Well, your computer is a Von Neumann machine that does not separate data from instructions, right? The techniques that secured it are not directly applicable to LLMs but in principle there must be ways to track them.
All the while, Linux is going towards reproducible builds (Debian just announced it as a policy). This is of course the only sane way for FOSS, and, I believe, the only sane long term approach in any case. Security by obscurity, while not worthless, is just a thin mitigation layer.
By the way, build-time randomization is ineffective in light of AI analysis---it needs to be per-binary-run, in the style of KASLR.
Sorry, SM-4 not SM-1, was a full emulation of 11/40, with UNIBUS, and all. There were DEC copyright strings latent in some system files. It was a pretty good copy, but quite unreliable, and the reason was quite pedestrian---the connectors! It was a good lesson on how the entire technology chain needs to be high quality for the final product to work well.
Another example I forgot: the first Soviet nuke was directly copied from the stolen Fat Man design. Of course later they did novel stuff, especially the fusion designs of Sacharov et al.
It is well known that KGB got hold of the Concorde blueprints, so yeah, not a direct copy but certainly a lot of influence in that design. Again. the details like engine performance made the difference: apparently Tu144 had to continuously use afterburners to stay supersonic. It was also quite unreliable---I've heard that towards its end of life it was just flying cargo and airmail.
No, they disassembled German optics industry plants in 1945, moved them to the Soviet Union and started cranking out great cameras based on German designs. I've heard that some Soviet cameras had Leica labeled parts inside.
Stuff like that happened repeatedly: GAZ Chaika was a copy of Packard; SM-1 computer was a copy of PDP 11/34; Tu-144 looked just like Concorde, etc. etc.
Important factor is that AC at given nominal voltage V swings between 1.41V and -1.41V, so it requires let's say 40% better/thicker insulation than the equivalent V volts DC line. This is OK for overhead lines (just space the wires more) but is a pain for buried or undersea transmission lines; for that reason, they tend to use DC nowadays.
An advanced AI rack might use 100kW = 800V 125A, requiring gauge 2, quarter inch diameter---this isn't your lol speaker wire. Actually, I apologize, I realized I may be talking to a serious audiophile, didn't mean to disrespect your Monster cables.
The skin depth by the way is sqrt(2 1.7e-8 ohm m / (2 pi 400Hz mu0))=~3mm for copper---OK for single rack, but starts to be significant for the type of bus bars that an aisle of racks might want.
As for efficiency, both 400Hz transformers AND fancy DC-DC converters are around 95% efficient, except that AC requires electronics to rectify it to DC, losing another few percent, so the slight advantage goes to DC, actually.
As for merging power, remember that DC DC converter uses an internal AC stage, so it's the same---you can have multiple primary windings, just like for plain AC.
Maybe LLMs are like a next evolution of a rubber ducky: you can talk to it, and it's very helpful, just don't expect that IT will give you the final answer.
Pretty much every laptop on the planet will run Linux. Maybe your optics are tinted because you seem to be a Mac person, and Linux support for newer Macs has known issues with low power modes.
I note how your 12+ hour claim was reduced to 5 hours when you actually put it to real work. It's still impressive, of course, but 5 hours aren't out of reach for Ryzen laptops either.
BTW, I have a RISC-V platform with 8 1.6GHz CPUs that uses under 5W under full load; on your 100Wh battery it would last for 20 hours. It's not a complete system, and performance lags behind Apple/Intel CPUs, but I think in few years RISC-V may take a bite out of both.
Linux will run on most platforms, so just pick up a fast, lightweight laptop, and select a conservative power profile for longer battery life and less heat, and don't run 32-thread machine learning jobs on it.
A 12-hour laptop battery life is a little bit of a red herring: yes, you can get it on efficient ultrabooks and MacBooks, with light use like web browsing or office work, on low brightness and minimal background apps. This is true on MacOS, Windows and Linux. The first two may be better at handling low power modes on hardware peripherals, but OTOH on Linux I have a better control over background tasks.
I have an absolute trash travel laptop from last decade, running Fedora Linux, and it lasts for multiple days if I keep it mostly closed and just open it for whatever browsing/editing I need on the road.
You could run on 5V with a boost voltage converter to 12V. For extra credit, you could run the USB-PD off 5V, negotiate 12V and only then switch it to the load.
Normally a wafer would have die-sized spaces for test structures used for optical, electrical, chemical and other tests. Think the TV test card https://en.wikipedia.org/wiki/Test_card
