Makes you wonder how much power a random internet comment about a niche/new topic has, at least temporarily. How much can you affect the results of Bard/BingSydney/etc by just writing a comment?
Someone edits a factoid into a Wikipedia entry or posts it on a forum somewhere. People see it, believe it, and spread it. Then maybe the mainstream media cites it.
Well at that point it’s officially “real”. And if you try to argue against it people will use those “real” sources to “prove” the made up factoid.
I remember reading about someone, an actor or an author. Wikipedia had their birthday wrong. But writers had used Wikipedia to find the date and included it in articles. Wikipedia now uses those articles as proof that the incorrect birthday is correct.
An ouroboros of proof.
They did get copied, and they didn't always get caught, and poof, the fake road is now real since it exists in multiple "independent" sources.
Planning officer: you want to build 20 houses on the outskirts of [town Google made up]. Looks legit to me.
Planning officer: you want to build 20 new houses in the middle of an agricultural area. Let me think on that.
That kind of thing happened to the former German minister of defence Karl-Theodor zu Guttenberg, whose full name is
Karl-Theodor Maria Nikolaus Johann Jacob Philipp Franz Joseph Sylvester Buhl-Freiherr von und zu Guttenberg[0]
and not
Karl-Theodor Maria Nikolaus Johann Jacob Philipp Wilhelm Franz Joseph Sylvester Buhl-Freiherr von und zu Guttenberg
which the Wikipedia article claimed and the news magazine Der Spiegel cited in an article. This article then in turn became the source for the slipped-in surname Wilhelm on his Wikipedia entry.
While his real full name already sounds totally made up Der Spiegel actually published a correction regarding his name.[1]
[0] https://en.wikipedia.org/wiki/Karl-Theodor_zu_Guttenberg
[1] https://www.spiegel.de/politik/deutschland/in-eigener-sache-...
They’d invent a “fact”, spin up a webpage with that fact, add the fact to Wikipedia and cite the webpage to add plausibility.
Then one of them would occasionally Google to see if any “reputable” sources were repeating that “fact” (after reading it on Wikipedia) and add/swap their references in to strengthen the Wikipedia text.
As long as the “facts” were trivial enough, it worked.
Furthermore the gas lighting is real and my biggest frustration. Asking about a feature in a program, ChatGPT assuring me it is there, me following the instruction, not finding said feature in said menu and ChatGPT assuring me "Ohh, you must be using a newer version, where the feature was removed". Only it was never there. Happened like 3 times already.
To be fair, that's a decent approximation of the response humans give when confronted with evidence their memory doesn't match reality.
Computers were better in the old days, when they were simple and logical and not human-like at all.
It’s not at all bad, in general many people work together in a global complex society, but, oh boy, lots of evolution to be had if we want a better, more kinder world.
My current pet peeve is the emotional reactivity it elicits of any social media in its worst forms.
As great and optimistic as ST:TNG was, don't forget that in the later seasons, they basically resorted to "which crew member goes crazy in this episode?", I guess because the writers ran out of ideas.
I don’t understand why some people seem eager to use it as a search engine or replacement for a Stack Overflow or Quora type site.
20% of the time I have to google, but that’s made more efficient by using specific terms mentioned by chatGPT - getting the right terms in a google search is half the battle. It does seem archaic already for google to give me a list of links and I have to go through them all manually deciding if they answer the query.
So you either know enough to spot the error or you verify all of it.
Another example. I had a long session with it where I was trying to get it to write a gitlab ci config file where the artifacts to be deployed to Netlify were already present at the root of the repo. As an addition it had a requirement to use 'sed' to rewrite a domain name recursively. This is the first and only time chatgpt was useful to me although it made a lot of mistakes that needed correcting.It was more useful than a google search because I wasn't able to find the correct format with it as most examples assumed a build step.
One thing that was frustrating is how I would give it requirements and it would incorporate them only to lose a prior requirement when fixing a latter one. I ended up hand fixing one of the later configs rather than using the final one it gave me.
For example I needed to use a node image because netlify-cli was required (the only custom thing necessary from the image) but it would revert to using an Alpine image and point to a now archived repo with a nonexistent release url for gitlab's cli.
When it fixed something else it would break the find .. sed combination and add nonsensical things like *. html to it rather than working on all files recursively.
In the end, through a combination of AI helping and the user (me) chiming in and trying to reason about it I was able to get somewhere.
Not ideal but not useless.
In other words, follow-up questions and cross-referencing are a must. You can't take any of ChatGPT answers as true without verification. You also need to be really careful when asking questions, because if you give it too many assumptions, it's absolutely great at confirming your biases for you.
I find it to be a great "rubber duck" that can help you see and understand the relationships between facts that you might have missed otherwise. OTOH, it'll be many, many years before I'm comfortable asking it about things I'm not already mostly familiar with or can verify easily.
For me personally, no particular reason. I just want to see what all the hype is about. And using it as a general "suggest me solutions" type of thing, so you know the specific keywords to research further is certainly useful I would say. It just often breaks down if you ask for something specific.
If I ask it to write code, it results in the same error by using features that don't exist.
ChatGPT gives me a useful answer for my questions about these API changes every time in seconds. It is only right about 80% of the time, and I'm upgrading to 1.19 which is after the knowledge cut off, but it is insanely helpful.
See "Let them eat cake", "Napolean is short", "inflation is transitory".
And then these people get mad when we don't find them as always trustworthy. Some things aren't even small omissions, others vastly misrepresent ideas so as to be wholly false.
cow