What you’re missing here regards diversity and volume.
Diversity: because language models are in some sense capturing an average (even if that average can be from a subset of training data), its comments will not be very diverse. I suspect if you try sampling away from the center of the distribution, you’ll find the quality degrades to nonsense.
Volume: the biggest problem is that you’ll end up with complete overwhelm of plausible comments. Language models cannot reason, which means they are incapable of producing the best insights (they can probably provide insight through “monkeys at typewriters” effects).
So the smart human who has analysed the subject really well, who currently has to rise above a certain amount of background noise, will now get completely swamped by it.
Previous poster said that GPT is to writing what calculators are to math. That's an interesting point, is this an analogy we can reason with?
My terse response was "no, because calculators are correct all the time whereas GPT is incorrect a lot of time". So, that reasoning can't be used. Need a different one.
This is the best argument I've seen against my own position. I don't think I agree, which leads me to think that the error I have made was in disagreeing with the art galleries. I guess they also didn't want to deal with a deluge of the banal.
For all but the shortest fragments of text, I suspect the answer to this will be "yes" and this will be achieved in a short timeframe. Interact with Chat-GPT enough and you soon begin to tire of its middle-of-the-road prose. Because it is in some sense a statistical average of its inputs, its writing feels very monotonous.
Of course, you can try to push it in the direction of something more interesting... "write in the style of X" so that it will regurgitate from a subsection of its training input, or push it onto a niche topic. Across a wide range of inputs, though, I predict it will be easily classified.
As training GPT will take far longer than training a GPT-detector, this is an arms race that the detectors will win for the most part. Anyone wishing to use GPT to generate middling essays to cheat at school will have to spend a lot of time to disguise the text (perhaps repeatedly running against open-source detectors) and then take the risk that detection does not improve between their submission and review.
Well, that is certainly false, because Chat-GPT was trained on the web and would surely have been exposed to content explaining policies on HN. So if this was generated by Chat-GPT (and it either was or was written by a human impersonating Chat-GPT's banal writing style) then it's another example of why it should be banned.
I feel like Pokémon go style games are massively underexploited.
Not just augmented reality but using cities as playgrounds. Nokia did some interesting things with this in the late 90s.
A fps might be problematic, but what about non violent trading (certain locations have mines etc, you can use resources to construct factories visible in augmented reality etc.) essentially a parallel world economy and reality that sits atop the physical word.
That’s a good point. Relatedly, Jaron Lanier says something quite insightful on the topic that gets to the root (pun) of the problem.
If you think all information should be free, like all software, then the creators of information are not being financially rewarded for creating it. That’s how you end up with cloud data centres making huge profits whilst most OSS projects struggle for cash. It’s also where the business models of Instagram, Facebook, Twitter, and TikTok arise.
Lanier makes a rather odd statement about how web links should be two way to solve this problem. I get his drift, but a better idea is that the web should be built on a more rigorous foundation (see Joe Armstrong’s talk - “the mess we’ve made” - for how it could work). But the general idea is correct - in an information economy either everyone is fairly rewarded for the valuable information they provide, or else you end up with giant companies capturing all the value instead, which leads to gross inequality.
Tbf to Stallman, he is more focused on constraining the worst commercial uses of software, but his fundamental idea that (as per this post) “knitting patterns should be free”, originally something that sounded like an exciting and inspiring idea, has turned out to be absolutely disastrous. He was maybe philosophically correct but in practice, the idea does not serve the best interests of most people.
Information shouldn’t be free (!!! I said that on HN) but rather information should be provided at a cost that is based on both (a) the cost of its provision and (b) the economic value that recipients of that information derive from it. So, for example, OSS authors writing code that underpinned Google’s empire or someone’s SaaS, should get a (small) slice of that pie.
I think once you recognise this, you end up in a really exciting place: like, how could this work in practice? What new concepts do we need to apply to information to do it? Some of the ideas of crypto should come in here: we need to be able to identify information uniquely, for example, so we can track its value. We need to know how many SaaS products are using Bootstrap or whatever. A new type of economy based on fair reward for information is possible. And for once, we do actually need a technical solution to a social problem.
I didn’t cover Yemen, because I am not as familiar with the details of the conflict, and from what I do know it seems to be a very complex situation involving many actors.
Much as I’ve defended the US in some other cases, I think that view on Iraq is somewhat naive. Yes, some wanted to get rid of an evil dictator, but it was motivated more by geopolitics than any sense of altruism.
Well, intervention in Yugoslavia wasn’t a war started by the US.
The Kuwait war was after an invasion by Iraq and at the invitation of the government there, and the US were restrained enough to halt at the Iraq border.
The Syrian war was not started by the US, and it has played a minor part there. In fact, it is Russia that has indiscriminately bombed Syrian cities and targeted civilians and hospitals there.
Afghanistan was in response to 9/11 and the refusal of the Taliban to end their protection of a terrorist organisation that had repeatedly attacked American targets internationally before their deliberate targeted murder of thousands of American civilians. This does not seem comparable to Ukraine.
The Iraq war stands out for me as the unjustified and imperialist war of modern times. It seems to me that the atmosphere post 9/11 was used as a pretence. But even Iraq does seem different: Hussein was a brutal dictator and, at least initially, Americans were welcomed as liberators. In Ukraine we are seeing widespread civilian resistance to the invasion. But Blair and Bush should answer for their war crimes.
By far the biggest difference here, I think, is that by being a European war and through the potential for further conflict and nuclear conflict, this war has far larger international ramifications and strategic risks attached to it. To see European cities bombed is to be reminded of what has happened in the past and that it can happen again. That is why it is receiving so much attention.
This is whataboutism and it is so tired. A good propaganda tactic though, and the Russians are using it as an attack vector. Westerners seem really vulnerable to this form of propaganda, possibly due to the success of previous information warfare led by Russia.
Bringing up some other terrible thing for comparison does not change in any way whatsoever the severity of what is happening in Ukraine. Whataboutism propaganda is designed to muddy the waters around that indisputable fact.
Diversity: because language models are in some sense capturing an average (even if that average can be from a subset of training data), its comments will not be very diverse. I suspect if you try sampling away from the center of the distribution, you’ll find the quality degrades to nonsense.
Volume: the biggest problem is that you’ll end up with complete overwhelm of plausible comments. Language models cannot reason, which means they are incapable of producing the best insights (they can probably provide insight through “monkeys at typewriters” effects).
So the smart human who has analysed the subject really well, who currently has to rise above a certain amount of background noise, will now get completely swamped by it.