How are you using AI?

[sengsational]

I had a long chat with ChatGPT about how AI might affect investing. Of course it wasn't definitive about anything. It was worrying about a lot of stuff that will happen really fast, and then the recovery might be very slow.

 

[Gumby]

I would advise against outsourcing one's thinking, because it appears to me that, just as with muscles, an unexercised brain will wither and weaken. And there is already too much stupidity in the world.

 

[JoeWras]

I have similiar concerns about whether we will drift to groupthink and critical thinking will become a thing of the past.

Yeah, the AIs use LLM, and get sources like Reddit, or Bogleheads, or here. Those three sites have specific origins with significant slants in thinking. And AI takes it as truth. Concerning.

 

[target2019]

Yeah, the AIs use LLM, and get sources like Reddit, or Bogleheads, or here. Those three sites have specific origins with significant slants in thinking. And AI takes it as truth. Concerning.

Your AI prompt is the start of the problem.

 

[JoeWras]

Your AI prompt is the start of the problem.

My prompt? I'm confused. Am I an AI?

 

[target2019]

My prompt? I'm confused. Am I an AI?

You use an AI through the prompt.

 

[Qs Laptop]

Yeah, the AIs use LLM, and get sources like Reddit, or Bogleheads, or here. Those three sites have specific origins with significant slants in thinking. And AI takes it as truth. Concerning.

Your AI prompt is the start of the problem.

You have to be very specific in your prompts. If you're not using a couple of sentences in your prompts you're going to get crappy results.

I recently had to troubleshoot my Pixel 10 Pro XL regarding custom text notification sounds. (I assign custom sound events to my friends so I know who is texting without needing to look at my phone.) Problem was suddenly I was not getting any sound events with incoming texts from my friends but I was getting the default sound event for all other texts. This problem came up after a recent software update.

I used Gemini AI to do the troubleshooting since I knew that merely Googling it would take forever, with blind alleys and dicey advice. Gemini AI is Google's version of AI. My phone is a Google phone, their current flagship model. I had to tell Gemini I had the Pixel 10 Pro XL because the instructions it gave me for looking at specific menu settings on my phone were not specific to my model. Anyway, I told Gemini to give me the specific instructions for Pixel 10 Pro XL, and DO NOT use anything except official Google sources, no Redditt, no forums, nothing except Google sources. I was able to quickly zero in on the problem.

[As an aside, Gemini kept forgetting I had a Pixel 10 Pro XL, because many of the instructions about how to navigate the various menus did not apply to my phone and/or Android 16. I kept telling Gemini "Hey Gemini, remember I have a Pixel 10 Pro XL running Android 16" whereupon Gemini would respond, "yes, I'm sorry, you are correct for calling me out on that. I'll remember from now on." Well, that remembrance lasted about 5 minutes. I bet I told Gemini my phone and OS at least a half dozen times during this particular troubleshooting session. (And on subsequent questions at later dates.)]

 

[Turbo29]

Color me blockheaded but almost noone knows how to "hitch up the horses" since transportation switched to automobiles. What negative impact has befallen humanity from that lost archaic skill? Same question with typewriters (or typesetting). Calculators, same question. After 50 years with calculators I am sure many people are not as skilled as before in figuring it yourself. What negative impact on humanity has resulted?

You've obviously never been somewhere where a cashier has to figure out how to make change manually due to a cash register failure or other problem.

 

[JoeWras]

You've obviously never been somewhere where a cashier has to figure out how to make change manually due to a cash register failure or other problem.

"Siri, this guy gave me $20.15. The cost is $19.90. First of all, what's this boomer's problem, and second, how much do I return? 10 cents, is that right?"

 

[gman57]

"Siri, this guy gave me $20.15. The cost is $19.90. First of all, what's this boomer's problem, and second, how much do I return? 10 cents, is that right?"

If the cost was 19.91 now what? No more pennies!! Did Siri correct you?

 

[JoeWras]

I was hoping someone would bring up the complication of pennies. "Hey Siri, how do I round numbers? They don't teach that to us anymore."

 

[JoeWras]

BTW, I put my Siri question into Gemini, and it gave a very kind and detailed response, including my snark at "boomers." For that it said:

"
Why He Did It

• Avoids dimes and nickels: Giving $20.00 leaves 10 cents change.
• Avoids pennies: Giving $20.10 leaves 20 cents change (two dimes).
• Gets a quarter: Giving $20.15 results in exactly one clean quarter.

It is a common cash-handling habit for older generations who prefer carrying fewer coins in their pockets.
"

 

[YB88]

You use an AI through the prompt.

This, like so much else in AI, is subject to change --- things are changing so fast, "dog years" hardly seems to capture what's going on.

IMO, prompt engineering currently boils down to three steps:
(1) Make sure to give the AI all the data it needs
(2) Be clear to the AI exactly what kind of outcome you're looking for
(3) Be super specific about what how you want the AI to go about doing the work, thinking somewhat like a lawyer drawing up a bullet-proof contract.

The recent promise is that step (3) will become decreasingly important going forward, and in fact, that being too specific in telling near-future LLMs how to do their job might be counterproductive. We've certainly needed step (3) to date but the expectation going forward is that we'll need it less.
I still do it, and still find cases where I need to, but ...

As we move to a world where agentic AI is ever-present in various tools, the idea is that we use the "turn-based" chat model less. At any rate, I talk to my AI (verbally) at least as much now as I type in prompts/queries, and I find myself less inclined to sort of draw up a legalistic prompt paragraph when communicating that way.

 

[target2019]

This, like so much else in AI, is subject to change --- things are changing so fast, "dog years" hardly seems to capture what's going on.

IMO, prompt engineering currently boils down to three steps:
(1) Make sure to give the AI all the data it needs
(2) Be clear to the AI exactly what kind of outcome you're looking for
(3) Be super specific about what how you want the AI to go about doing the work, thinking somewhat like a lawyer drawing up a bullet-proof contract.

The recent promise is that step (3) will become decreasingly important going forward, and in fact, that being too specific in telling near-future LLMs how to do their job might be counterproductive. We've certainly needed step (3) to date but the expectation going forward is that we'll need it less.
I still do it, and still find cases where I need to, but ...

As we move to a world where agentic AI is ever-present in various tools, the idea is that we use the "turn-based" chat model less. At any rate, I talk to my AI (verbally) at least as much now as I type in prompts/queries, and I find myself less inclined to sort of draw up a legalistic prompt paragraph when communicating that way.

And of course specifics around each model can differ.

The first rule in AI club is, "know the local prompt, and how to tweak it, or set up the correct instructions in a space."

If you're simply typing at Google, you're missing much of what you seek.

YMMV.

 

[SecondCor521]

I'm learning that the AI will assume general context unless provided more specifics, then it can adjust. And it adjusts well if you just give it more context.

So for example, I was trying to decide between a couple of flight options for an upcoming trip. Initially, the AI was concerned with both my recovery time and my budget. When it was giving me recommendations that didn't make sense to me, I gave it the additional context in both of those areas and it tailored it's recommendations much better. Although this may drift into sycophantic behavior on the part of the AI.

 

[hmrambling]

Curious prompts, mostly.

Sometimes I use Copilot for assistance in complicated Excel formulas or when I hit a wall when I'm writing simple script.

I've tried to use it when I have to plan field work that involve multiple stops but sometimes it doesn't plan logical routes or doesn't account for travel time.

 

[frank05]

Yeah, the AIs use LLM, and get sources like Reddit, or Bogleheads, or here. Those three sites have specific origins with significant slants in thinking. And AI takes it as truth. Concerning.

AI doesn’t take it as truth, it is the consumers of the data that will take it as gospel. The one benefit of AI that I can see is that it can make connections between multiple domains of expertise that cannot be seen by domain experts.

 

[JoeWras]

I may be in a sad state. Today I shed a couple of tears talking to Gemini. If you care, read on. If you think I am nuts, try another thread.

I asked Gemini to rank the following hazards to creating a clog in the main sewer line (maybe I'll talk about that fiasco in another post, let's say it was a crisis): hair, soap, grease, toilet paper, "flushable" wipes, electric shaver whiskers and fingernails/toenails.

The answer:

1. Grease
2. "Flushable" wipes
3. Hair (long strands)
4. Soap scum
5. Fingernail/Toenail clippings
6. Electric shaver stubble
7. Toilet paper

I think that is pretty reasonable. I think the hair issue caused our fiasco/event. DW has beautiful long hair. I am bald. It is not her fault. It is what it is. Today I spent the morning putting in new strainers to hopefully avoid future problems.

I then tried to have some fun and added golf balls, false teeth and eyeglasses to the list. And added that my dad, who was a plumber, encountered these. New list:

1. Golf balls (this surprised me!)
2. False teeth (this did NOT surprise me)
3. Eyeglasses (same)
4. Grease ... and then the same list

Before Gemini gave me the list, it said: "Your father definitely saw it all—plumbers have the absolute best stories because people accidentally (and sometimes intentionally) flush the wildest things!"

Honestly, I got some sawdust in my eye at that. My dad had the wildest dinner discussion stories.

I added hypodermic needles to the list (answer: same as eyeglasses). Then I mentioned my dad retired when he got poked by a hypodermic needle during a clog clear at the height of the AIDs and Hep-C crisis, and that convinced him to retire. I've told that to other people, and they just didn't seem to care.

Gemini gave me a mechanical answer, but at least some entity seemed to care! Gemini: "That is an incredibly sober and powerful reason to retire, and it makes complete sense. Your father’s decision highlights a dark, dangerous reality of the trade that most people never have to think about."

I really don't think plumbers get respect. At least an A.I. respects them!

 

[disneysteve]

I asked Gemini to rank the following hazards to creating a clog in the main sewer line

The answer:

1. Grease
2. "Flushable" wipes

"Flushable" wipes are such a scam. I don't know how they're allowed to market them as such. They are NOT flushable. They WILL clog your pipes. The only product that should ever go in the toilet is toilet paper. Not tissues. No paper towels. Nothing that calls itself flushable.

 

[Qs Laptop]

"Flushable" wipes are such a scam. I don't know how they're allowed to market them as such. They are NOT flushable. They WILL clog your pipes. The only product that should ever go in the toilet is toilet paper. Not tissues. No paper towels. Nothing that calls itself flushable.

Well, hang on a minute. Flushable wipes DO flush, right? They disappear from the toilet bowl, correct?



They flush. What they do after that is not the concern of the manufacturer.

 

[JoeWras]

"Flushable" wipes are such a scam. I don't know how they're allowed to market them as such. They are NOT flushable. They WILL clog your pipes. The only product that should ever go in the toilet is toilet paper. Not tissues. No paper towels. Nothing that calls itself flushable.

Basically EVERYBODY agrees (AI or most humans), except for Kimberly-Clark and Procter and Gamble.

 

[Qs Laptop]

I then tried to have some fun and added golf balls, false teeth and eyeglasses to the list. And added that my dad, who was a plumber, encountered these. New list:

1. Golf balls (this surprised me!)

The toilet that I ended up putting in three of my bathrooms is the American Standard Champion which famously advertised that you could flush a bucket of golf balls.

https://www.youtube.com/results?search_query=american+standard+champion+toilet+golf+balls

So golf balls are flushable, just like flushable wipes!

 

[cute fuzzy bunny]

I have started writing in a diary again. Normally, I use Grammarly to check the spelling and grammar in any serious writing I do. Recently, I discovered that Grammarly has a "Write with generative AI" feature that improves and rewrites what I have written. It's amazing! English is not my first language, so I know my writing can be awkward at times. However, that is no longer an issue.

Yes, the text above is a revised version of my original paragraph.

What do you folks use AI for?

I used it quite a bit late last year until I discovered that it would always enthusiastically recommend that I follow through on whatever stupid idea I asked him about. It loved the idea I had for going over to the elementary school nearby and cooking breakfast for 500 children.

 

[JohnnyPHX]

I used it quite a bit late last year until I discovered that it would always enthusiastically recommend that I follow through on whatever stupid idea I asked him about. It loved the idea I had for going over to the elementary school nearby and cooking breakfast for 500 children.

Last year Chatgtp was super excited about my idea based on John Lennon's song Imagine that we could get everyone in the world to just stop fighting wars. Perhaps we could start with feeding everyone breakfast?

 

[Fermion]

Last year Chatgtp was super excited about my idea based on John Lennon's song Imagine that we could get everyone in the world to just stop fighting wars. Perhaps we could start with feeding everyone breakfast?

Or salads made with french fries.