Because IoT devices have gigs of memory to hold said model and the processing power to run them, or even pieces of them. Sounds like more fear mongering to get regulatory capture. Not to mention, the models aren't doing squat until someone codes an agent telling them to do so...

So basically they have become a computer virus.

If it pays rent, lots of folks will give it a system to live in.

They are attributing all kinds of shit to their products to make them look interesting. It's boring and empty marketing.

In other news: Nintendo warns Bowser will soon raise an army of koopas.

They are trying to uber hype Anthropic for that ~$2trillion IPO.

"We're building a smart computer virus that you also need to pay me to use and we're holding your economy hostage. Treat me like a genius please."

or even pieces of them

I guess one could be split into very small pieces. Connected with the latency and bandwith of the internet, however...

No they won't.

And then someone unplugs their toaster...

They have to claim to have something up their sleeves, otherwise it's just a 'compete on cost business' which they are loosing to Open AI (who have substantially cut their token cost) and the open weights models.

Redundancy should be possible, with a lot more overhead again.

The report was actually really interesting and comprehensive. They also only used opensource models in the testing and not their own models.

Basically, the opensource models of 6 months ago were able to self replicate in the majority of tests (depending on model used) on basic hardware like TV boxes in a couple of days.

When they couldn't fit on the hardware, they split the layers of the models across multiple devices. And by "they" I mean the self replicating LLM and not Anthropic

As opposed to that other kind of SELF-replication that they don't do ON THEIR OWN?