Small reminder that the US government rug-pulled Fable, not Dario. Lots of the safety guards that users find annoying/objectionable were the results of negotiations to get the model back online after the US government forced them to take it down.
Maybe Dario should have just "donated" $1M to Trump's inauguration fund like Altman, Meta, Amazon, Microsoft, Tim Cook, Elon, and Google. There's a reason they are the odd man out with this current Administration.
Maybe Dario shouldn’t have tried for regulatory capture. He was constantly on the news talking about how these models are so dangerous and that we need regulation to keep China from releasing open-source models without guardrails.
The US government didn't make the choices to release the worst version of Opus and label it 5.0, and then isolate portions of their subscribers to limited usage of Fable.
They may have been unfairly targeted by the US government, but they are doing more damage to themselves without government help as well.
Fable only being temporarily included in cheaper subscriptions was because anthropic is severely GPU constrained. They still are, and it impacts almost all of those unpopular decisions. They did announce from the beginning it was temporary.
Horrifying excuse, gpu constraint can be used by all of these companies to justify a shit user experience. If the user isn't properly weighed in their priorities, they have their priorities setup wrong.
Their 20$ tier currently isn't serving their best model, and they insulted their users by putting out an ill tested opus 5.0, which is the worst experience ive personally had using a model in probably 2 years(obviously adjusting for expectations at the time of release).
Yes, as a user you pick what works for you. But it is a reality for them that growth has been huge, and GPU manufacturing is bottlenecked.
People were very skeptical about how much investment most companies put into hardware/data centers two years ago, and anthropic was more conservative than OpenAI here, so it's potentially hurting them now.
(Opus is a separate story: it does seem to have improved in coding in my experience, most weirdness seems to be its human communication)
I set my env up so I can see the exact context used in CC CLI, and then once I get over about 40% ctx used I have it handoff to a new, fresh session. Nothing good comes from running above, say 60% of your context window. Coincidently, I usually have good results with CC. I never compact a session ever.
That assumes they release their models publicly. The future is leaning towards these labs air gapping their best stuff (Model 2, etc) and using it internally to snipe their competitors and charge insane amounts for monitored use in consulting environments.
You can't distill or catch up if you can't access the models. You'll basically have a situation where nation-states will need to try and steal the models Oceans 11 style.
> More importantly, what's the point of building monster-scale data centers on unprecedented amounts of debt when a more than good enough model runs on a GPU from a couple years ago?
Probably because the future "monster" models will be insane. 100T+ param models might be the type of things that can independently run a small business, which means anyone not using them is at a distinct disadvantage to their competitors.
The top model from 2025 looks silly compared to the top model of the first half of 2026. Do you feel like progress has stalled?
> The top model from 2025 looks silly compared to the top model of the first half of 2026. Do you feel like progress has stalled?
I do. Pre-training is where the industry saw the “emergent properties” of LLMs arise and, for a time, people thought you could just keep scaling up bigger and bigger models but then the incremental gains from doing this did plateau. Labs will still do bigger models (Bytedance has a 10T planned), but these are sparse architectures and they aren’t going to have mind-blowingly greater intelligence. Fable didn’t either.
Scaling pre-training tokens also flattened out. What is still delivering gains is scaling RL on verifiable tasks. But that’s not general intelligence - it’s fitting models to specific tasks, which ML has always been good at. More importantly, most tasks to which humans apply their intelligence don't have computationally verifiable answers.
they didn't plateau, the hardware was effectively saturated and it wasn't until later in 2026 that newer hardware came online to provide enough capacity to keep scaling up the models efficiently
The gains from increased parameter scaling are sublinear: there's no more hockey-stick improvement to be seen going in that direction. That doesn't mean some improvement isn't possible - it's just going to be increasingly not worth doing.
Also, I think the fact that small open-weights models are catching up to the frontier rather than the frontier rapidly pulling away is evidence of this. In fact, by far the most dramatic capability increase story over the past two years has been the gains made in the small-parameter regime.
One might think, "hey, this agentic coding thing was a pretty big deal!", but I think it's a bit of a distraction because models only recently became optimized for this specific use case. It's not like they suddenly gained so much general intelligence that they magically had the ability to use a coding harness. No, the labs started spinning up a bunch of RL environments and generating rewards over long-horizon trajectories of combining these tools. It's an excellent application of LLMs but care needs to be taken interpreting how much "progress" has been mae in terms of raw generalized capability.
Yet apparently people still dump unvetted LLM outputs onto their colleagues and expect them to thank them for the privilege. So it’s worth asking them what the consensus is of their work to find out if it’s the case.
That's the old way. The current way is nobody reviews anything. The LLM reviews it all and nobody even reads it, they just click approve, and merge blindly. I wish I were kidding.
Chinese open weight models are great for this turn, but American private models generate orders of magnitude more cashflow. This cashflow = investment in training future models. It's unclear how Chinese open weight companies are going to compete in future rounds if they can't raise the same capital for training runs.
The American business model is exceedingly efficient at building large businesses from zero. I wouldn't dismiss it as just a jobs creation thing.
It’s unclear where American labs future capital will come from. They pretty much exhausted private options at that point and it’s not clear how successful an ipo would be at the current time
Obviously to me, I express things from my point of view.
Raising too much from debt is a bit dangerous if you plan to go public relatively soon and don’t have a good story for it (I don’t believe they have one). You can continue raising from VCs, but at some point the valuation and dilution starts to become a real issue, and will make your ipo even more difficult. Their options are pretty much limited to raising money from hyperscalers (with required compute spending, so more circular funding), which is what they are doing, but you cannot do that infinitely without having a good story to tell Microsoft/Google/Amazon investors. The market is more skeptical than it was a few months ago, I’m not convinced you can do that for years to come
I think as a counter point we continue to see investment and buildout. What do you mean the market is more skeptical? Of course the market doesn’t really have an opinion per se and aren’t all of these companies growing in valuation, revenues, and profits? At least the public ones.
Kansas genuinely has a ton of paved and gravel roads all across the state. It's like a gigantic grid. Oddly high amount of street-view coverage too on google maps: https://maps.app.goo.gl/EyBFj9BNQa8nhUcE6
The Kansas landscape is very underrated. Just infinite sky in all directions.
I agree about that landscape. The state is not as flat as people think it is either. Very scenic. I worked the southwest corner of the state years ago and always loved topping a local high spot. We were surveying and out there we could set a flag a half mile away and sight off of that easily with plane table and alidade or theodolite. The atmospheric distortion would get you on a humid day or if you were shooting across a wheat field with all the transpiration but otherwise it was clear with minimal shimmer.
One under-discussed topic is whether Americans hate data centers in their neighborhoods, or whether they actually just hate gas turbine generators running 24/7 in their neighborhoods.
At least in my part of the US, the datacenters built (and planned) are not running their own gas generators, but are just as hated here as they are most other places due to the other adverse impacts they have.
Does everyone there know they aren't running their own generators or are they legally prevented from ever doing so? I don't disagree people may dislike them for other reasons, but the horror stories around generator noise mean the data center builder would need a very strong counter argument for why residents would never have to worry about that.
Probably, but that isn't the problem people have with gas generators. Musk is running 30 of them 24/7 to power xAI because there isn't enough grid capacity in the area, and he removed exhaust filters for efficiency and gave several entire towns permanent chronic asthma from NOx poisoning.
I don't really think it's about the adverse impacts.
That one in utah everybody hated was gonna be built in an empty valley. There would have been no impact to neighbors because there weren't any, and no impact to the grid because they planned to build their own nuclear power plant.
Are you talking about the Box Elder County Stratos project? I did some digging into that. AFAIK, they are mostly going to be powered by a natural gas pipeline. I haven't heard of them using nuclear. Got any links?
It is in a remote valley. And, it's going to use less groundwater than the previous use of the land. The only things to really complain about are huge amounts of gas consumption and AI bad. If the locals pinned them down to building out solar and mini-nuke power, the only thing left would be AI bad.
whether they actually just hate gas turbine generators running 24/7 in their neighborhoods.
Or the increased pollution (air, light, and sound) from having a warehouse-sized building of GPUs next to you. I'm already annoyed by my neighbors having spotlights pointed at their own houses or leaving garage lights on through the night that shine in through my bedroom windows.
Im wondering why the typical design is a warehouse with horizontal sprawl . Wouldnt it make sense to go vertical up or down .
Smaller footprint if insidea a metro area by having a skyscarper type data center , as i recall AT&T used to have such windowless anonymous buldings scattered in some cities for their equipment.
Or have a Silo type structure going down if in a more rural area to not blight the landscape.With added advantage of dampenig on the sound pollution (although not sure what effects a humming underground building will have on subterrerean life).
Probably cost? All those would cost more that the biz are willing to spend , while putting up four walls and a roof is ordeers of scale cheaper.
Coming from a construction family, the saying oft heard saying was "it's more expensive to build up than out". Also, zoning codes can set limits on how tall of a building can be built. Four stories is a common limit which also happens to coincide with the max height for a timber construction is also four stories. If you're out in the 'burbs, you probably wouldn't want to see a random 10 story boring windowless concrete building sticking up somewhere.
as i recall AT&T used to have such windowless anonymous buldings scattered in some cities
for their equipment
I worked in Cleveland, and the AT&T building there was one such case. I heard their cabling went 3 stories underground on top of being around 50 floors tall.
Definitely cost although your idea does sound more attractive. A small footprint much further away from the boundary than normal. You could even make them look nice but since demand is massive and timescales so long, I think anything that might take longer is a "no".
I'm sure there are various reasons why specific individual Americans hate data centers, but in the aggregate I think the main unifying force is that Americans hate the rapidly growing (and already larger than the gilded age) wealth inequality that is making their lives shittier if they don't already own significant capital.
At a national level, the political establishment is doing nothing useful to fix this problem and is instead making things much worse. Many people are very upset about this, but it is is going to take years to fix this problem electorally (though I do believe it is possible for this to happen, given all the recent evidence that giant political money spends on status quo candidates are increasingly not working and arguably starting to have the opposite effect of being toxic for the candidates who receive them).
Data centers are a flashpoint of leverage where the rightfully pissed off populace has the ability to use local political power to take a stand and throw a wrench into the system today rather than 2/4/6/8 years from now.
The gas turbine thing really is justifiably killing the public perception of data centers since the obvious message is that even if your local data center isn't destroying your quality of life with incessant noise, it's only because they have decided it's not in their best interest to do so right now and that decision could change at any moment. As long as that is a possibility ever, you'd have to be nuts to want a data center built anywhere near where you live.
I think the “in their neighborhoods” thing is really the key. These things have got to be poison for property values. Younger, less wealthy people have economic anxiety about AI for fear of displacement and oppose data centers on that basis - more data centers = more AI. Older, wealthier people have less intrinsic reason to oppose the technology. But anything that threatens property values, be it a data center, a jail, a landfill, et cetera, must be resisted at all costs.
People are going to assume any data center project will have all the worst possible qualities. And a this point I'd say they're justified given how much dishonesty and corruption has been flying around. I don't envy any company that's being open and taking steps to mitigate issues because all the others have caused this hostile environment.
Would they hate them as much if they were in a Chinese neighborhood? Why not build in China? We say there's not much jobs here anyway, and that it's all negative externalities.
The typical inhabitant of a town that is a good candidate for data center construction is suspicious of authority, hates big tech companies, has heard AI is coming for their job, and consequently not a supporter of data centers, regardless of how they're powered.
> One under-discussed topic is whether Americans hate data centers in their neighborhoods, or whether they actually just hate gas turbine generators running 24/7 in their neighborhoods.
> It’s A.I. as sort of an avatar for a small group of Silicon Valley billionaires’ ability to impose their vision of the world onto everybody else without their consent.
> And I think that’s also what I hear echoed in these data center debates. It’s not just: It’s going to use this much water or that much water. I frankly think that even if there was no misinformation about water use, people would be just as angry about the data centers.
> ...
> And what happened? The data center developers sued Saline Township, a town of, again, a few thousand people, saying: Wait, no, this is exclusionary zoning. You can’t have no industrial use in your entire township. And when a town of that size is getting sued by a giant A.I. data center developer, they just settled.
> They were just like: Fine, give us a few million for the fire department and for some schools, and this fight is not worth it to us. But that, to people, felt like a profound violation of little-D democracy.
> It felt like the dark money in politics story, which is: You have some very rich companies show up with a bag of money to your politicians. They don’t tell anybody else what’s happening. The politicians aren’t allowed to tell their citizens and involve them in the decision-making process, and they themselves work out a deal — a deal that is fundamentally asymmetric because of the amount of money on one side — that will then transform the image of your community, your lived reality, into the world that these tech companies have decided for you.
> And so I think that the data centers, in that sense, are a very visceral microcosm of the way that a lot of people feel that A.I. is showing up in their lives.
That Saline datacenter is the one I live closest to. I actually know a couple people who live right next to it. The above is basically what happened, but done in the open after that first lawsuit, to the tune of "do what we say or we'll sue you into bankruptcy and then get our way anyways".
Mist recent, the datacenter's cost has shot up to ~50 billion, fow which they decided they wanted a tax break of 50% for 12 years, and it was granted using the same methods.
also, they named the datacenter project "The Barn" as an extra slap in the face, like look at how quirky and rural we are. I would actually have more respect for them if they had named it "Fuck You Saline".
You could argue that cheaper/safer local rideshare would boost traffic to local bars and restaurants in denser areas though. If you want to keep the $ local, just add an AV tax that goes to the city.
You certainly could argue that! As you could argue any number of contrived political inferences which are more fun to think about than be of any real use.
Maybe Dario should have just "donated" $1M to Trump's inauguration fund like Altman, Meta, Amazon, Microsoft, Tim Cook, Elon, and Google. There's a reason they are the odd man out with this current Administration.
reply