It’s interesting how the ruling class gets to develop these technologies, know there are downsides, but shovel the problems their products create onto governments and the people.
They are not looking to society to solve those problems though, they are fearmongering legislators into restricting the technology using laws they will (in parallel) shape, thus creating a legislative moat to strangle competitors with.
This needs to be repeated over and over. OpenAI is not actually worried about existential risks of AI. If they were, they would be developing their AI very differently.
OpenAI has realized that talking about existential risks in AI is a good way to:
A) distract from immediate problems that OpenAI is not willing to take seriously (spam, prompt injection/hijacking, reliability issues, bias, costs and sustainability, etc...)
B) make their products sound more advanced than they actually are (not that GPT isn't impressive, but it's a far cry from anything resembling AGI). OpenAI benefits tremendously from creating the impression that it's always about a month away from a serious breakthrough on AGI, that's how you get VC investment money.
C) to encourage regulators to construct moats around the industry that will hold back open models that don't have the same privacy risks and that could be more specialized and could spur more innovation towards decreasing computing and environmental costs and moving away from the "shovel everything into a thousand GPUs that you can only afford with an investment from Microsoft" business model that OpenAI currently enjoys.
And I'm sure on a personal level Sam Altman enjoys talking about this and enjoys thinking about these high-minded concepts and worrying publicly about how dangerous the thing he runs is. But it's also strategic; Sam Altman is not really fundamentally scared of any of this. He's just figured out how to piggyback on top of a small community of people who are sincerely scared of AGI (and also a larger population who has only ever thought of AI risk in terms of sci-fi stories and who is highly susceptible to buzzwords about AI risk) in order to serve his own ends.
And of course, to re-characterize the extremely well-studied and well-documented fields of AI misuse as if it's some brand new concept called "societal misalignment" is a very Sam Altman thing to do -- but it's an effective PR strategy for OpenAI to act like they're taking practical concerns seriously when in actuality they're ignoring most of the people raising those concerns and hand-waving the issues that their product has right now. And it's an effective way of absolving OpenAI of responsibility. After all, it's not OpenAI's fault that its product is unreliable and can be hijacked by a malicious PDF and is extremely useful as a low-quality spam generator. It's just that society itself is subtly "misaligned" (ignoring the fact that "alignment" is a word in AI that has meanings and is used in a specific way that isn't really all that applicable to society).
Typically under articles like this HN debates existential risks. And fine, whatever, but I don't want anyone to lose sight of the fact that OpenAI is not authentically participating in that discussion. Their goals are not your goals.
The government foists downsides on us all the time, it’s nothing new. The FAA diversifies itself at the expense of your safety and we’re forced to use them if we want to fly. At least we have a choice if we use OpenAi’s products
You have no choice. OpenAI impacts all of society. From AI-generated articles you may read, to AI-generated imagery, etc.
OpenAI is saying "we build it, you fix the problems we create".
They're no different from coal mining operations that product significant environmental impacts then go out of business leaving it to the local community/government to clean up.
7 comments
[ 22.9 ms ] story [ 844 ms ] threadOpenAI has realized that talking about existential risks in AI is a good way to:
A) distract from immediate problems that OpenAI is not willing to take seriously (spam, prompt injection/hijacking, reliability issues, bias, costs and sustainability, etc...)
B) make their products sound more advanced than they actually are (not that GPT isn't impressive, but it's a far cry from anything resembling AGI). OpenAI benefits tremendously from creating the impression that it's always about a month away from a serious breakthrough on AGI, that's how you get VC investment money.
C) to encourage regulators to construct moats around the industry that will hold back open models that don't have the same privacy risks and that could be more specialized and could spur more innovation towards decreasing computing and environmental costs and moving away from the "shovel everything into a thousand GPUs that you can only afford with an investment from Microsoft" business model that OpenAI currently enjoys.
And I'm sure on a personal level Sam Altman enjoys talking about this and enjoys thinking about these high-minded concepts and worrying publicly about how dangerous the thing he runs is. But it's also strategic; Sam Altman is not really fundamentally scared of any of this. He's just figured out how to piggyback on top of a small community of people who are sincerely scared of AGI (and also a larger population who has only ever thought of AI risk in terms of sci-fi stories and who is highly susceptible to buzzwords about AI risk) in order to serve his own ends.
And of course, to re-characterize the extremely well-studied and well-documented fields of AI misuse as if it's some brand new concept called "societal misalignment" is a very Sam Altman thing to do -- but it's an effective PR strategy for OpenAI to act like they're taking practical concerns seriously when in actuality they're ignoring most of the people raising those concerns and hand-waving the issues that their product has right now. And it's an effective way of absolving OpenAI of responsibility. After all, it's not OpenAI's fault that its product is unreliable and can be hijacked by a malicious PDF and is extremely useful as a low-quality spam generator. It's just that society itself is subtly "misaligned" (ignoring the fact that "alignment" is a word in AI that has meanings and is used in a specific way that isn't really all that applicable to society).
Typically under articles like this HN debates existential risks. And fine, whatever, but I don't want anyone to lose sight of the fact that OpenAI is not authentically participating in that discussion. Their goals are not your goals.
OpenAI is saying "we build it, you fix the problems we create".
They're no different from coal mining operations that product significant environmental impacts then go out of business leaving it to the local community/government to clean up.