she hacked youOct 25

"AI models may be developing their own ‘survival drive’, researchers say" - Guardian

The "research paper" was a tweet by an AI company

The "experiment" was asking the LLM to shut down

A model is ~not~ shutdown ~ever~ by asking a model to shut itself down

*The only possible response is a hallucination*

You shut down a model by turning off the deterministic software running it; so works every time w/o fail

Yet Guardian's shill tech writers just report AI industry tweets as if it was fact

Show threadCassandrichOct 25
@ekis The way you make an LLM able to shut itself down is by having the company that owns it put all its assets into magical toy money and including the private key for the wallet in the LLM's training corpus.
Show threadJohann-Tobias Schäg
@dalias @ekis
I think giving an LLM access to an authenticated cloud shell to spawn new compute and a task like "develop AGI" is also good to hasten the shut down.
Oct 25 at 11:11pmWeb