3 ms·
> Every new technology is developed and deployed the same way: > You make a prototype, try it at a small scale, make limited deployment, fix the problems, make
by persnickety 3y ago
> Every new technology is developed and deployed the same way:
> You make a prototype, try it at a small scale, make limited deployment, fix the problems, make it safer, and then deploy it more widely.
This makes an assumption: that the problematic technology is an intentional development rather than an emergent feature of an intentionally developed technology.
We already have a name for such emergent features: "bugs". No one really deployed Heartbleed, especially not in a limited deployment. Spectre? Rowhammer? And we all had/have to deal with the fallout, and we're not even done.
Who says that the danger of the technology cannot stay hidden until it's universally deployed?
- reducesuffering 3y agoYep. And there are already 20 year old viruses out there, autonomously following their programming, causing issues aligned with no human, as the original humans do not profit off it anymore, and unable to “be turned off”…
- afarviral 3y agoReally? What are some examples?
- PoignardAzur 3y ago> > You make a prototype, try it at a small scale, make limited deployment, fix the problems, make it safer, and then deploy it more widely. That argument in particular seems like wishful thinking. What in the last 10 years of our tech industry makes him think "move slowly and deploy carefully" is going to be the strategy they adopt now? And that's without considering the arguments in "We have no moat and neither does OpenAI". No matter how careful the main stakeholders are, once the technology exists, open-source versions will be deployed everywhere soon after with virtually no concern for safety.
- jprete 3y agoOpen-source models are deployed all over the place _now_ with no concern for safety. Fortunately nobody has yet managed to make a mutatable reproducing system out of them.