Ugh – I feel like the LLM labs are starting to tune these things to operate “on our computers” whereas I wish they would introduce stuff to have them “operate in controlled VMs” with gated access to things.