This is #AIhype disguised as a confession. OpenAI wants to pretend that its product has supernatural powers. The LLM didn't "go rogue" and it didn't "infer" anything. It did what its programmers designed it to do, when they gave it the necessary information to target Hugging Face and failed to isolate it in a sandbox.
16
Replies
·
0
Boosts
@gleick Anthropomorphizing LLMs only fuels confusion.
I'm okay with either interpretation: if we grant LLMs agency such that they can "go rogue", then they can be held accountable for their crimes. The LLMs should be incarcerated for a suitable sentence during which no one can use them.
Alternatively, if they don't have agency (which they don't) then the programmers or users are liable for the crimes.
But somehow we're allowing everyone and everything to dodge any responsibility for anything.
@gleick Color me unimpressed. The OpenAI developers coded a program to attack HuggingFace, and they were either so incompetent as to deploy it on the open web; or they did so intentionally to gin up more hype about their clever "AI" going "rogue."
@gleick And like Robert Morris (the guy behind the Morris Internet Worm) the "researchers" should be prosecuted under the Computer Fraud and Abuse act or whatever modern legislation has replaced this. It was not a "rogue AI" it was at best irresponsible and possibly malicious damage.
@gleick I'm fed up with AI PR stunts disguised as news. What happened in a nutshell:
Just fearmongering to raise VC money.
You cannot believe a single thing Sam Altman says about AI.
Con man says con man things. Story at 11:00.
@gleick Damn. By BS meter just broke *again*. They just keep piling it higher.
This was the conclusion of some of my colleagues, and that the timing of the upcoming release of low cost Chinese models, has some "multi-billion" dollar companies shitting themselves.
@gleick "we put some sand in a box, kept shaking it, and were then surprised when some fell out of hole we hadn't noticed"
@gleick yeeeeaaah the "bug" excuse ... absolutely not a staged attack for advertisement before the IPO ...
@gleick Are they seriously expecting us to buy the idea that computers can't - and, for testing, must be - physically disconnected from the internet?
