- Tweets 652
- Following 1,067
- Followers 63
- Likes 634
Harold retweeted
There is ~30GW of LLM compute in operation worldwide. The AI industry says it needs to ~4x this number in 5 years. This is both physically impossible and deeply misguided.
The vast majority of improvements in AI do not come from scale at this point; they came from software engineering—training, harnesses, tool integration, loops. And even with all of this effort, what have we really gotten? Where are the productivity gains? Where are all the jobs it’s replacing? Where are all the scientific discoveries? Where are the products it’s made?
NONE of the things we were promised have happened. How is adding more chips going to make this business make sense?
Answer: It won’t.
The core problem for the industry is that engineering doesn’t move nearly as much money as capex. The industrial incentives all point towards building sheer scale rather than better software.
Jensen Huang needs data centers, not better chatbots. He couldn’t care less if it’s “AGI” or not. As long as he’s giving banks an excuse to finance his chips, that’s all that matters.
The hyperscalers and the frontier labs sold scale as the “moat.”
They convinced the market to make a $5T+ bet on it. That was dead wrong and we’re all about to see just how wrong it was.
LLMs are an experimental software feature that metastasized into the greatest misallocation of resources in all of human history.
I am now full doomer—not because chatbots are going to become “superintelligent,” but because we bet the entire economy on the absurd science fiction story that they will.
We are a deeply stupid species.
What do you notice about the other industries?
PHYSICAL INFRASTRUCTURE. All of those “bubbles” left infrastructure that added to GDP over many decades.
Before you say “data centers,” the vast majority of the cost is in GPUs—which begin depreciating rapidly after 2-3 years. Data centers aren’t railroad tracks; they are temporary money furnaces.
I don’t know how we get out of this without a great depression. It’s the most absurd mass psychosis in human history.
6
19
40
2,575
Harold retweeted
stepping out on a limb to say that a lot of folks here could be overestimating the near future progress of AI. Seeing a lot of capitulation, awe and wonder, felating-the-Basilisk and it all has the same character of the "Bitcoin is going to $2 million" from a few years back
13
2
54
2,741
Harold retweeted
LLMs are trained on human feedback to mirror you.
They are programmed to accommodate your wishes.
The entirety of the internet is in the training data.
There is no difference to a chatbot between a mass murderer and a researcher—its algorithm will try to please you either way.
Giving any rando on the planet access to a supercomputer with a sycophantic linguistic interface was always a terrible idea. The chatbots won’t kill us—but the psychotic people they create will.
4
46
119
3,670
Harold retweeted
The mass delusion we're experiencing is a product of the decades-long campaign against the humanities in general and critical theory in particular.
These people simply *do not know how to think*, except by doing statistics
"The simulation machine does such a good simulation it must actuallybe real" is the best argument that their most brilliant mind can come up with
2
12
67
2,734
Harold retweeted
It's spreading: Meta data center bonds crashing to B2/B levels (7.5%) despite A+ rating. Market calling bluff on residual value guarantees.
Oracle Bonds Plunge To Record Low As $30 Billion CapEx In Peril On Data Center Turmoil zerohedge.com/markets/oracle…
59
319
1,582
202,649
Harold retweeted
1) Resign from lab, citing safety concerns
2) Immediately post link to blog.
It's that easy, folks.
3
3
35
1,067
$META bonds fall to speculative junk credit rating levels as company faces severe bankruptcy risks in the next 5 years.
As I promised months ago.
It's spreading: Meta data center bonds crashing to B2/B levels (7.5%) despite A+ rating. Market calling bluff on residual value guarantees.
5
17
109
6,181
Harold retweeted
What happened to parents groups. If you released an album in the 80s that mentioned Satan every mom in the U.S. would be trying to kill you lmao. Now they let these companies mind rape their kids and no one cares because the adults are addicted to it too. Fucking crazy
1
2
25
480
Harold retweeted
There are a few people like Ezra Klein, Nate Silver, and this guy who got super weird and started being controlled opposition.
That’s because they joined a doomer cult, and are mad that people are figuring it out.
In Effective Altruism, this is called an “infohazard.” It’s information control, and it’s common to all cults.
EA is just Scientology with a slightly better sci-fi backstory. Instead of Xenu that kills everyone, it’s Claude.
I think referring to people who are saying “there are non-trivial but sub-50% odds of something going badly wrong” as “doomers” is a bad epistemic practice in a pretty obvious way that makes me think less of arguments and analysis from anyone who does it.
8
35
121
6,568
While I disagree, I understand the arguments for considering what the LLMs do as thinking and think it’s a reasonable stance.
This position, however, is alien. Why would creating thinking machines, even ones that think better than us, unseat the importance of thinking for man?
The natural reaction now is to defend thinking as uniquely human. An alternate response is that thinking has borne enormous weight in our view of what makes us us, that pillar has been dealt a real blow, and it must now bear less weight than before, and something else bear more.
3
2
12
766
Harold retweeted
Those impressed by the "breakthroughs" in research eagerly announced by AI frontier labs apparently fail to consider how enormously monied they are. If regular academic labs enjoyed that sort of funding we'd see great things too but with considerably better quality and integrity.
16
9
56
3,516
Harold retweeted
“If this thing that will never happen, happens, it will prove my ideology correct”
It pains me to say this, but if we actually “get AGI,” the resulting boom in industrial capacity from robotics and massive society-wide wealth creation will look like a total vindication of neoliberalism.
The discourse about financialization and offshoring being a generational mistake may be replaced by a very different historical interpretation: that the late 20th and early 21st centuries were an enormous capital-accumulation phase, freeing up civilization-scale pools of liquid capital that could ultimately be deployed to accelerate and bring about the most consequential technological phase-change in human history.
In that telling, what looked like deindustrialization and betrayal from inside the period (I grew up in the Rust Belt, I know) becomes the prelude to reindustrialization on an almost unimaginable scale.
None of this is inevitable. The needle has certainly not been threaded. China and x-risk remain enormous contingencies. But if it is, I suspect the future will look back on the last 50 years very differently than we do today.
2
1
18
585
Harold retweeted
I hate this so much. What risk are we even talking about anymore? This has nothing to do with superintelligence extinction risk.
This debate is killing me. I have dozens of kids with zero tech experience arguing with me. Just… fuck.
45
10
195
5,501
