Facebook has many employees in the EU. What they try in the US they'll try in Europe if they think they'll get away with it.

A union would allow Facebook workers to push back against company bullshit and back it up with industrial action. And yes unions can be useless and exist to leech off members. But in this case, it is a clear cut example of where a union is needed and it is needed NOW.

A gold sprayed phone that will chip and flake in no time, assembled from the cheapest Chinese parts, with a firmware that will doubtless be terrible and filled with spy/malware.

This is a clear cut situation where workers should unionize to push back on this bullshit.

European employees can complain about violation of their privacy from this tracking. I expect lawyers could also make a strong case that training an AI amounts to constructive dismissal and take Facebook to their country's labour relations tribunals over it.

If software can read (or write) arbitrary RAM in a machine then you've got WAY bigger problems than some plaintext passwords.

Well if they're on a spending splurge they can just purchase new hardware similar to what they're on already, install the XCP-ng (or Proxmox etc.) and migrate most of their VMs with relative ease. Or keep some or all of the hardware. I'm sure some VMs need more thought, precaution & contingency than others but for most it should be a straightforward process. But it makes no sense to jump from one horrible vendor to another if the intention is to NOT be reamed for $$$. Maybe mainframes offer some savings at scale, but also massive investment and vendor lockin.

Moving over to IBM where the bullshit is as strong as Broadcom sounds like an insane choice. Companies already own hardware that they use to run their virtual machines so why not keep the hardware and just switch to different virtualization software like XCP-ng?

How do you "vet" a model and what are the criteria for pass or fail? What is the point of a model like Mythos if some government body could decline it because it's too good at what it does?

And of course there is the usual bullshit of Trump declining AIs based upon corruption in his own administration & personal dealing.

You just know that this "rule" was passed immediately after several senators placed bets on whether or not Congress would pass a rule limiting trading on prediction markets.

I'd love if software like LibreOffice would be able to plug into a model and there be a simple way to kick it off.

The option should be off by default, the experience should not be degraded by being off and the models that the AI feature points at should be "open" insofar as possible and be accessible on or offline. i.e. I can download the model and run it locally if I wish.

Open source really needs a licence akin to GPL for models and their training data to share and share alike. There is an opportunity also here for sites, authors, artists etc. to tag content to invite and consent to using their data for training open models. Will it ever be as good as these massive gazillion billion parameters LLMs produced by commercial companies? No. but it might suffice to do a lot of interesting stuff.

The comparison with audiophiles is apt. People spending ridiculous sums of money on contraptions and gizmos that do absolutely nothing to improve the sound except in their own subjective imagination.

Every supermarket does coffee in various roasts & blends to suit all palates. Find one which is nice and stick with it. It's not rocket science. Coffee snobbery just leads people down the path of wasting money on expensive brands, gadgets & contraptions.

China is eating its own dogfood. And nuclear power might be part of the answer too but power generation can and is a blend of different sources.

And if you're so stupid as to believe solar and wind are a "scam" then more fool you.