zlacker

Exactly, I also struggle to take seriously the "security" concerns of an organisation that releases this product with plugins and no proper way to restrict them and what they can do. The prompt injections are just ridiculous and show a complete lack of thought going into the design [0].

"move fast and break things"

It very much feels like they are trying to build a legislative moat, blocking out competitors and even open source projects. Ridiculous.

I don't fear what this technology does to us, I fear what we do to each other because of it. This is just the start.

0: https://twitter.com/wunderwuzzi23/status/1659411665853779971

> Let ChatGPT visit a website and have your email stolen.

> Plugins, Prompt Injection and Cross Plug-in Request Forgery.

>>samwil+(OP)
You can host your open source AI project in Belarus or Russia or whatever. You can even VPN into the country and pretend you're actually developing it there. And we also have Tor, I2P, etc. to hide where you're coming from. And we will likely have anti-stylometric tools so it will be very difficult to identify the author of the code.

So it's unlikely the government will be able to put a stop to it. Especially given AI is a technology that many will find very useful. They can't even put a stop to child sexual abuse material on the Internet, material which is universally hated. How are they ever going to stop AI development then? It's all going to go underground, on the darknet.

And restricting the development of open source software likely will be found unconstitutional, on First Amendment grounds, in the US. And it's likely to spur civil disobedience as well.

>>182716+h1
Last time the US was afraid of Germany developing nukes this resulted in nukes actually being developed and then promptly stolen by the Soviets.

Admittedly, at the time, socialist/communist were to some degree supported by the U.S. intellectuals - were’ve been quite a few that were sympathizing. So recruitment was easy. Hopefully this is not the case now. As it is quite clear that there is no good reason to support Putin, unless you happen to be his buddy, crony or an asset (i.e. Trump).

>>startu+44
What in the world are you talking about

>>182716+h1
> They can't even put a stop to child sexual abuse material on the Internet, which almost nobody wants. How are they ever going to stop AI development then?

They can't stop CSAM because of the tradeoff against privacy, not for technical reasons. AI development requires expensive, specialized hardware. Just not really comparable things.

>>idopms+T6
> AI development requires expensive, specialized hardware.

Does it though? I can run LLaMA on my desktop. Training requires hardware which is more expensive if you want it to run quickly, but it's in the range of tens of thousands of dollars. That's not beyond the means of many individuals, much less organizations. And in a few years it will be hundreds of dollars.

>>idopms+T6
It might not require specialized hardware in the future as GPUs get more powerful and we have other techniques such as LoRA for fine-tuning the models. We might see a distributed training [1] effort harnessing thousands of gamer GPUs worldwide, as well. All of this powered by open source software. Also there could be advances in the training software making it vastly more efficient.

1. https://arxiv.org/pdf/2301.11913.pdf