OpenAI CEO warns that GPT-4 could be misused for nefarious purposes
theregister.com
theregister.com
Personally I think that’s the only way to avoid the “everyone being homeless and jobless” scenario we’ve all been describing.
If Microsoft have cheap labor from bots, and you do too, it levels the playing field dramatically. You won’t have to pay money for access to similar capabilities
The Luddite’s were angry because they were replaced by machines they couldn’t afford to own themselves and profit from. With open source machines, this doesn’t happen. We all have machines and then we all have more equal opportunity.
If someone's job is eliminated because they were replaced by a bot, how does having their own bot help them? They're still out of work.
Or are you suggesting that everyone in the future will have to become AI engineers?
I'm not being argumentative here. I'm not understanding, and want to understand how this would be applied to help people in that position.
The only way I can think is if the solution is that they will all be AI operators, in which case the solution on offer is another variant of "retraining" -- but not everyone will be suitable for that sort of work. But that doesn't sound right, so I think I'm not understanding.
The means of production is both the software AND the hardware.
So we have to assume you’ll be able to run them on much more efficient hardware soon.
Anyway, I’m not necessarily talking about individuals running their own, but groups of people.
No, people, with all their diverse backgrounds, expertise and ingenuity will come together and build what's required for their survival and because of open source technology, they will also have access to the same advanced AI systems as M$.
I feel like I've had a similar conversation when I was talking to someone arguing that the 2nd amendment was imperative for citizens to defend themselves against the government. Like bro, how is your shotgun winning against an aircraft carrier.
For all other kinds of wars, yes, the government can crush you. That was true even in the 1700s, and has only gotten more true today.
Not only can scammers do this, but they will prefer to do this purely for cost reasons. Scammers have to turn a profit, too. And nation state actors will do it for traceability reasons.
So the really bad actors won't use OpenAI anyway. So OpenAI should stop lobotomizing their service for the non-nefarious actors who are just trying to make the next generation of software.
Sure, go ahead and put subtle steganographic fingerprinting in the OpenAI output, so long as it doesn't affect quality. This will hinder the low-tech bad actors who can't run LLaMa themselves, and won't hurt the good actors.
But it's the most infuriating thing to try to build on the GPT4 API when its "helpful assistant" chat type mindset occasionally breaks the workflow.
Looking at what happened with stable diffusion, there's no way this can be kept under wraps. Weights WILL leak and trigger a cambrian explosion of open source innovation.
Their moat, if any, will be in the boring things: regulatory capture, integration, whatever you call Oracle's business model of "pay through the nose and blame us if something goes wrong". And at that point it'll be just any boring regular tech company.
They offer serverless GPU deployments, pretty hassle free (compared to using AWS or azure)
A few days ago before it was taken offline I did some tests I did with Alpaca-stanford. If I can run a GPT-3 level LLM in my machine, I won't even bother looking at GPT-4. Yes GPT-4 is impressive, but GPT-3 and Alpaca are already good enough IMO.
I'm becoming increasingly wary and anxious, especially when we already seem to live in a pretty fragile world. People already are distrusting of everything, even from our core institutions...add in AI's ability to take millions of jobs and create fiction from real events in a way that looks like reality, and we could be walking into a catastrophe. Having a bunch of newly unemployed people that don't know what to trust seems bad.
This is going to be so abused heavily, I think people are absolutely naive if they think this will be a smooth ride into the future. AI is going to cause massive problems in the next decade, and tech companies that create it will not suffer any consequences for it, they'll actually grow to be much more powerful. I don't know the answer, but worrying about it and acknowledging this is a likely outcome isn't a crazy idea. As I said, a bunch of upset and distrustful people that can no longer make ends meet because AI took their job is a great way to cause huge conflicts. We've lived in a relatively peaceful time, we take that for granted, it isn't always like that.
Vastly fewer people believe in outright lies now than a couple of decades ago. Fringe theories of all kinds used to be commonly accepted, much more than they are now. Religiously motivated homophobia is a good example. And I’d claim that making it easier to spread falsehoods made it so, by training people to be less naive.
That is very new, at least in how far spread it is. There are people that think there is no war in Ukraine at all and this is all a "Wag the Dog" situation. If you aren't hearing these things, go to a local bar and bring it up. I guarantee you will find people you'd not suspect believing something absolutely devoid from reality, even in "blue" areas. You will be shocked at the volume of people that agree or chime in with something even more crazy. It is something I've noticed which started around 2020, and it's very alarming.
With AI this will become even more confusing and widespread, as it's easier and easier to fake things or come up with "facts" written in long form that aren't based in reality at all. Gen Z might become immune to this, but every other generation before it? I don't think so. Now imagine millions of people that believe things like this that now have no job, or if they do it's lower paying. It has the potential to be very bad. This might not happen in 2023 or even 2025, but at the speed AI is currently moving, it can't be more than a decade away from massive displacement and realism in generation that will threaten many threads of society.
AI is a general-purpose accelerant and force multiplier. It provides a mechanism for automating and deploying at scale, a set of attacks on our society, against which we have little experience and almost no defense—nor even any good means of detecting, at least not until post-mortem forensics.
The most obvious harmful avenue for this is venal criminality (which will be awful) but the real danger is in the political sphere.
There is already widespread use of AI for disinformation purposes in e.g. the Ukraine war.
I have been saying for the last N months or so, my immediate concern with AI is not AGI but augmented intelligence applications which are leveraged enough to be destabilizing.
In specific, I believe the 2024 election cycle in the US will be decided by AI.
We aren't ready for this.
Hell no. You are probably thinking about things like 'misinformation'. Let me tell you that all such concerns are totally unfounded:
People just buy into whatever already fits their existing bias. Even if they are lies or proven lies. They don't care if something was a lie. If there is more stuff that confirms their existing bias, they will shout louder. If there is less, they will shout less. But they will still vote the same way.
So when the 2024 elections happen and there is a lot of misinformation, everyone will just buy into ! whatever ! confirms their existing bias - be it truth or be it a lie - and vote in the exact same manner they were going to do before that misinformation.
While this reduces the concerns about misinformation and/or the effects of the AI, it also suggests that objective politics is difficult because people are not affected by facts, truth, or even their own prior experiences.
Electric cars have no direct emissions but increase mining operations in certain parts of the world. Or they can be used to plow through a public gathering of people. Or with some rewiring to electrocute someone to death. If you think hard enough, you can find nefarious purposes for almost any household item that's made your life easier.
I think what's important to focus on are the "net" benefits but the outliers feed into our emotional response.
As in children digging in pitch black mine shafts with their hands. Sometimes referred to as "Artisanal Mines"
The reason outliers feed our emotional response is a survival/skin in the game mechanism. Parroting Taleb, all it takes in ruin once and the game stops. It’s not unreasonable to be hyper focused on reducing long tail risks with potentially catastrophic and unknown results. Caution and fear is warranted here.
Tech is amoral, neither moral nor immoral.
Great power / responsibility and all that.
I think they probably started out worried about safety and misinformation and nefarious actors, but I think this comment reveals their present intentions: "A thing that I do worry about is … we're not going to be the only creator of this technology. There will be other people who don't put some of the safety limits that we put on it."
Translation: "the competition will be able to duplicate our efforts, and we may not have any way to keep ahead of the pack. Please, pretty please, worry about safety so that we can sell you our safety tools."
How or what do we train on? How can it improve itself.
Drunk driving was legal for years, I think it's a good thing to be wary of how things can change and to be ready, as a society, to deal with unintended consequences of this new technology, even if we assume it to be a worthwhile net good.
As an amusing tidbit: a reporter asking people about making drunk driving illegal, a little before that came to be a reality: https://www.youtube.com/watch?v=W_tqQYmgMQg
Truly, the real heroes are OpenAI!
> Some people, however, are confused by the startup's behavior. If the technology is as dangerous as OpenAI claims, why is it readily available to anyone willing to pay for it? Still, Altman added: "A thing that I do worry about is … we're not going to be the only creator of this technology. There will be other people who don't put some of the safety limits that we put on it."
Translation: We are happy to ensure that it only produces right think, right thought, and right information. We will not allow double minus bad things to come from our LLM.
Bastards, I suspect the issue was that the prompt contained the word "monkey" and I suspect since this is an USA company they put tons of safe filters in, from all the big groups(conservatives, religious, liberals ) so nobody gets offended and cause a PR issue.
TL:DR they are filtering what we can do, and when they do mistakes we still pay for it.
I guess we won't be having discussions with the AI about "Expeditionary Force" book series.
The stupid shit is that there is no flag for us to know is moderated , and the chatbot response with a similar message but in different wording, so you need to guess that the response was moderated.
Equally informative, equally valid, equally pointless.