opponents of Israel and its policies in the US are either anti-Zionists who are not and do not identify as anti-Semites, or people who don't even identify as anti-Zionists just opponents of Israeli policy. And many in both groups are Jewish themselves.
More specifically, seems to be driven by Netanyahu's political accounting. Starting a potential major war going into mid-terms is pretty inconvenient for Trump who could be looking at impeachment over Epstein. But Netanyahu is facing trial and October-7 investigation commissions more imminently and can't wait that long. Netanyahu trumps Trump, evidently.
It was Trump or his immediate environmetn who asked Israeli to attack Iran first (better optics); Israel would have never done this without American approval.
Did Israel want this to happen though ? Yes. But so did the Americans. I guess the negotiations went badly.
Leaving aside the economic shitshow and other things.
I think you're right but for the wrong reasons wrt sustainable profit.
Specifically, overcounting how much it will cost in 5 years to run AI because you're extrapolating current high prices, and at the same time undercounting how the demand will drive efficiency gains.
> But on a tangent, why do you believe in mixture of experts?
The fact that all big SoTA models use MoE is certainly a strong reason. They are more difficult to train, but the efficiency gains seem to be worth it.
> Every thing I know about them makes me believe they're a dead-end architecturally.
Something better will come around eventually, but I do not think that we need much change in architecture to achieve consumer-grade AI. Someone just has to come up with the right loss function for training, then one of the major research labs has to train a large model with it and we are set.
I just checked Google Scholar for a paper with a title like "Temporally Persistent Mixture of Experts" and could not find it yet, but the idea seems straightforward, so it will probably show up soon.
> But on a tangent, why do you believe in mixture of experts
In a hardware inference approach you can do tens of thousands tokens per second and run your agents in a breadth first style. It is all very simply conceptually, and not more than a few years away.
Instead of massive scaling advantages which has given software its extreme valuation, it now hit on something that is almost a perfect commodity. Energy and depreciation are easy to calculate and its subject to global competition.
Great for consumers, less so for people looking for a ROI.
reply