It likely will have impact on Trump's crypto coins, and US tech giants' CEOs which were posing with Trump all the time. It almost for sure will have impact on Musk which represents SpaceX. It looks to me that it is partially tech related.

Right, faeries and pixies are feeding carbon free electricity to the grid during nights!

Or are you arguing for nuclear? If so I'm not against it.

I guess you can do something with wind turbines and geothermal too. Well, wind is even less reliable than solar. Solar at least shines every day though very poorly in the cloudy winters. But geothermal could work.

I agree with you about nuclear. I think it is a better option overall. At least it works as well in winter as it does in summer. Nuclear is even better in winter. It works fine even in summer if money are spent on cooling towers instead of heating up nearby rivers. Not sure about fusion. It looks like a pipe dream. But I believe that fission can work well if more research and investment is put into it. And some more standardized regulations across countries. Something which would streamline government approval and eliminate custom modifications.

That being said, solar works OK for homeowners in spring and summer. It does not work in autumn and winter. If you are located near the equator it works OK all the year. Though hurricanes can mess with it.

London latitude is about 52 deg. My friend's latitude is about 48 deg. It is about the same. If cloud cover averages are about the same as well then your power generation will lacking a lot in winter. His experience is that it is about 10x lower around winter solstice compared to summer solstice.

Notice that winters at his location are typically cloudy and summers are typically with clear skies. That does not sound appealing but it is good. Cloudy winter days mean early morning temperatures typically around 23 deg F. Clear sky winter days mean early morning temperatures as low as -13 deg F. I prefer cloudy winter before sunny winter all the time! But winter clouds are terrible for solar power.

Solar works at least for spring and summer. My friend disconnects from grid for spring/summer and uses only his solar and battery backup. He connects to grid for autumn/winter.

You are lucky to live in a location where electricity is still cheap during nights. In my location, the electricity is cheap only during lunch time and very early afternoon. It is very expensive during evening and stays quite expensive until morning.

Well, my 10x claim is very simple. A friend of mine has solar for his house. Enough solar that he completely disconnects from the grid for spring and summer. He reconnect to the grid for autumn and winter. His solar is built from non-movable panels and a battery backup for a few days. He lives at 48 deg latitude. He told me that the production around winter solstice is about 10 times smaller than around summer solstice.
I do not really need any studies. I have an approximate number directly from the source which lives without grid about half of every year ... and connects to the grid for the other half.

Turning off the hardware during winter is not an option. What kind of return-on-investment would it be if they do not operate during winter. They many billions on the hardware. Winter electricity price increase is peanuts compared to that. Either the homeowners will run of money or data centers will. We shell see :)

Quite a heavy emphasis should be put on the distribution. There are winters that cover huge areas of the Earth.

Solar isn't enough when the sun doesn't shine, i.e. evenings, nights. You end up having to over-provision capacity in the summer just to have barely enough in the winter - that is, if you aren't near the equator and actually have seasons.

Grand-grand-parent forgot about big additional energy storage.
Solar without storage means no electricity after about 6 PM. Depending on the location one needs solar generating about 10x the required power in summer to have just enough power in winter. That 10x is for latitude around 45 deg without storage capacity for half a year. Good luck with solar only!

Long life and prosperous distillation future to (not only) Chinese open weights models. Hopefully they will not stop releasing them ... anytime soon at least.

Having the code and the training data open would be much better but don't let the perfect be the enemy of the good.

Microsoft at least recognizes that memory safety features of Rust are a good thing and it is worth using (despite existence of unsafe keyword). If it would be according some slashdot users we would be still using assembler because real programmers do not make (stupid) mistakes.

Well, MTP will take a bit of memory lowering maximum inference context. But not much, probably only about 30k tokens. I'm not sure. I know that a friend runs Qwen3.8 27B with MTP and full context 262k on Radeon AI PRO R9700 32GB. He gets speed about 29 t/s. He used some of the Q6 quantizations. Not sure which one. He is likely using Q8 for KV chace, not sure. KV cache precision should be at least the same (preferably higher) than the weights precision. He also switched to Q5 model (at least for a while) because of some bug influencing inference speed (not because of memory size).

Many people consider even Q4 additional error rate acceptable. It is typically recommended as good compromise between model capability and memory required. Depends on the application. Higher compression seems to degrade reasoning from bigger contexts (e.g 100 kB of loaded source code) more than just answering a question directly. But Q8 seems to be as good as F16 almost always.

Q3_K_XL model with Q5_1 KV cache without MTP will give you about 110k token context on 16 GiB VGA., 21 t/s speed at low context (will drop to about 16 t/s when context is filled to about 60k tokens).

Yeah, 32 GiB is on the low side, but it is usable with small models. E.g. Qwen3.8-27B-Q6_K_M has KL divergence to F16 only 0.002 and will allow for full training context. It will do for some applications.

The subscription is a tax on stupidity :D
The more clever ones just install llama.cpp (or something similar).