6 ms·

GPT-3 is just a tool. If a product that uses it messes up you can hardly blame it on the tool but rather on that product or the company developing it (here Nab

GPT-3 is just a tool. If a product that uses it messes up you can hardly blame it on the tool but rather on that product or the company developing it (here Nabla).

s3tz 5y ago
Not exactly, it's a service with unpredictable output. This isn't like a knife where you know what will happen depending on how you use it.
- egeozcan 5y ago
  Chatting with random people on the internet is much more unpredictable. Try writing on 4chan that you are depressed and gasp in horror how horrible humans can be when anonymous. Okay maybe it is indeed very predictable what would happen on 4chan, but I hope you get my point. Here the AI is just a tool, we shouldn't categorize them as some "special" software.
  - bsjks 5y ago
    If you ask in certain boards and make an effort to write your post detailing what’s going on, you will only receive helpful responses.
    - egeozcan 5y ago
      Those boards have better moderation, something you need regardless of AI, which is my point. AI doesn't add more danger to the situation, you just need to moderate as usual. If you are using some computer program to add more content to your web site, you will just have more content to moderate. News aggregators do have this problem for example, as sometimes automated crawlers post sensitive content and that needs to be marked & deleted before being published. Saying, "oh but our algorithm can cause depression, so we moderate the access to it" doesn't make sense, as any content can cause it and needs to be moderated. That's why content moderation is one of the hardest problems of this age. You use computers to do it? Computers develop biases. Humans? Same! It's very hard to scale and is a problem only remotely related to AI-generated content because AI generating content increases the input to your moderation system. Now I hope it is clear what I'm trying to say :)
      - speedcoder 5y ago
        Does "moderation" require bias? For that matter, can knowledge exist without bias?
        egeozcan 5y ago
        > Does "moderation" require bias? Objectively yes. Information on correctly peeling a pineapple is ok? Processing chicken to cook it? What about a dog? A rat? Fish? Insects? Is killing millions of bacteria with a single drop of chemical okay? Ways to commit suicide? What if this video is for prevention? When it comes to moderation, we can't even set a consistent rule set to "deal with it all (c)". We just wing it, and hope it matches the expectations of our majority of our users... and the government(s)! Yeah governments have rules, but they tend to change towards the expectations, and a lot of rules for interpretation. It's a very hard task. > For that matter, can knowledge exist without bias? Umm... I don't know what that would mean. Isn't science there to reduce the amount of bias in knowledge? So, maybe, no? We can hope to reduce it, but we cannot get rid of it? At this point I don't really know what I'm talking about to be honest :)
  - _s_a_m_ 5y ago
    > Chatting with random people on the internet is much more unpredictable Not particularly the best comparison, we are talking here about applications and services which we sell and we have to make sure that we have the confidence that we can trust our own systems and the customer can rely on our word and engineering skills. Right now we are still in the primordial soup of AI. As long as we do not have proper methods to verify and certify our models, they are dangerous and unpredictable.
    - egeozcan 5y ago
      Please see my other response: https://news.ycombinator.com/item?id=29576981 https://news.ycombinator.com/item?id=29576981 As a summary, generating content isn't new and just increases your moderation load. AI is not something "special". It's just software, at the end.
      - ceejayoz 5y ago
        That’s a bit like saying machine guns aren’t an improvement over bows and arrows because they both throw projectiles, isn’t it?
        egeozcan 5y ago
        I'm not talking about improvements. You need gun control with bows and arrows, and you still need gun control with machine guns. That's what I'm saying.
        ceejayoz 5y ago
        And what we're saying is stuff like GPT-3 may be improvements on the level of archery --> machine gunning when it comes to online trolling, misinformation, and the like. Casually dismissing it with "well it's just extra moderation load" is a mistake, I think.
        egeozcan 5y ago
_s_a_m_ 5y ago
You can frame anything being "a tool". But if you want to stay within your logic, then this is a very specific "tool", it is supposed to directly interact with people, not machine to machine, and as such it can have consequences which are unpredictable. And no, human to human interaction is mostly NOT unpredictable, we act according to social norms and have common sense, otherwise you're not a proper human. This is not a new issue but is heavily researched in the certifiable AI area. If you just turning million knobs randomly based on an architecture and the biased data you have, of course you don't really known what you are doing.
KarlKemp 5y ago
It really doesn’t matter what it ‘is’ or if it makes sense to speak of things in moral categories. There are plenty of ‘tools’ that are/should be restricted, from guns to Tucker Carlson. It can obviously have some effect on the so-called ‘real world’, or people wouldn’t pour money into developing & using it. It would be a unique feat to be powerful with no chance of any of the effects being harmful. From there, it’s just arithmetics: what harms and benefits do we expect with what probabilities, and how does this change depending on the distribution scheme?

GPT-3 is just a tool. If a product that uses it messes up you can hardly blame it on the tool but rather on that product or the company developing it (here Nabla).

s3tz 5y ago

Not exactly, it's a service with unpredictable output. This isn't like a knife where you know what will happen depending on how you use it.

egeozcan 5y ago
Chatting with random people on the internet is much more unpredictable. Try writing on 4chan that you are depressed and gasp in horror how horrible humans can be when anonymous. Okay maybe it is indeed very predictable what would happen on 4chan, but I hope you get my point. Here the AI is just a tool, we shouldn't categorize them as some "special" software.
- bsjks 5y ago
  If you ask in certain boards and make an effort to write your post detailing what’s going on, you will only receive helpful responses.
  - egeozcan 5y ago
    Those boards have better moderation, something you need regardless of AI, which is my point. AI doesn't add more danger to the situation, you just need to moderate as usual. If you are using some computer program to add more content to your web site, you will just have more content to moderate. News aggregators do have this problem for example, as sometimes automated crawlers post sensitive content and that needs to be marked & deleted before being published. Saying, "oh but our algorithm can cause depression, so we moderate the access to it" doesn't make sense, as any content can cause it and needs to be moderated. That's why content moderation is one of the hardest problems of this age. You use computers to do it? Computers develop biases. Humans? Same! It's very hard to scale and is a problem only remotely related to AI-generated content because AI generating content increases the input to your moderation system. Now I hope it is clear what I'm trying to say :)
    - speedcoder 5y ago
      Does "moderation" require bias? For that matter, can knowledge exist without bias?
      - egeozcan 5y ago
        > Does "moderation" require bias? Objectively yes. Information on correctly peeling a pineapple is ok? Processing chicken to cook it? What about a dog? A rat? Fish? Insects? Is killing millions of bacteria with a single drop of chemical okay? Ways to commit suicide? What if this video is for prevention? When it comes to moderation, we can't even set a consistent rule set to "deal with it all (c)". We just wing it, and hope it matches the expectations of our majority of our users... and the government(s)! Yeah governments have rules, but they tend to change towards the expectations, and a lot of rules for interpretation. It's a very hard task. > For that matter, can knowledge exist without bias? Umm... I don't know what that would mean. Isn't science there to reduce the amount of bias in knowledge? So, maybe, no? We can hope to reduce it, but we cannot get rid of it? At this point I don't really know what I'm talking about to be honest :)
- _s_a_m_ 5y ago
  > Chatting with random people on the internet is much more unpredictable Not particularly the best comparison, we are talking here about applications and services which we sell and we have to make sure that we have the confidence that we can trust our own systems and the customer can rely on our word and engineering skills. Right now we are still in the primordial soup of AI. As long as we do not have proper methods to verify and certify our models, they are dangerous and unpredictable.
  - egeozcan 5y ago
    Please see my other response: https://news.ycombinator.com/item?id=29576981 https://news.ycombinator.com/item?id=29576981 As a summary, generating content isn't new and just increases your moderation load. AI is not something "special". It's just software, at the end.
    - ceejayoz 5y ago
      That’s a bit like saying machine guns aren’t an improvement over bows and arrows because they both throw projectiles, isn’t it?
      - egeozcan 5y ago
        I'm not talking about improvements. You need gun control with bows and arrows, and you still need gun control with machine guns. That's what I'm saying.
        ceejayoz 5y ago
        And what we're saying is stuff like GPT-3 may be improvements on the level of archery --> machine gunning when it comes to online trolling, misinformation, and the like. Casually dismissing it with "well it's just extra moderation load" is a mistake, I think.
        egeozcan 5y ago

_s_a_m_ 5y ago

You can frame anything being "a tool". But if you want to stay within your logic, then this is a very specific "tool", it is supposed to directly interact with people, not machine to machine, and as such it can have consequences which are unpredictable. And no, human to human interaction is mostly NOT unpredictable, we act according to social norms and have common sense, otherwise you're not a proper human. This is not a new issue but is heavily researched in the certifiable AI area. If you just turning million knobs randomly based on an architecture and the biased data you have, of course you don't really known what you are doing.

KarlKemp 5y ago

It really doesn’t matter what it ‘is’ or if it makes sense to speak of things in moral categories. There are plenty of ‘tools’ that are/should be restricted, from guns to Tucker Carlson. It can obviously have some effect on the so-called ‘real world’, or people wouldn’t pour money into developing & using it. It would be a unique feat to be powerful with no chance of any of the effects being harmful. From there, it’s just arithmetics: what harms and benefits do we expect with what probabilities, and how does this change depending on the distribution scheme?