It's likely possible to extract model weights from the chip's design, but you'd need tooling at the level of an Intel R&D lab, not something any hobbyist could afford.
I doubt anyone would have the skills, wallet, and tools to RE one of these and extract model weights to run them on other hardware. Maybe state actors like the Chinese government or similar could pull that off.
I wouldn't call that size a small power bank. That chip is in the same ballpark as gaming GPUs, and based on the VRMs in the picture it probably draws about as much power.
But as you said, the next generations are very likely to shrink (especially with them saying they want to do top of the line models in 2 generations), and with architecture improvements it could probably get much smaller.
Top of the line models will need more weights and more transistors, so the shrinking factors will be competing with growing factors, I'd expect them to keep maxing out the ASIC sizes to whatever is economically feasible.
Maybe they're numbering the models based on internal architecture/codebase revisions and Sonnet 4.6 was trained using the 4.6 tooling, which didn't change enough to warrant 5?
I do wonder why Apple chooses not to lock down the Mac to just Mac OS like all their other hardware? I'm sure the sales from people who intend to run something other than MacOS look like a floating-point error on the scales Apple operates.
You replied to your own question. Locking down the system for 3 users worldwide and making sure it stays locked is not worth the effort.
Just not publishing the specs is enough to delay so much the effort that those machines are out of warranty and have depreciated so much by the time they are supported that they aren't competitors to the mac ecosystem anymore.
I don't think it is possible to have a locked down development machine. You have to be able to run arbitrary code on a development machine so they can never lock it down like iOS is.
There are plenty of other ways they can be less open and hackable than Linux but it can never get to the point of the iPhone.
That’s a reasonable take. The never part seems strong though.
If I may offer a slight consideration? “arbitrary code vs arbitrary signed code”.
What’s realistically stopping Apple from requiring all code and processes be signed? Including on device dev code with a trust chain going back to Apple and TPU / Secure Enclave enforcement
They don't because it's a floating-point error now. But with the continued enshitification of MacOS, it likely won't be in the future, and they just may lock it down. But being so hostile to the hacking community would do more harm than good, so I doubt that they would do so even if Linux use on Macs grew to >1%.
I've found that doing this on laptops is often more problematic, the OS itself will usually boot fine, but you might have issues with drivers for supporting hardware like the GPU, audio, etc.
I recently helped my GF by proofreading something she wrote, which is a primarily Hebrew (RTL) Word document with English terms like units, numbers, and unpronouncable chemical names sprinkled in.
If I had a dollar for every time MS Word failed to correctly handle the BIDI mix and put things in the wrong order, despite me reapeatedly trying different ways to fix it, I'd be richer than Microsoft.
On the contrary, Google Docs, LibreOffice, and pretty much every text box outside of MS Office can effortlessly handle BIDI mixing, all thanks the Unicode Bidirectional Algorithm [1] being widely implemented ans standardized.
> Flatpak is on life support; there isn't much going on there but it still works.
Since when is Flatpak on life support? Everybody (except Ubuntu) is pushing for it, from the regular desktop distros like Feodra all the way to image based distros like KDE Linux and Fedora Silverblue.
I doubt anyone would have the skills, wallet, and tools to RE one of these and extract model weights to run them on other hardware. Maybe state actors like the Chinese government or similar could pull that off.
reply