How do folks feel about Chat GPT texts on the forum?

[fredrussell]

Dunno if this has been discussed before, but there’s been a few threads of late that include a Chat GPT ‘summing up’ of the thread subject. I don’t think of myself as a technophobe, but I’m not sure how I feel about this and I’m interested to know what you lot think.
Personally, I feel the point of forums is the batting back and forth of a subject and the reader can take from that discussion whichever he or she feels are the salient points. Once in a while things get heated but on the whole it’s a good way to get to the bottom of a subject, the more participants the better.
When a Chatbot text is copied and pasted into a thread I feel it’s a bit like Moses turning up with the ten commandments during a community discussion on morality.
Anyway enough from Luddites Anonymous - I realise this technology is here to stay, just wondered if any of you have any thoughts on this?

 

[Sandy]

I am way to old to start using it, so not even looking at it.

 

[GHA]

Dunno if this has been discussed before, but there’s been a few threads of late that include a Chat GPT ‘summing up’ of the thread subject. I don’t think of myself as a technophobe, but I’m not sure how I feel about this and I’m interested to know what you lot think.
Personally, I feel the point of forums is the batting back and forth of a subject and the reader can take from that discussion whichever he or she feels are the salient points. Once in a while things get heated but on the whole it’s a good way to get to the bottom of a subject, the more participants the better.
When a Chatbot text is copied and pasted into a thread I feel it’s a bit like Moses turning up with the ten commandments during a community discussion on morality.
Anyway enough from Luddites Anonymous - I realise this technology is here to stay, just wondered if any of you have any thoughts on this?

Perplexity provides the relevant links to the info sources which are generally very precise and of interest to the topic and tbh the replies produced are generally better & more useful than most of the posts on here. For finding out things web chat rooms aren't really the place to go anymore

 

[Bouba]

I was watching a documentary about how AI can transform teaching at school....like many people I first treated AI as a gimmick...computers having a conversation, it was only polite banality, so it was like a real person who wasn’t really saying anything. Then it evolved into something that could write essays for students so they could cheat at homework, but again it wasn’t anything profound...just drivel mixed with keywords or phrases. Now it’s morphing into something that will either save mankind or destroy it. So I am wary....although I can’t till Elon perfects it for my car

 

[greeny]

Interesting question. In the short term could this be the death of forums such as this? Maybe.
In the long term - Chat GPT has to draw its answer from somewhere. To stay current then there has to be the up to date source of information to draw from. If not from forums, manufacturers data, internet Q and A sources, how else does it formulate its answers?
It reads, it compiles in a very efficient way and answers. No up to date source data means no up to date answer.
Maybe I'm just getting too old for all this.

 

[Bouba]

Interesting question. In the short term could this be the death of forums such as this? Maybe.
In the long term - Chat GPT has to draw its answer from somewhere. To stay current then there has to be the up to date source of information to draw from. If not from forums, manufacturers data, internet Q and A sources, how else does it formulate its answers?
Maybe I'm just getting too old for all this.

Exactly right....AI needs three things, incredible processing power, vast amounts of raw data and huge storage facilities

 

[greeny]

Exactly right....AI needs three things, incredible processing power, vast amounts of raw data and huge storage facilities

Not just raw data, current accurate data if the answers are to be worth anything.

 

[KevinV]

Current "AI" is just a glorified search engine and content aggregator - Google on steroids.

It's fed by content scraped from the rest of the web, so is only as reliable as that source. Feed it 10 articles stating grass is orange and 1 stating it's is green and it will tell you that the consensus view is that grass is orange. It will get even less reliable as it starts ingesting it's own content. Even 5 years ago using the internet to do research was fraught with danger - I found a string of 10 or 11 articles (each referring to each other) saying the same (wrong) thing, eventually traced back to a single unproven wild claim. AI will make tracing things like this harder as it is essentially a "black box".

The old computer addage of BS in, BS out does of course also hold true for
forum members on many subjects - simply regurgitating what we have previously been told.
Where the forum comes into its own is when members share actual experience, and creative thinking - neither of which a bot can do (yet)

 

[GHA]

Current "AI" is just a glorified search engine and content aggregator - Google on steroids.

It's fed by content scraped from the rest of the web, so is only as reliable as that source. Feed it 10 articles stating grass is orange and 1 stating it's is green and it will tell you that the consensus view is that grass is orange. It will get even less reliable as it starts ingesting it's own content. Even 5 years ago using the internet to do research was fraught with danger - I found a string of 10 or 11 articles (each referring to each other) saying the same (wrong) thing, eventually traced back to a single unproven wild claim. AI will make tracing things like this harder as it is essentially a "black box".

The old computer addage of BS in, BS out does of course also hold true for
forum members on many subjects - simply regurgitating what we have previously been told.
Where the forum comes into its own is when members share actual experience, and creative thinking - neither of which a bot can do (yet)

Except it's incredably useful. Perplexity anyway, I set up a laptop with a new ssd to dual boot today with Linux , be very precise what's wanted then taking photos of the screens on the way if anything didn't make sense, show it the image, "now what do I do" . Bang on replies.
Astounding really.
Showed it a picture of a friend's land rover "what's this" no mention it's even a vehical -" looks like a mark one land rover by the resessed headlights and colour..made between.... " .. Loads of very precise details of how it came to the reply without even being told what the image was.
In 2 seconds blows 99% of the replies on here into touch without the bias and grumpiness.

 

[ylop]

In 2 seconds blows 99% of the replies on here into touch without the bias and grumpiness.

Well it may still have the bias… it’s entirely possible for it to very convincingly tell your the wrong answer.

 

[Bouba]

I think that if anyone does use AI in a post...then they should acknowledge it. It’s only fair on the others....

 

[Frank Holden]

I find the

Well it may still have the bias… it’s entirely possible for it to very convincingly tell your the wrong answer.

Yes, Like a B/s Artiste who knows 90% of what he is talking about but makes up the rest.
I gave Perplexity a run on two subjects I am pretty savvy on.
It handled finding time of Meridian Passage and calculation of the latitude remarkably well.
Next up I asked where I could find Travelifts in Chile.
For travelifts it offered up reasonably accurate info, ignored two and mentioned one that has not existed for some considerable time.
Sources were a 16 year old Noonsite post, my own writings on Cruiserswiki, and the phantom travelift was it seems from a site called Portbooker.com which appears to be a dud link.

So it looks like a fancy search engine to me.

Re using it on the forum? Akin to answering a question by just posting google links

 

[Frank Holden]

having seen that ChatGTP is different to Perpelexity I gave it a shot and asked the identical question 'Where are there travelifts in Chile?'
This was its response- a fast food cafe in a shopping mall in La Serena!?

 

[Misterbreeze]

Have a look at this when deciding if it's just glorified google. This is a reputable fact checking website, not tabloid scaremongering AI Models Were Caught Lying to Researchers in Tests — But It's Not Time to Worry Just Yet

 

[ducked]

There, you'd never have found that otherwise.

BIG PITA in academia, where I was peripherally operating until fairly recently, due to its facilitation of industrial-scale plagiarism and fiction.

On fora I suppose there was always a certain amount of BS, but at least there was someone to call BS on.

I had Firefox's rezident ABS module back up an assertion I made recently that burning diesel in an engine intake tract (which I THINK I've seen in a manual somewhere) was sometimes used for preheating,

Which was nice.
It gave its sources, which was nice too.
Unfortunately, none of the sources appeared to support what we were saying.
I suppose it was just agreeing with me.

 

[Frank Holden]

A

Have a look at this when deciding if it's just glorified google. This is a reputable fact checking website, not tabloid scaremongering AI Models Were Caught Lying to Researchers in Tests — But It's Not Time to Worry Just Yet

As I stated Perplexity is just a tricked up search engine and is quite different to ChatGPT. It lists the sources of the information it quotes.
.

 

[garymalmgren]

For finding out things web chat rooms aren't really the place to go anymore

Oh yes they are!!!!
I just found out that Web chat rooms are not the place to find out things,

Merry Christmas all.
gary

 

[KompetentKrew]

I've seen so much nonsense from AI / ChatGPT - no doubt I'll give it another chance at one point.

A year or two ago we started to see a lot of suspicious comments in a forum I moderate - they just sounded a bit off to me, and we realised that evangelists were copy-pasting from ChatGPT or whatever. They could sound incredibly authoritative but you'd notice something in the middle that was completely wrong.

I've also noticed a certain kind of waffle to AI comments - it's like they're doing a student essay, and the first paragraph is just filler to help make up the character count. "A sailboat is a useful vessel for travelling along the coats and across the sea - not only is it exhilarating to travel powered only by the wind, but it reduces your carbon emissions. Seafaring is not without its dangers however, and inexperienced sailors should take lessons from a knowledgeable captain or an RYA or ASA accredited school. A skipper needs to be aware of tides, currents and changing weather conditions, not to mention rocks, wrecks and shallows."

 

[requiem]

As I stated Perplexity is just a tricked up search engine and is quite different to ChatGPT. It lists the sources of the information it quotes.

Good description! Now that I think about it I think you've hit the nail on the head: use the LLM to generate queries, pull the results, and use the LLM to summarize them.

Another analogy I might use is that LLMs are similar to a very lossy image compression algorithm paired with a re-touching algorithm; as when you see an older jpg that's been re-saved a number of times (or for the older crowd, a photocopy of a photocopy and so on), and then use photoshop to clean it up and fill in the bits that were lost.

For example, I have a few LLMs on my laptop. The smallest is only about 1.8 GB in size and the next larger is about 7 GB. Both can confidently give me answers about random historical events. The size of English Wikipedia articles, current text in compressed format, is around 24 GB, and I very much doubt it was the only source for training the models.

One might say this is similar to human memory, where details of past events are often "filled in" by the brain (which is also why eyewitness testimony isn't nearly as reliable as many might think). It also suggests a model for improvement, where topical data can be queried rather than "recalled". This would also be a good approach for mathematical topics: a model that formulates a question, passes it to a tool such as Wolfram, and then returns the result would be more accurate than one that attempts to "solve" it directly.

 

[Refueler]

If AI becomes the norm ... if it gets better at being accurate ... then most of the fun will go from forums ... we'll lose the alternative ways people come up with to solve things ....