Am I the only one that has NEVER used an AI specific program like ChatGPT, Claude etc. (NOT a Poll)

[GenXguy]

AI keeps getting the same things wrong, even after I school it, and it agrees I'm right (like updated information from a law passed a year ago), then a later chat, it gets it wrong again. And then it forgets things I just told it a few messages back.

 

[ShokWaveRider]

AI keeps getting the same things wrong, even after I school it, and it agrees I'm right (like updated information from a law passed a year ago), then a later chat, it gets it wrong again. And then it forgets things I just told it a few messages back.

That is because a HUGE amount of info on the web, (Where AI gets it's info from is unverified)! But folks (AKA Sheeple) keep falling for it.

 

[COcheesehead]

The problem more and more is people are stopping at the AI answer. They’re ignoring that those answers often are not from reliable sites.

When I do a Google search I only click on the links to sites that I know to be trustworthy and find my answers there.

The reason I laughed was the post said the summary is often wrong, but the source of the wrongness (is that a word?) is the links. So damned if you do, damned if you don’t.

 

[explanade]

AI keeps getting the same things wrong, even after I school it, and it agrees I'm right (like updated information from a law passed a year ago), then a later chat, it gets it wrong again. And then it forgets things I just told it a few messages back.

AI doesn't "forget" because it's not capable of anything analogous to human thinking or information processing.

For all the compute it's used, all the energy, it's not retaining information accurately to use with future queries.

 

[explanade]

The reason I laughed was the post said the summary is often wrong, but the source of the wrongness (is that a word?) is the links. So damned if you do, damned if you don’t.

No the links do not always match the summaries.

That was my point.

So you have to look at the original sources, SEVERAL of them, to synthesize the conclusion, which may or may not align with the summary.

Not only is it making people lazy, because they're starting to just go with the summaries, it's reducing their mental capacity as the studies show.

 

[COcheesehead]

No the links do not always match the summaries.

That was my point.

So you have to look at the original sources, SEVERAL of them, to synthesize the conclusion, which may or may not align with the summary.

Not only is it making people lazy, because they're starting to just go with the summaries, it's reducing their mental capacity as the studies show.

I think my mental capacity is going down from just posting on this forum, but that is neither here nor there.

 

[disneysteve]

The reason I laughed was the post said the summary is often wrong, but the source of the wrongness (is that a word?) is the links. So damned if you do, damned if you don’t.

Not really. The AI summaries give the links to where they supposedly got that information. Numerous times I have clicked on the link, read the entire page, and the info in the summary simply wasn't there. The summary didn't accurately represent what the page actually said.

 

[GenXguy]

AI doesn't "forget" because it's not capable of anything analogous to human thinking or information processing.

For all the compute it's used, all the energy, it's not retaining information accurately to use with future queries.

Of course, it's all 1's and 0's, but I hope you got the general AI point I was making.

 

[GenXguy]

Not really. The AI summaries give the links to where they supposedly got that information. Numerous times I have clicked on the link, read the entire page, and the info in the summary simply wasn't there. The summary didn't accurately represent what the page actually said.

Exactly. I think I mentioned this earlier in this thread, that the links won't have the information AI is spewing, and I know I can't trust AI without confirmation. I'm having to correct it in seemingly every session I have with it.

 

[GenXguy]

That is because a HUGE amount of info on the web, (Where AI gets it's info from is unverified)! But folks (AKA Sheeple) keep falling for it.

Or it's not using the updated information, such as some changes in the ACA beginning 2026, so it provides outdated info. And I've also had AI repeatedly refer to 2026 changes, as if they are going to take effect in the future, and I remind it that we are already in 2026, so those changes were already in effect, such as "expiring enhanced subsidies".

 

[Pellice]

I use it mostly for cooking discussion and book recommendations and such. Mostly for fun. I got good advice when I had food poisoning last December.

Had a great discussion today about the waterfront of ancient Byzantium. It was fun!

 

[KarenF]

I use AI so I don't fall behind. I had it check my business spreadsheet for errors and to scrape the Amazon Kindle page for books I have read. I use it for other things too, including travel. I don't save chats. It is the way of the future, and in my opinion, even if you don't see a need to use it, you should try.

 

[Cerys]

Other than what Google responds with to my searches, I have had no reason to use an AI specific program, let alone subscribe to or converse with one. I really do not see the value to me. I have never had a burning urge to even try one. Enough of my data is out there without adding to it. I have no desire to join the AI herd.

My niece and nephew use it all the time. I tend to think of it as what we learned very early in computerese -- garbage in, garbage out.

I feel strongly about that because I use talk to text all the time, and with the introduction of AI it has, as I like to say, sucked on toast.

I did play with it once, asking the generic question of my name and tell me about myself. It knew nothing, and made amazing stuff up, none of which was true.

But I have no interest in continuing on with this. In fact having AI integrated to talk to text has had a Deliri0us effect on my speech. I'm just going to leave that mess up there... the word it can't get is deleterious. It works best if you speak very slowly, but still does not do such a good job. It's also not a natural way to speak, and if you try to use a natural speech pattern you spend probably as much time fixing it as you would have typing it out in the first place.

I find myself choosing that option more lately. But I am a terrible typist.

 

[TrvlBug]

I am now currently using AI (Alphabet's) to research my next car. Looking to add the 2027 BMW and MB new SUVs being built on EV platforms to my short list containing the Lucid Gravity and Rivian R2. I know about the BMW, but not the MB so decided to waste some of my time. AI tells me because manufacturing is being done in the US, "the electric GLC is highly anticipated to qualify for the $7,500 federal EV tax credit for buyers who choose to lease or who meet the federal purchase requirements". When I called it on that statement, it tells me I am "completely correct and thank you for catching that slip".

This is what scares me about AI...the garbage it spews. I was very specific in my query that these cars were in the build process and not yet available, which it referenced in its answer to me.

 

[ILikeStarTrek]

Other people have discussed what got them to start using AI, here is my 2-cents.

I travel a fair amount and take a lot of pictures. I may have been in a place like London or Paris that has lots of famous buildings or landscapes. When I get back, sometimes I can't remember the name of a building (or mountain or whatever) I was taking a picture of.

Google has an image search where you can upload your image and it will search for matching images (with titles/names). Probably about half the time, if the name of the building / mountain / etc is known at all, the normal image search will find it. But for the ones it fails on, I found that clicking on the AI mode would often find it! The interaction is helpful, as sometimes it doesn't know initially but if I tell it what city I was in, it will then find what I'm looking for. There were times where I have asked it more questions related to the image and generally the AI mode was quite helpful.

I make sure to verify the answer by doing a separate (non-AI) search on the name of the building/mountain/etc it found, and see if pictures of it really do match my picture or not. Over the past year, I have identified at least several dozen of my pictures. I can only think of two times that it gave me a building that was simply wrong when I compared other pictures of that building to my picture. (One of those times, I ended up figuring out the correct building. I spent about a half hour doing further dialog with the AI because it refused to accept that it was wrong! I finally "convinced" it, but it took some doing!)

Once I started using AI regularly for image searches, I started using it for normal searches if the regular search results weren't helpful. (Hard to quantify but maybe about one out of ten of my general informational searches.) Note, for regular searches I use DuckDuckGo instead of Google, but DuckDuckGo doesn't have the image search that Google does - hence I have to use Google for images. Both of the engines have the optional AI mode you can open after getting search results. And there are times that interacting with AI is useful to give better results, as others here have noted.

 

[TrvlBug]

Delving deeper, it provides WRONG info. I specifically asked for other EVs meeting my criteria, but only those built or being built on a true EV only platform. You guessed it, other than the Lucid Gravity, all the other suggested cars were ones built on the multi platform. It didn't even suggest the Rivian!!!

Edit to add: And it suggested a car that has been discontinued.

 

[Jeb-NY]

I do use the free AI and find it very useful, but would not ask it to do creative things accept the occasional birthday cartoon for friends. What I do find it useful for are the following.

I primarily use Claude and Gemini.
I worked in software before retiring 20 years ago and I find my mind needs a bit of help in that field now. I have written web pages for myself and clubs since most people knew what the web was, I think the first was around 1990. I was using Gemini initially to help refresh some old web HTML that I had done for a club I belong to and was not completely happy with the results. My son recommended Claude as being better with code and such. Claude had done a great job of cleaning up old code and switching to more modern HTML style, with good naming and comments so I feel comfortable modifying and extending. Also writes scripts to modify multiple files, that I use to be good at and have forgotten too much. In my real job I did do coding in higher level languages but have no real need for that anymore, but I think Claude could probably help with that also, at least roughing out the frame work.

I now use Gemini for general stuff and Claude for technical help.

Claude has helped me to learn several software packages, even showing me how to modify one of them, a fairly complex public domain database package.

I have used Gemini to compare cars that I was interested in and how to use some of the advanced features of the one I bought, it can find things in the 500+ page fine print PDF manual faster than me.

Both are good at finding details in manuals in seconds, that would take me a long time to find. I made a mistake using H&R Block Income Tax software this year. I knew roughly where to look in the program and the online directions and couldn't figure out what I did wrong. Was about to call H&R, when almost as a joke I asked Gemini. It came back in a few seconds and told me a box I forgot to check and what page it was on. That was the mistake I made, all was ok then. Another time when I was sure what I needed was in the manual because I found it once several years ago, a feature of my smart TV. After digging through manuals for 20 minutes or so, I asked Gemini and seconds later I was told where the hidden feature was located.

Maybe if I wasn't using a 78 year old brain I wouldn't use AI as much, but I find learning new things keeps my brain going better than playing computer games.

So using it for things it might know how to do, or is in a manual that is online, saves a lot of time.

 

[ERD50]

I've had some great success with AI (but some frustrations along the way), and I'm learning that it often just doesn't 'think' the way you'd expect. I recently gave it a very specific request (looking for a thread size for the legs on a specific part from a 'classic/retro' musical instrument from the 1970's a Hohner Clavinet D6), it came back with a very solid sounding reply, but on further investigation on my part, I knew it was wrong. So I started asking it what went wrong...

Part of its reply - the part that I put in bold really blew me away! :

Why I Blew the Answer​

I don't "know" facts by looking them up in a verified master database every time you ask a question. Instead, I function on pattern recognition, predicting the most likely next words based on a massive dataset of text.

In my training data, things like "Hohner," "vintage electric pianos," "metric threads," and "M10" are all closely clustered together in discussions about vintage instrument repair. Because "M10" is a standard metric size frequently discussed in those circles, my system stitched those pieces together into a response that sounded authoritative and plausible, even though it was factually incorrect for the specific part you asked about.

The illusion of confidence is my biggest flaw. I will state a wrong answer with the exact same certainty as a right one because, to my underlying architecture, they both look like highly probable sequences of words.

So I asked: "OK, how do I ask for better answers for a case like this. I thought adding "Clavinet" and "D6" would keep you away from general things like "vintage electric pianos,"?"

AI replied:

In human communication, adding specific details like "Clavinet" and "D6" acts as a filter to discard irrelevant information.

With an AI, it ironically can do the exact opposite. When you narrow the topic down to a highly specific, niche piece of vintage hardware, the pool of actual data I have to draw from shrinks drastically. Instead of saying, "Hey, I only have two forum posts about this and they are a bit vague," my architecture tries to bridge the gaps by blending data from related topics (like other vintage electric pianos or other Hohner parts).

I asked for further tips:

1. Use the "Anti-Hallucination" Clause​

By default, AI models are optimized to provide an answer rather than no answer. You have to explicitly flip that switch in the prompt.

Example Prompt:"What is the thread size on the legs of a Hohner Clavinet D6? If you do not have the exact specification from a service manual or a verified parts database, explicitly state your uncertainty or say 'I don't know' instead of guessing."

I started a new chat with that form, and it told me it didn't know. Better than the wrong answer!

Good news is, I picked up some 8mm x 1.5mm pitch nuts and bolts, and a matching tap at the local big box store, and was able to clear out the threads well enough to get the legs to screw in correctly - one was just so jammed up I couldn't get it started.

Now, am I motivated enough to refurbish this thing? It's in good shape, but need a total refurb, new hammers, strings and damping. But it's less than $200 for that - it's the labor (Hohner Clavinet Refurb Kit). Lots of work, but it might be worth ~ $7,000 when done. But I wouldn't do it for the money, it would have to be for the 'fun' of it, with the money being an 'excuse'.

 

[frayne]

I view AI programs as tools and like any other tool, they are only as effective as the people who know how to use them. A hammer in the hands of a novice can pound nails and fasten boards together, but in the hands of a skilled craftsman, it can be used to create high-quality, useful products. AI is no different.

 

[DayDreaming]

I've had some great success with AI (but some frustrations along the way), and I'm learning that it often just doesn't 'think' the way you'd expect. I recently gave it a very specific request (looking for a thread size for the legs on a specific part from a 'classic/retro' musical instrument from the 1970's a Hohner Clavinet D6), it came back with a very solid sounding reply, but on further investigation on my part, I knew it was wrong. So I started asking it what went wrong...

Part of its reply - the part that I put in bold really blew me away! :

So I asked: "OK, how do I ask for better answers for a case like this. I thought adding "Clavinet" and "D6" would keep you away from general things like "vintage electric pianos,"?"

AI replied:

I asked for further tips:


I started a new chat with that form, and it told me it didn't know. Better than the wrong answer!

Good news is, I picked up some 8mm x 1.5mm pitch nuts and bolts, and a matching tap at the local big box store, and was able to clear out the threads well enough to get the legs to screw in correctly - one was just so jammed up I couldn't get it started.

Now, am I motivated enough to refurbish this thing? It's in good shape, but need a total refurb, new hammers, strings and damping. But it's less than $200 for that - it's the labor (Hohner Clavinet Refurb Kit). Lots of work, but it might be worth ~ $7,000 when done. But I wouldn't do it for the money, it would have to be for the 'fun' of it, with the money being an 'excuse'.

Which AI was this? I've never seen one admit to the reasons for it's mistakes like this one did.

 

[ERD50]

Which AI was this? I've never seen one admit to the reasons for it's mistakes like this one did.

That was Google's Gemini. But I specifically had to ask it why it was wrong (and I'm mostly using the 'extended thinking' option).

I'm learning a lot about how to get the best out of these tools. Like any other tool, there is a skill set associated with it. Maybe later I'll start a thread about how to get the most from AI, or AI tips and tricks (too click-baity?) . I'm also not sure if any/all of this translates to other AI systems that are out there, I suspect they all have their quirks.

 

[audreyh1]

I've had some great success with AI (but some frustrations along the way), and I'm learning that it often just doesn't 'think' the way you'd expect. I recently gave it a very specific request (looking for a thread size for the legs on a specific part from a 'classic/retro' musical instrument from the 1970's a Hohner Clavinet D6), it came back with a very solid sounding reply, but on further investigation on my part, I knew it was wrong. So I started asking it what went wrong...

Part of its reply - the part that I put in bold really blew me away! :

So I asked: "OK, how do I ask for better answers for a case like this. I thought adding "Clavinet" and "D6" would keep you away from general things like "vintage electric pianos,"?"

AI replied:

I asked for further tips:


I started a new chat with that form, and it told me it didn't know. Better than the wrong answer!

Good news is, I picked up some 8mm x 1.5mm pitch nuts and bolts, and a matching tap at the local big box store, and was able to clear out the threads well enough to get the legs to screw in correctly - one was just so jammed up I couldn't get it started.

Now, am I motivated enough to refurbish this thing? It's in good shape, but need a total refurb, new hammers, strings and damping. But it's less than $200 for that - it's the labor (Hohner Clavinet Refurb Kit). Lots of work, but it might be worth ~ $7,000 when done. But I wouldn't do it for the money, it would have to be for the 'fun' of it, with the money being an 'excuse'.

That is very informative and helpful. Thank you.

 

[TrvlBug]

The more I use AI the less I trust it. Last week I asked about a distressed plant and it blamed me for my error in not shaking the bottle of herbicide before applying it. This herbicide has always worked wonders in past years (I always followed the directions on the bottle!). At the nursery a few days later I talked with the plant expert, he laughed and said WRONG, what the problem is is a severe mite infestation this year and he advised how I should treat it.

DH does not trust AI, having worked with expert systems for years. His moto: trust but verify.

 

[PaulR888]

Exactly. I think I mentioned this earlier in this thread, that the links won't have the information AI is spewing, and I know I can't trust AI without confirmation. I'm having to correct it in seemingly every session I have with it.

Financial educator, Paul Merriman, said in a weekly podcast that he uses AI and in the process he tells it what sources he wants it to pull from, sources Paul calls "truth tellers" that he has confidence in. I have not tried that, but interesting.

 

[Lorenzo]

I have not used an AI specific tool like ChatGPT--or if I did try ChatGPT it was very briefly when it first made the news, as I cannot recall. I would like to use these tools, but I just can't think of anything to ask them to help me with. Others have posted some examples of uses that I thought were very clever, and I made a mental note of them in case such circumstances arise. Maybe my projects and interests are either too straightforward to benefit from more than a skim of a few Google search results or too consequential to trust any source other than a human expert.