People used to say that you were scammed if you used Wikipedia to check something. I use ChatGPT: it provides useful deep suggestions in my research. It cites sources when I ask it. Like a knowledgeable but fallible PA. I am a sceptic: I say "check that" sometimes "sorry I made a mistake". - ThreadSky

richardclegg.bsky.social • 67 days ago

People used to say that you were scammed if you used Wikipedia to check something. I use ChatGPT: it provides useful deep suggestions in my research. It cites sources when I ask it. Like a knowledgeable but fallible PA. I am a sceptic: I say "check that" sometimes "sorry I made a mistake".

Comments

misterabk.bsky.social•67 days ago

OK, yes, I think that's an extremely silly way to do research. Sorry! If going up a mountain to pray for inspiration helps you finish a paper, then that's fine with me. But I don't believe it actually, in a meaningful sense, works.

shermarama.bsky.social•66 days ago

I think there's a mismatch in what the word 'research' means here. I suspect ABK is thinking 'find the end result' while RGC is thinking 'find different sources of input'. In RGC's meaning, you can't find the end result online because *nobody* knows it yet...

misterabk.bsky.social•66 days ago

I don't *think* that's what I thought Richard meant. I understand that he's not just asking for answers to science questions. But, I take his point that I'm not a scientist doing research, I'm just a guy standing in front of generative ai asking it not to be shite.

shermarama.bsky.social•66 days ago

But also, sure, I didn't think it meant that either, before starting doing research, but didn't know there would still be a big perspective shift on sources and what anyone 'knows'. I think researchers (inc. AI ones) can forget others maybe haven't had this, and so don't auto-apply it.

shermarama.bsky.social•66 days ago

Yeah, me too... I've mostly worked at the interface between academic research and putting it into practice, and for every Richard, I see loads of degree-but-not-PhD qualified people doing Dumb Shit with GenAI, and not knowing they are. It's horrible to work with.

richardclegg.bsky.social•66 days ago

I am thinking of publishable science but yes. Questions where the definitive answer is not known as yet but could be established with good methodology.

richardclegg.bsky.social•67 days ago

I say "I am interested in improving ways to embed graphs into multidimensional space using a time dimension give me ten relevant papers and some suggestions" - it finds ten papers that are real and relevant and it makes suggestions that are helpful.If it did not do that I would not use it for that.

misterabk.bsky.social•67 days ago

I don't think it's a good idea, but I concede that experts can make judicious use of it, like a fuzzy search engine. But the vast majority of people aren't experts. I couldn't tell if those papers were relevant. What stops me publishing reams of slop and future models being trained on the slop?

richardclegg.bsky.social•67 days ago

If you didn't use chatGPT for papers you would use Google or (better) Google scholar. As someone who used both for decades I swapped because AI results were more relevant. I'm a grumpy man in my 50s I'm not swept up in the giddy thrill of the new - it is just better at finding relevant research.

misterabk.bsky.social•67 days ago

OK, fine. I remain doubtful of whether the widespread adoption of the Search Engine That's Often Wrong And You Have To Be A Sceptical Expert To Use Effectively will be a boon for society.

richardclegg.bsky.social•67 days ago

The search engine is already the search engine that is often wrong. :) Follow the first Google hit and ask yourself how often that is right. At least we moved on from the days when your nan would warn you off the internet like it was a cave of ghosts and thieves. (Though maybe she had a point).

richardclegg.bsky.social•67 days ago

With respect I use it and I do scientific research. You don't use it (or pretty clearly not much) and you don't do scientific research. One of us is making faith based statements without particularly good knowledge and I don't think it is me. (Anyway, I enjoy your material either way.)

shermarama.bsky.social•66 days ago

I hear what you're saying - there are definitely specific text-domain things where it's useful - but mentioning those cases doesn't stop all the many and wide misunderstandings and misuses, even among people that both you and I would hope could keep perspective.

shermarama.bsky.social•66 days ago

Personally I'm just kind of holding on until the flap has died down and the niche useful cases have settled in, like where the self-balancing tech went after Segways. But it's teeeedious dealing with everyone who thinks it's their new best friend the magic answer machine, right now...

artisanloaf.bsky.social•67 days ago

the good thing is we have precedent for these arguments.

richardclegg.bsky.social•67 days ago

Yeah. I mean I would not cite Wikipedia for a paper but I am long past being embarrassed that Iook there for maths. (At least in a few places because I wrote part of that page.)

artisanloaf.bsky.social•67 days ago

GenAI/LLMs are just a tool, they can be used well or used poorly, can enable extra productivity/learning, or enable laziness.

the dotcom bubble and hype didn't make the internet worthless.

I have no incentive to convince people, better for me if people do not understand the capabilities tbh.

richardclegg.bsky.social•67 days ago

Oh God. I can remember 1999/2000 crash and people gloating "see, we knew buying things on the internet was a fad."

tabletopcave.bsky.social•67 days ago

No, they used to say that Wikikpedia was an open source resource that anyone could add stuff to, so you shouldn't use a wikipedia entry as primary source (and that still applies). Most pages on wikipedia has citiations you can/should go a check when using it.

tabletopcave.bsky.social•67 days ago

The big problem with AI models are that when asked for sources they are very likely to make up false sources and false citiations (the models have no concept of true/false). So you end up with texts you have difficulties to verify. People not going being an AI text is easily duped by false info.

tabletopcave.bsky.social•67 days ago

We just having a big scandal in Norway when a "report" about school structure in Tromsø was obviously made by AI which cited non-existing books and qoutes from real researchers.

tabletopcave.bsky.social•67 days ago

The adminstration hadn't compiled research about how schools could be structured, they inputed what they wanted the report to say to an AI model to get a "report" to support for shutting down schools. And when giving that input the AI model spouted some nonsene that fulfilled the "brief".

tabletopcave.bsky.social•67 days ago

That's the big problem, AI doesn't actully do anything besides compiling stuff based on input. The person inputing requests need to be skilled both in what and how they choose to input, and be able to check if what the AI delivers is true. For most people pushingAI, that's widely undercommunicated

richardclegg.bsky.social•67 days ago

Not sure when you started to use it. The change over in attitudes was around 2005/2006. I knew Professors who were pretty virulent about it. "Should we ban students using it or at least tell them not to ever do that". (Not citing it, using it at all.) I was covertly writing bits of it at the time.

tabletopcave.bsky.social•67 days ago

Again using your personal expierence doesn't scale up to include all of humanity. I was a student between 2001 and 2010, I know first hand how wikipedia came in, how it was used and how it was not used. And the same caveat for wikipedia use exists today as it did back when it launched.

richardclegg.bsky.social•67 days ago

Absolutely: experience is different with age. You were a student and found it super useful and probably most of your friends used it. I was in the staff meeting convincing elderly profs that it was not evil we should not ban it or warn students against using it but tell them not to cite it.

richardclegg.bsky.social•67 days ago

No believe me they *used* to say "Wikipedia is nonsense because anyone can edit it." "You look at Wikipedia you fool." "I saw in the paper the other day that it said the prime minister is made of jam". There was a genuine hostility. It abated after a comparative reliability study vs Britannica.

tabletopcave.bsky.social•67 days ago

Why should I believe your experience as the only truth? I was a student when wikipedia launched, and know precisely how wikipedia was described and used on my university...

tabletopcave.bsky.social•67 days ago

And wikipedia is still nonsense if you go there without know what wikipedia is. in Taskmaster UK series 5 Mark Watson added a sentence about hats being machines. Of course those things are nonsense.

lizardky.bsky.social•66 days ago

By "cites", I think you mean "makes up".
One of literally hundreds of such examples.
https://bsky.app/profile/nateraymond.bsky.social/post/3l6pk5htcfz26

lizardky.bsky.social•66 days ago

A few more, collected.

https://bsky.app/profile/lizardky.bsky.social/post/3ldgysasqec23

Comments

Posting Rules

Reply