And you really don't have to. I just went to one a single meeting, signed a single piece of paper, paid 60€ (probably still <100€ these days) and was done with it. But as GP already alluded to: I didn't choose the structure that is very well known to be the complicated one.
> I'm not sure what this obsession is at the moment with 'fat pensions'.
Pretty sure it's a concerted effort to force the EU countries to adopt private/financialized pension schemes and abandon the still dominant socialized model. The huge new inflow into the financial markets would make some people very very rich, and conveniently would also transform low pensions from a societal problem into a problem of bad personal foresight instead.
(Not accusing GP in particular, they probably just picked it up in the media)
Many people, especially in the tech field, have false perceptions about the inner workings of the human brain. We aren't rational automatons receiving exactly the data that has been sent out by the other automatons. That's for a multitude of reasons, the most obvious of that is the fact that only a very miniscule part of our thought is conscious (about 2% is the last number I've read about it). Even the fundamental inner workings of the brain differ from the idea you alluded to. Our brain isn't just a parser interpreting the data we receive - instead it is a black box constantly predicting what happens next, and only uses sensory input, both from the outside and from the inside of the body, to validate or falsify the prediction [1]. One of the obvious side effects if this is for example our tendency to ignore facts that don't fit to our current worldview.
So if I know that these things are as they are, and use them to communicate more successfully, is that manipulation? Then it would also be learning manipulation if kids are sent to school to learn how to write well, or how to do a presentation.
I had a situation with my kid a while ago. They were already tired, but had to take a shower. When I proposed that verbally, they denied. Then I showed them the warm water coming out of the showerhead, and they instantly agreed. So I got what I wanted (the kid getting clean), because I knew how to communicate successfully. But that isn't manipulation: I didn't lie, I didn't have a personal advantage at their cost etc. I just made it easier for them to anticipate what taking a shower would feel like.
So perhaps the distinction should be: If I can honestly and wholeheartedly argue to myself that my intentions are to the best of all participants, then that is communication. If I only care about my outcome, or even want to have adverserial outcome for the others, then that is manipulation.
But we can't use "not noticing" some mode of communication as part of the definition of manipulation simply because we all notice almost nothing consciously, compared to the sensory input we get every second of our lifes.
[1] A pretty approachable book about that, written from a researcher: How emotions are made, from Lisa Feldman Barret
Ok, I choose to not use Visa/Mastercard in the US, and I want to subscribe to some saas. What do I do now? Or do you mean "choice" as in "you can always choose not the breathe or eat"?
The whole "race" narrative is silly. It is all built on the assumption that one country (corporation, actually) somehow creates AGI and thus, essentially, the singularity. Great for raising VC, apparently, but at its core this is magical thinking.
Even if any of the US corporations would eventually end up in a scenario where their revenue is at least as high as their inference cost, what harm would that do to the other contenders? It's not as if there is any kind of network effect here that would exlude them from market participation.
Fair enough, appreciate the detailed response! Can you elaborate why other quantizations weren't affected (e.g. bartowski)? Simply because they were straight Q4 etc. for every layer?
I currently run the qwen3.5-122B (Q4) on a Strix Halo (Bosgame M5) and am pretty happy with it. Obviously much slower than hosted models. I get ~ 20t/s with empty context and am down to about 14t/s with 100k of context filled.
No tuning at all, just apt install rocm and rebuilding llama.cpp every week or so.
The privacy/data security angle really is important in some regions and industries. Think European privacy laws or customers demanding NDAs. The value of Anthropic and OpenAI is zero for both cases, so easy to beat, despite local models being dumber and slower.
To add to that, the current take that the US could just walk away from the conflict is incredibly naive - Iran will decide when this is over, and it won't be before the November elections. Before the US attacked, blocking the strait was only a potential, now Trump gave Iran the chance to prove that they are capable of doing it. And why on earth would Iran now give that away for free?
In Germany there was zero investment into the electric infrastructure, but the power allowed to flow from the panels into the grid is currently limited to 800W for this type of system. Seems to work fine. Larger systems still need a license.
They paid about $10B on inference and had about $10B in revenue in 2025. The users and numbers of zeroes on those numbers are not relevant. What is relevant is the ratio of those numbers. They apparently are not even profitable on inference, wich is the cheap part of the whole business.
And cost of inference tripled from $3B in 2024 to $10B in 2025, so cost of revenue linearly grows with number of users, i.e. it does not get cheaper.
Of course they bundle R&D with inference pricing, how else could you the recoup that investment.
The interesting question is: In what scenario do you see any of the players as being able to stop spending ungodly amounts for R&D and hardware without losing out to the competitors?
But only if you ignore all the other market participants, right? How can we ever reach a point where all the i.e. smaller Chinese competitors perpetually trailing behind SOTA with a ~9 month lag but at a tiny fraction of the cost stop existing?
I mean we just have to look at old discussions about Uber for the exact same arguments. Uber, after all these years, still is at a negative 10 % lifetime ROI , and that company doesn't even have to meaningfully invest in hardware.
IMO this will probably develop like the railroad boom in the first half of the 19th century: All the AI-only first movers like OpenAI and Anthropic will go bust, just like most railroad companies who laid the tracks, because they can't escape the training treadmill. But the tech itself will stay, and even become a meaningful productivity booster over the next decades.
Anthropic seems to want to both own and eat its stolen cake.