Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

G/O media will make more AI-generated stories despite critics

vox.com

Post reply on HN
Print view View original

91–100 of 108 posts

Re: G/O media will make more AI-generated stories despite critics

#91

Post by dghlsakjg »

TillE wrote: post #64
munificent wrote: post #61

Earlier quoted context omitted.

> It's not really different from getting a content farm to write you a bunch of crap, just cheaper and faster. Using a gillnet to catch 1,000 fish in an hour is not really different from using a rod and reel to catch a few a day. It's just cheaper and faster. But the gillnetting can easily lead to extinction and death of an ecosystem whereas recreational fishing rarely does. Scale matters.

We already have plenty of non-AI autogenerated spam which is apparently good enough to please Google, LLMs just make that slightly better. So the scale already exists, and we're really just talking about quality.

SEO spam has already made a tremendous impact on the quality of google results (it used to be that a bad result was a good website about the wrong topic, now a bad result is a bad website about the right topic, and there are A LOT of them).

Being able to generate 100x the spam will certainly hurt things even more.

Re: G/O media will make more AI-generated stories despite critics

#92

Post by dghlsakjg »

2OEH8eoCRo0 wrote: post #80

I think that nuance is key. An AI-generated story with factual errors is obviously bad but I wouldn't mind if reporters were able to feed facts they've found, data dump style, to an AI that could generate a narrative that is then reviewed.

The kind of straight news story that this works for is the kind of story where writing it is the fastest part already.

Getting the actual facts to put in the story normally took far longer when I was working in a newsroom.

Longer, harder to write stories, like deeply researched news, or long features are not something that AI can do (yet).

Re: G/O media will make more AI-generated stories despite critics

#94

Post by Barrin92 »

circuit10 wrote: post #70
gwright wrote: post #29

Earlier quoted context omitted.

I think it confuses things immensely to anthropomorphize large language models. LLMs don't lie or tell the truth they just spit out text that is in alignment with the training model. Don't give them agency they don't have.

It’s a reasonable way to describe it. LLMs are designed to talk like humans so using human terms to describe them works quite well

it's also not a great term for the same behavior in humans. There's the often quoted distinction by Harry Frankfurt about lying, which is knowing the truth but intentionally subverting or hiding it, and bullshitting, which is talking without any regard to truth or falsehood whatsoever.

It's a distinction worth making as people are much better at spotting lies than they are at spotting bullshit. The bullshitter may even be accidentally correct. AI models in particular will make up things in random places where nobody is sceptical or has their guard up because they see no reason for a lie, but bullshitters don't need one.

https://en.wikipedia.org/wiki/On_Bullshit

Re: G/O media will make more AI-generated stories despite critics

#95

Post by tsol »

Karellen wrote: post #82
gwright wrote: post #51

Earlier quoted context omitted.

That is an interesting point. I'm not sure why, but "hallucinate" doesn't bother me as much as "lie". Maybe it is because of all the ancillary baggage that "lie" has related to our current culture wars (lies, misinformation, disinformation, gaslighting, etc.) I just asked (another anthropomorphic verb) ChatGPT for a better word and it came up with "fabricate". > The term "fabricate" could be used to describe AI's ten…

I always thought that "confabulate" would have been the best word to describe what LLMs do: > Confabulation is distinguished from lying as there is no intent to deceive and the person is unaware the information is false. Although individuals can present blatantly false information, confabulation can also seem to be coherent, internally consistent, and relatively normal. > Can include autobiographical and non-personal…

Confabulate is more of a neurological term so I think that's why it never caught on, but I've always thought that was the perfect word for what's happening

Re: G/O media will make more AI-generated stories despite critics

#96

Post by BoorishBears »

JohnFen wrote: post #71
BoorishBears wrote: post #68

Earlier quoted context omitted.

It doesn't. Hallucination in this context means 'an unfounded or mistaken impression or notion'. There's one side of the current zeitgeist that over-anthropomorphizes these models. But there's another side that seems to be terrified that LLMs could be anything resembling intelligent Both sides are pretty emotion over facts though. A lot of their "human-like" behavior is emergent from things that don't work the same w…

> It doesn't. Hallucination in this context means 'an unfounded or mistaken impression or notion'. Correct. Which, as near as I can tell, isn't the correct characterization of what's happening with LLMs. > But there's another side that seems to be terrified that LLMs could be anything resembling intelligent I certainly have no such fear. I say that only to indicate that my biases are not of that sort.

The first sentence seems contradicted by the second no? If you don't have that fear then why bother playing around the semantics of "impression or notion"?

Is it an unfounded or mistaken cluster of tokens which resemble an impression or notion then?

At the end of the day it's producing an output which systemically was intended to be accurate, but ended up not being accurate through a mistake in generation: That's a hallucination.

Re: G/O media will make more AI-generated stories despite critics

#98

Post by constantcrying »

circuit10 wrote: post #70
gwright wrote: post #29

Earlier quoted context omitted.

I think it confuses things immensely to anthropomorphize large language models. LLMs don't lie or tell the truth they just spit out text that is in alignment with the training model. Don't give them agency they don't have.

It’s a reasonable way to describe it. LLMs are designed to talk like humans so using human terms to describe them works quite well

Lying is intentional. A statement which is untrue is not by itself a lie, the thing which makes it a lie is the intention of the person.

AIs do not have intentions, the only thing in an AI is a statistical model of how likely words appear in certain contexts.

"Hallucinating" is a much better term, even though it still pretends a staristical model is akin to a human.

Re: G/O media will make more AI-generated stories despite critics

#99

Post by gwright »

JohnFen wrote: post #65
gwright wrote: post #51

Earlier quoted context omitted.

That is an interesting point. I'm not sure why, but "hallucinate" doesn't bother me as much as "lie". Maybe it is because of all the ancillary baggage that "lie" has related to our current culture wars (lies, misinformation, disinformation, gaslighting, etc.) I just asked (another anthropomorphic verb) ChatGPT for a better word and it came up with "fabricate". > The term "fabricate" could be used to describe AI's ten…

"Fabricate" has the same problem as "lie" and "hallucinate". All of those terms imply a cognition that isn't actually happening. I think the best way to refer to these things is the more accurate "error". The LLM isn't lying, it's in error.

I think "fabricate" is much less aligned with cognition.

It is completely reasonable to say that "This machine fabricates widgets." or "This algorithm fabricates pseudo-random song lyrics." without being worried that the machine or the algorithm is exercising "cognition".

Re: G/O media will make more AI-generated stories despite critics

#100

Post by gwright »

scarface_74 wrote: post #74
gwright wrote: post #29

Earlier quoted context omitted.

I think it confuses things immensely to anthropomorphize large language models. LLMs don't lie or tell the truth they just spit out text that is in alignment with the training model. Don't give them agency they don't have.

Yes and when you’re using “serverless” services there are actually servers running somewhere. At this point, it’s not insightful for someone to mention either on HN.

If we are going to expand the discussion to general inaccuracy in human communication we are going to need a much larger thread.

I do agree with you that "serverless" is a misnomer and not a useful technical term. I'm guessing the marketing team wasn't fond of "dynamically provisioned and auto-scaled pool of virtual machine instance-based services".

Post reply on HN