AI and the SBO Forums
- Thread starter Dave
- Start date
ChatGPT thinks the third one is AI, based on “Stylistic density and rhythm” and “Human-like imperfection vs. LLM polish”. That sounds plausible to me, but I don’t feel very confident about it. But I see the point you’re making - it won’t always be possible for mods to separate pure AI from AI polished personal writing or pure human writing. For that matter I used ChatGPT to answer your question, so would that make my post AI? It’s a thorny problem. Clearly we don’t want to be bombarded with AI slop, but what’s the best way to do that, maintain human communication on the site, and get people as many accurate answers and as much helpful dialog as possible?
Not thorny at all. I think most of us can id AI almost off the bat, in spite of John's examples.
Dialog by humans is pretty easy to spot.
Nautically speaking: ARGH = Artificially Regenerated Garbage Ha!
If members chose to post material; including AI generated content, I would hope they would identify the source, but would not suggest banning such posts, even if the AI source is not identified. While AI can be very helpful in researching the Internet, and/or composing a text message, AI does generate errors. So that poster puts their reputation a risk, and the rest of us need to disagree with wrong posts as we often do.
Gary, this is a community and my position requires me to deal with violations of the Terms of Use first, and maintain some decorum in the forums and not to dictate what the norms should be. (And some backend stuff that doesn't involve content.) In terms of the AI topic, my role will be to distill the comments into guidelines involving the use of AI in posts, if the community thinks that is what's needed. Prior to making the guidelines public the moderators will have the opportunity to review and comment on them. Hopefully, we, the community can come to a reasonable consensus on this issue.
On an important side note, this is a great community. The members freely share their experiences and knowledge and rarely is it necessary to intervene in a discussion that has gotten out of hand. The most common "problems" the moderators have are:
- Selling in the forums
- Duplicate posts in the forums
The AI is only going to get better though. Compare the responses from 3 year old models to current ones and the differences are pretty significant. We don’t really know where it will be in 5 years, and at what point the improvement might hit some wall or level-off, but I’m sure it won’t always look like it does today. I think this will be a double edged sword - the amount of misinforming responses might decrease, but it might also get harder to tell AI from human. Or maybe the answers will get good enough that we won’t care whether they were AI or human, at least for some questions.
One related thought - there are a lot of different questions on the forums and maybe AI is more acceptable for some answers than others. If someone asks a basic question like “how often should I lubricate my Volvo shaft seal”, maybe an AI answer that regurgitates the manual is good enough, if whoever is posting it knows or verifies the answer is correct. But I certainly wouldn’t want an AI post in the Wednesday Photo Day thread that’s meant to show personal experiences and real images.
Nice. Also we don’t want any Badly Informed LLM Generated Examples (BILGE), and when we’re talking about anchors we don’t want any Derivatively Recycled Incoherent Formulated Text (DRIFT).Nautically speaking: ARGH = Artificially Regenerated Garbage Ha!
(Disclosure - acronyms may be ChatGPT inspired
)
All the more reason to ban the use of AI, 100% on threat of death. Let's keep SBO related to people and not machines. Let's stop this crap HERE and NOW.
Hang 'em high
.
Since this is the Sailboat OWNERS forum, I think if AI owns a sailboat, it should get to participate.
B T Dubs, I just drove all the way out to Ohio to buy my third sailboat. Now I own two Mariners and a Windmill. No AI involved.
Here's hoping I can get on the water first good day in the Spring.
-Will
B T Dubs, I just drove all the way out to Ohio to buy my third sailboat. Now I own two Mariners and a Windmill. No AI involved.
Here's hoping I can get on the water first good day in the Spring.
-Will
Nice ones.Nice. Also we don’t want any Badly Informed LLM Generated Examples (BILGE), and when we’re talking about anchors we don’t want any Derivatively Recycled Incoherent Formulated Text (DRIFT).
(Disclosure - acronyms may be ChatGPT inspired
I didn't use it for my effort, I made it up.
I'll bet you did, too.!
Hmm... How do we know "Hal 9000" isnt moderating the forum now and uses the name "Dave" to throw us off?
TBH, I read that and it immediately makes me think, "AI output, edited to look human"
If you don't use AI, you would presumably learn more during the research, and know how accurate the result might be.
With AI, people just tend to simply regurgitate the output, and not bother checking the sources it lists.
I don't want someone without any practical experience, posting a solution they.simply asked a LLM.
You can often spot that now. As output becomes more convincing, but not necessarily more accurate, it will get harder to separate the wheat from the chaff.
There have been recent reports that AI is actually increasing workload for businesses, due to the need to double check output. Ask the idiots at Deloitte how that goes.
Yup. Often serious errors. Not always obvious ones.
Considering it doesn't have any intelligence and simply plays the statistical odds to generate output.. I suspect that as more people post AI hallucinations as fact, and that is eventually hoovered into the next LLMs... it will just get increasingly "stupid" while looking more convincing.
Like an echo chamber of lay people discussing medical info, they have no clue about. The people who are actual experts are increasingly drowned out by the majority who have no clue.
Small models, trained on verified info, on a narrow subject, may make sense.
Large Models based on the average person's online blather, not so much. My sarcastic post from ages ago, asking "Fellow Captans" is likely in there. Heaven help us.
Last edited:
Personally...
Why make it complicated ?
No thanks, but likely unavoidable.
Seems ok now for the people that use SBO.
Why make it complicated ?
Only if they clearly note it's AI, and they include all the sources.
Yes, if it's Google's AI output instead of actual info from sources.
Last edited:
Anyone who believes everything that is posted from any source is true absolute gospel truth is going to be burned and very disappointed at some point.
All information from a site such as this should be viewed the same as a series of conversations in a friendly neighborhood bar. Some are well intended garbage, some just BS, some absolutely heavenly sent brilliance and the source should be awarded a Nobel Prize, and some a sarcastic attempt at humor that does not always translate well in this medium, most all of it is at least worth what you paid for it, some is a bunch of fluff from folks who just want to run their mouths (keyboards) about everything, and a bit is dangerous. Separating out the wheat from the chaff is the end user's duty.
All information from a site such as this should be viewed the same as a series of conversations in a friendly neighborhood bar. Some are well intended garbage, some just BS, some absolutely heavenly sent brilliance and the source should be awarded a Nobel Prize, and some a sarcastic attempt at humor that does not always translate well in this medium, most all of it is at least worth what you paid for it, some is a bunch of fluff from folks who just want to run their mouths (keyboards) about everything, and a bit is dangerous. Separating out the wheat from the chaff is the end user's duty.
I completely agree with the above, well, except I'm not sure I've seen nobel prize level comments... LOL
However, if I were to be hanging out in my local bar, I would be talking to humans, not machines.
For sure neither man nor machine is error proof. But there existed the original premise here that when someone posted, it was a person. I believe maintaining that distinction needs to continue to exist.
After reading your post, it is not clear to me your position w.r.t. the original question - clarify please.
dj
Saw a pirate walking down the street at the end of last month. It looked fake but I was afraid to ask.
Isn't this true of all responses in general, whether from AI or real people?
When responses from real people are off the mark usually other users jump in to straighten them out or clarify.
I assume the same happens when AI responses are off the mark so what is the difference here?
Don't get me wrong. I am not a particular fan of AI.
I do think it would be good if posters who cut and pasted an AI answer would say something about it being AI generated and whether or not the poster had vetted the answer.
So we can assume AI is not autonomously jumping in and posting answers and the problem then is that actual human users are just copying and pasting or paraphrasing AI answers without vetting them or saying that is what it is. Is that correct?
So I assume that means you have no way of really controlling the posting of AI answers except by asking users to identify them as AI generated and then admonishing them if they don't and they are found out. Is that right?
I
A bit off topic, but for fun I posted The NYT “Connections” game words to ChatGPT and Google Gemini.
If you are not familiar with the game, you get 16 words, and have to determine 4 sets of 4 words that are somehow “connected”.
The categories run from easy to very tricky.
I was impressed that both AI renditions took a screen shot of the 16 words and correctly put them into text and AI could read them.
Unfortunately, neither ChatGPT or Gemini correctly connected the various word choices. They did offer suggested matching, but they were not what the creators at the NYT were looking for. Interestingly, the AI responses “explained” or justified their answers, and even admitted that some of them were a “stretch”.
This game takes some thought, some gut instinct, and a wide knowledge of useless facts. And it requires making some “leaps” in logic.
AI isn’t up to the task...at least not yet.
Greg
If you are not familiar with the game, you get 16 words, and have to determine 4 sets of 4 words that are somehow “connected”.
The categories run from easy to very tricky.
I was impressed that both AI renditions took a screen shot of the 16 words and correctly put them into text and AI could read them.
Unfortunately, neither ChatGPT or Gemini correctly connected the various word choices. They did offer suggested matching, but they were not what the creators at the NYT were looking for. Interestingly, the AI responses “explained” or justified their answers, and even admitted that some of them were a “stretch”.
This game takes some thought, some gut instinct, and a wide knowledge of useless facts. And it requires making some “leaps” in logic.
AI isn’t up to the task...at least not yet.
Greg