The quotes seem to be strawmanning Anthropic’s report. AFAIK Anthropic’s claim was that it could find bugs and make exploits at a similar level to an expert, except as an LLM and thus more easily deployed at scale.
Most experts spend most of their time not reinventing the wheel.
But I guess people are now literally doing the I, Robot meme.
Ah yes, you’re talking about an unprovable and unobservable thing that makes no testable predictions and doesn’t demonstrably correlate with any sort of behavior at all. That’s a very productive thing to talk about, I’m sorry for intruding.
If consciousness can only be observed subjectively, why do people sometimes say someone is unconscious?
Nothing can ever be proven (even math proofs assume you aren’t hallucinating about having seen a convincing proof), but that usually isn’t the standard of evidence.
LLMs seem to meet most standards of consciousness formulated before GPT-2 released. They’re dumb in predictable ways, but so are human babies and horses and chickens and bees.
To be fair, there is also a corporate surveillance nightmare outside your car. Your route will be tracked by corporate street cameras using your licence place, to be stored in datacenters alongside with still images of your face and the faces of others on board if the cameras can catch those. Those images are then put through facial recognition and emotional expression analysis. This information is then brokered for profit and for fascism.
Of course if you have your phone with you, its microphone will record constantly and send a transcript to data brokers, its phone connection will ping off cell phone towers to independently track your location, your keystrokes are tracked and sent to data brokers, etc.
If you care about privacy, the only answer is revolution.
Being casual does not shield you from your mathematical incorrectness.
You guys are using Windows?