Also it's a sales tactic. "our AI is so clever, it's so scary!" wink wink

It just means they don't invest enough in proper sandboxing and does NOT diminish the abhorrent resource usage and blatant data theft and plagiarism committed by these companies, while they poison the well of information that is the internet for everyone else.

Meanwhile their faux experts feed us questionable information, but you should always check it yourself (right). If I ask any other expert for advice that turns out to be blatantly false, you would keep them accountable for damages.

The danger is that we can't seem to hold them accountable for anything, not that the thing is "dangerously clever". If someone builds a bad bot (regular software or AI/MCP/harness), they should be fully responsible. Just don't run it if you can't guarantee safe operations.

That it's hard for them to properly credit where their training data came from should be their problem and a hard prerequisite for publishing or allowing usage of models. Ripping off everything ever published online has never been OK in the first place. They are 'asking for forgiveness later' and should definitely not get it.

The training and exploitation of these models should be subjected to pollution tax just like any other industry. Water is becoming more and more scarce around the globe and humongous electricity consumption isn't magically pollution free because it isn't fossil fuels. The precious metals in all chips and datacenters are even further from "clean".

Don't even get me started on the societal and psychological impacts...

These companies (and to an extent their users) will never self-regulate, we need our governments to actually step up here.