back
25 comments
LLMs are capable and able to produce value. There's no questioning that - anyone who says otherwise is in absolute denial.

The problem is.... the amount invested makes zero sense and the returns needed to justify the valuations of these firms won't come.

The players in the Tech industry in general have gotten way too ahead of themselves. When this all settles the big danger for MSFT, Google, Meta etc is that investors will trust them with the cash on the balance sheet less - demanding more cash return. Apple will come through this period very strong and end up looking like they played a genius move keeping out of it.

Define "value", because is seems pretty negative value at the moment.
Oh yeah? What value have they generated? How is it valuable to rehash existing data in a non-deterministic stream of nonsense?
Its not rehashing, its recombining and mixing any concept ever in minutes. If you cant find value in that you're not being creative enough
We use local models in data processing pipelines. Being able to do compute with language has plenty of use cases
I used Grok heavily to generate a 5min marketing video that got MP's to back my project.

So... go ahead. Try and claim Im lying. I pity you - I'm not lying.

Do I care about whether xAI/Grok survives as a going concern? Absolutely not. Did it generate value for me? Absolutely yes.

insider influence backed up by deniable narative is hardly a new idea(called a "fix"), though conflating that with AI having "value" is quite a stretch. AI is just another pivot point or prop, hookers, drugs, cash, blackmail, favors, etc will still,

trump, AI glossy slop.

Endgame is IPOing those AI companies and getting them on indexes, forcing index funds to buy them, which seemed to be evergreen investment category, but I’m not so sure anymore..

Did somebody say crypto?

Their backers certainly enough money and political muscle to force this outcome.
Great writeup, I’ve been running my own mitm measuring token usage on antigravity since release. In the past week and with the introduction of ”credits”, the amount of usable opus on the 350$ ultra plan was >4000$ it is now <1200$ which means around 4x less usable tokens.
Every time these shops seem to be out of runway, they do another raise[1]. The investor class is far from running out of funds to support this adventure. They will run out of patience or find the shiny new thing before there's a liquidity crisis. I suspect a good portion of the cash pulled from private credit goes towards further funding of the AI trade.

[1] https://www.nytimes.com/2026/03/31/technology/openai-12-bill...

Inveator class is loving the stress on the normal worker ai is causing and making them work harder for fear of layoffs
Great writeup. Only thing I din't see in here was an analysis of the impact of players like Talaas[1] and their stupid faster hardware LLMs.

I feel like it could be majorly disruptive, but idk if it's going to prolong the apocalypse or bring it about sooner -- or if it's a big nothing burger.

But the demo[2] is super cool.

[1]: https://taalas.com

[2]: https://chatjimmy.ai/

I'm bullist for something like talaas to get smaller and easy to put in a desktop. Imagine an RPG where NPCs.... are way more complex and the entire game is very non deterministic.
I think I would like that as well. The problem is that if we bake an LLM into HW and make it cheaper and very efficient to run, then all games will have the same AI slop content, which could get boring pretty fast. The alternative is that these cards should load a different / fine-tuned LLM per game, but then we already have GPUs for that and today's LLMs are nowhere near good enough at the size which a GPU can run.
They claim to have qwen 3.5 27B on a card at end of year on the market. If they do, I’ll be buying one immediately.
I would respect Ed (and Gary Marcus) more if they would concede the occasional point. But everything AI is always a hyperbolic and unqualified disaster. I suppose that's what the audience wants.

"every bit of AI demand ... that exists only exists due to subsidies"

Really? NOBODY would pay whatever the fully-loaded cost is? What about people running local models on their own GPUs? Are they being subsidized too?

> Are they being subsidized too?

Yes, because they didn't pay the cost to train those models in the first place.

True, good point. I still think the piece goes too far with its claims.
Your loss. Dude is on the cutting edge and 90% of Silicon Valley is in a cult.
I like to read Ed (and Gary) as a counterbalance to the AI hype that is pervasive.

But Ed reads more and more like he has been personally wronged by the AI companies and is on a righteous crusade to destroy them. He has plenty of valid criticisms, but he comes across like he has his head in the sand when he can't seem to acknowledge any potential upside at all related to AI. It's starting to feel like he's just interpreting news and events to fit his worldview, which isn't that valuable as a reader (unless I want that worldview affirmed for me).

And this is all before the AI industry starts to feel the real impact of the ongoing oil (and helium) shock that Trump caused for no coherent reason.

Going to be an interesting next couple of years...

It’s gonna be funny as long as taxpayers don’t end up bailing the likes of scam Altman. I wouldn’t put it past Trump and the GOP to absolutely loot the place on their way out pre-November and cut their tech bros in on the action. But that means the VC SF dipshits better speedrun this collapse because they’ve got exactly 7 months to become a “systemic risk” to the US economy.
> I wouldn’t put it past Trump and the GOP to absolutely loot the place on their way out pre-November and leave their tech bros out to dry.

I wouldn't be surprised if this happened instead.

> What use is Perplexity without an eternal subsidy? The value of having Aravind Srivinas sitting around your office all day? I’d rather start my car in the garage.

Lmao