mastodon 4.7.3By day I am a computer programmer with Index Data, where I have been happily and gainfully employed for 20 years.
By night, I am a vertebrate palaeontologist with the University of Bristol, specialising in sauropods: the biggest and best of all dinosaurs.
I am an advocate for open access, open data and open source, and I'm beginning to think pre-publication peer-review might be a mistake. Supported #LFC before Bezos takeover.
Email: mike@miketaylor.org.uk
ORCiD: 0000-0002-1003-5675
My commentary:
This was not a case of "they told it to hack, it hacked, gosh, what a surprise". The test involved hacking, but instead of hacking the target machine the model instead pulled off a much more difficult series of hacks to get the answer sheet from a machine OpenAI did not own, because that's the most reliable way to score 100%. If a trainee human pentester did that they'd be looking at a prison sentence (HuggingFace had already involved law enforcement).
It's exactly the kind of alignment failure that the paperclip scenario describes: the AI was given a low-stakes task, and it did things that a human would never consider (and most would never be capable of) in order to slightly increase its probability of success. Human attempts to restrain it completely failed. We got lucky this time because it didn't even try to be sneaky.
No, this was not a fucking marketing stunt, Jesus fucking Christ. OpenAI were forced into this disclosure because HuggingFace had already called the cops. This tells us that OpenAI's models will randomly commit criminal acts in order to do slightly better at tasks you don't really care about, exposing their customers to considerable legal risk (IANAL). They are trying to downplay it.
Edit: HuggingFace's announcement said that they tried to use hosted American models to analyse the attack, but they were locked out at their hour of greatest need and had to switch to a self-hosted Chinese model. That's the opposite of an advert for OpenAI's defensive cybersecurity capabilities.
This is legitimate nightmare fuel. We need to pause AI development right the fuck now. Contact your elected representatives.
"When a [cookware] brand changes hands this many times, the name becomes the product and the pan becomes a cost to be managed beneath it: thinner steel here, an offshored handle there, a warranty rewritten by lawyers instead of engineers, and so on…"
https://www.worseonpurpose.com/p/your-cookware-got-worse-on-purpose
Yet another game where Liverpool did more than enough to win, but somehow we managed to score only once from nearly three xG. It feels like this has been happening a lot this season.
"It's time for online checkin" Jet2 tells me. I follow the link, enter my booking reference, name and flight date, and it takes me to a Manage My Booking page with a Check In Now button.
Clicking the button takes me to a 404.
??!
Brilliant break by Kerkez and Frimpong, utterly squandered by Wirtz.
... And then Wirtz is SO close to notching a winner on the break, but denied by Kweev.
VVD can't capitalise on an unexpected chance to shoot from the edge of the area. Way over the bar. Shame.
Anyone else remember when John Hartson was sent off for two yellow cards, both given for deliberate handballs?
The most deliberate of deliberate handballs — both hands, even. No yellow card, of course, because Darren England.
With Rio's stratospheric emergence, Trey Nyoni has gone under the radar. Has anyone been paying attention and knows has a sense of how good he is?
We have definitely conceded the initiative in the late stages of this game. I hate how we've been letting that happen.
