Is anyone else getting tired of AI?

3,477 Views | 63 Replies | Last: 4 days ago by Lathspell
hph6203
How long do you want to ignore this user?
AG
"Give me your opinion on this thing." Is way different than "Answer this constrained question."

Using it to work on an Unraid server/services running on it has been way easier than searching for answers/teaching myself things I would only need to do once or very infrequently.
AustinAg2K
How long do you want to ignore this user?
Flashdiaz said:

it's not a bubble. It's being used and sold on a scale we're simply not at yet. It is still getting refined and will be better as time goes on.

Isn't that pretty much the definition of a bubble?
AustinAg2K
How long do you want to ignore this user?
satexas said:

AI can be fantastic - but it depends on what you're using it for and under what context.

If you're talking search engines, media generators - then it can be ASS.

If you're talking about private AI engines companies use for their websites, trained in their documents and input - then it can be ASS.

If you're doing finite things like coding with it - it can be fabulous and save TONS and TONS and TONS of time.

If you want it to re-write a letter or document for you, cleaning it up and making it more professional - it can be fabulous.


SUMMARY - The more "finite" or directed your question is, the more like you get back a finite exact response that's accurate. If it has to 'deduce' using chains of logic, then things can really get squirrely.

For the most part, I agree with this, which was kind of my point with the original post. For most things, it's really quite bad at them. There are a couple of things where it can be useful, like coding, if used correctly. However, I feel like it's being shoved down our throats as if it's true AGI, and we'll never have to work again. In reality, it's just a tool that can be used in very certain situations, but it can't do everything. It can't do most things, and yet people are trying to force it in everywhere.
TitanAGGIE09
How long do you want to ignore this user?
AG
The amount of work emails I get from individuals stating laws or referencing things that are blatant AI copy and paste now is astounding..
techno-ag
How long do you want to ignore this user?
AG
The left cannot kill the Spirit of Charlie Kirk.
nai06
How long do you want to ignore this user?
AG
I have lots of thoughts about AI and most are generally negative.

  • It's being sold as the solution to everything that doesn't really solve any problems or issues.
  • It uses a massive amount of resources to produce substandard results.
  • The more of it's own data it ingests, the worse their results get.
  • AI is being used to wrongfully deny medical claims.
  • People have committed suicide because the turned to AI for therapy instead of a licensed professional.
  • Grok is producing sexualized images of people without their consent, including CSAM.
  • LLMs rely on theft of people's work to function.
Outside of a few very specific use cases, AI is garbage. We are worse off as a society because of it.


IrishAg
How long do you want to ignore this user?
rjhtamu said:

AI is a tool. It will continue to improve. Just like a hammer, screwdriver or even a gun, sometimes results will not go as expected. User skill and intention will always apply.

I was just about to post this same exact thing. People treating it as more than a tool are the ones that are going to have problems with it. In addition, you need to know what tool will work best for what job, for example, just because you can get get a result using a general purpose tool doesn't mean that result will be good enough for professional job.

Like anyone using tools, you need to not only understand how to use them, but also understand what is the outcome you need/want? Just because I can work with wood some doesn't mean I should attempt to build cabinets for my wife's kitchen remodel (from a point of view of using the tools I have or assuming I have the skills).
IrishAg
How long do you want to ignore this user?
I'll also add, for those that seem to be pissed off at, or using very specific examples of the large company's LLMs doing stupid things (or being used stupidly), is that you're looking at probably less than 5% of the things that generative AI is used for, from a business point of view.

AI is just marketing buzz, this is all a completely leveled up version of ML and data science, where we are training models to interact with structured/unstructured data more like our brains do. To give interactions to humans in a more human way of consuming information.

Overall the crazy thing most don't realize (especially market research talking about a bubble), is that we are at the beginning of a paradigm shift, not the middle or end. Most people that are doing things with AI are still in LLM training and application building (or apparently in perpetual training mode like most of the name brand ones). Most companies haven't even gotten to the inference phase where you actually make money.

But, long story long, I can guarantee most don't understand AI as it relates to a business, hell I've got my hands in the mud due to information security, learning as much as I can and I still find that I've barely scratched the surface. So take everything you see and hear with a grain of salt on the subject.
AustinAg2K
How long do you want to ignore this user?
IrishAg said:

rjhtamu said:

AI is a tool. It will continue to improve. Just like a hammer, screwdriver or even a gun, sometimes results will not go as expected. User skill and intention will always apply.

I was just about to post this same exact thing. People treating it as more than a tool are the ones that are going to have problems with it. In addition, you need to know what tool will work best for what job, for example, just because you can get get a result using a general purpose tool doesn't mean that result will be good enough for professional job.

Like anyone using tools, you need to not only understand how to use them, but also understand what is the outcome you need/want? Just because I can work with wood some doesn't mean I should attempt to build cabinets for my wife's kitchen remodel (from a point of view of using the tools I have or assuming I have the skills).


The problem with saying it's just a tool is that it's not being sold that way. The Big Tech companies are saying it's basically AGI and can do anything, and dumbass CEOs are eating it up. At some point everyone will realize it's just a tool and there's going to be a major market correction.
IrishAg
How long do you want to ignore this user?
AustinAg2K said:

IrishAg said:

rjhtamu said:

AI is a tool. It will continue to improve. Just like a hammer, screwdriver or even a gun, sometimes results will not go as expected. User skill and intention will always apply.

I was just about to post this same exact thing. People treating it as more than a tool are the ones that are going to have problems with it. In addition, you need to know what tool will work best for what job, for example, just because you can get get a result using a general purpose tool doesn't mean that result will be good enough for professional job.

Like anyone using tools, you need to not only understand how to use them, but also understand what is the outcome you need/want? Just because I can work with wood some doesn't mean I should attempt to build cabinets for my wife's kitchen remodel (from a point of view of using the tools I have or assuming I have the skills).


The problem with saying it's just a tool is that it's not being sold that way. The Big Tech companies are saying it's basically AGI and can do anything, and dumbass CEOs are eating it up. At some point everyone will realize it's just a tool and there's going to be a major market correction.



But that's not exclusive to AI…name any big trend over the last 20 years and it's been sold to C-level as the thing to fix everything.

But again, this is like cloud compute or the smart phone level of tech paradigm shift with how it can fundamentally change multiple levels of business in every industry. So this whole idea of market correction OSS being done by analyst that really don't understand what this is. There will be dips, just like there was with cloud or phones, but it's not a bubble to burst like the dot com era.
nai06
How long do you want to ignore this user?
AG
IrishAg said:

AustinAg2K said:

IrishAg said:

rjhtamu said:

AI is a tool. It will continue to improve. Just like a hammer, screwdriver or even a gun, sometimes results will not go as expected. User skill and intention will always apply.

I was just about to post this same exact thing. People treating it as more than a tool are the ones that are going to have problems with it. In addition, you need to know what tool will work best for what job, for example, just because you can get get a result using a general purpose tool doesn't mean that result will be good enough for professional job.

Like anyone using tools, you need to not only understand how to use them, but also understand what is the outcome you need/want? Just because I can work with wood some doesn't mean I should attempt to build cabinets for my wife's kitchen remodel (from a point of view of using the tools I have or assuming I have the skills).


The problem with saying it's just a tool is that it's not being sold that way. The Big Tech companies are saying it's basically AGI and can do anything, and dumbass CEOs are eating it up. At some point everyone will realize it's just a tool and there's going to be a major market correction.



But that's not exclusive to AI…name any big trend over the last 20 years and it's been sold to C-level as the thing to fix everything.

But again, this is like cloud compute or the smart phone level of tech paradigm shift with how it can fundamentally change multiple levels of business in every industry. So this whole idea of market correction OSS being done by analyst that really don't understand what this is. There will be dips, just like there was with cloud or phones, but it's not a bubble to burst like the dot com era.


What makes you confident that any of the AI engines or LLMs will be able to overcome model collapse as they ingest more AI created content? I think that's likely to be exacerbated when companies can no longer train their models on stolen content.
IrishAg
How long do you want to ignore this user?
nai06 said:

IrishAg said:

AustinAg2K said:

IrishAg said:

rjhtamu said:

AI is a tool. It will continue to improve. Just like a hammer, screwdriver or even a gun, sometimes results will not go as expected. User skill and intention will always apply.

I was just about to post this same exact thing. People treating it as more than a tool are the ones that are going to have problems with it. In addition, you need to know what tool will work best for what job, for example, just because you can get get a result using a general purpose tool doesn't mean that result will be good enough for professional job.

Like anyone using tools, you need to not only understand how to use them, but also understand what is the outcome you need/want? Just because I can work with wood some doesn't mean I should attempt to build cabinets for my wife's kitchen remodel (from a point of view of using the tools I have or assuming I have the skills).


The problem with saying it's just a tool is that it's not being sold that way. The Big Tech companies are saying it's basically AGI and can do anything, and dumbass CEOs are eating it up. At some point everyone will realize it's just a tool and there's going to be a major market correction.



But that's not exclusive to AI…name any big trend over the last 20 years and it's been sold to C-level as the thing to fix everything.

But again, this is like cloud compute or the smart phone level of tech paradigm shift with how it can fundamentally change multiple levels of business in every industry. So this whole idea of market correction OSS being done by analyst that really don't understand what this is. There will be dips, just like there was with cloud or phones, but it's not a bubble to burst like the dot com era.


What makes you confident that any of the AI engines or LLMs will be able to overcome model collapse as they ingest more AI created content? I think that's likely to be exacerbated when companies can no longer train their models on stolen content.


Because that's not where the money is going to be at for most companies in AI. Most companies will build out their application with a specific LLM and RAG (Retrieval-Augmented Generation) based solution that is hyper focused on the data that they want the applications built on AI to manipulate, monitor, analyze, or whatever else they want to do.

Most AI scenarios for businesses (not the general public) isn't ChatGPT or Grok where is just a huge more sophisticated version of Google, it's utilizing very focused designed systems pulling in data from specific sources to accomplish business goals in a manner that is very human friendly. This is when you get past the training stage of LLMs and move into the inference stage, which is a completely different type of infrastructure needed (which is what most are building now)

Everyone is currently hung up on the training stage for a lot of the work being done (especially everything in the media), to understand a little bit better, I'll steal this that Gemini stole from a video about AI training vs inference:
Quote:

"AI training is the resource-intensive phase where a model learns patterns from vast datasets, while inference is the deployment phase where the already-trained model uses that knowledge to make real-time predictions or decisions on new data, like a student learning (training) then taking a test (inference). Training involves complex computations to optimize parameters, whereas inference applies those fixed parameters for quick, practical tasks, requiring less power but constant availability. "


Again, I don't profess to know this stuff even close to what the experts know, I just know what I've self taught because of where the market in infoSec is now and where I see it going. But, to me thinking "AI" is just a bubble and will fade, is the same as thinking "cloud computing" is just some computers in a data center somewhere, or that a smart phone is just a fad (and a very good example of how someone could judge something for the social abuse people have with it, and not realize how the technology fundamentally changes the world). With where AI is right now, I can't see how it won't end up a fundamental shift in how people use technology (whether they realize it's AI or not), just like Cloud Computing and just like the smart phone.

Only reason I'm passionate about the topic, is that I see so many people speaking with such confidence (especially financial analysts) that don't seem to grasp the totality of this as a paradigm shift in how technology will work.
Koko Chingo
How long do you want to ignore this user?
AG
KidDoc said:

I can tell you that Open Evidence has made complex medical searches amazingly easy and accurate for free. I'm not complaining.

Add to that AI Scribes and it has made my life much better. Little to no AI in my feeds because I don't click on them.

My wife is in medical and says their AI scribe also makes sure to add every issue patients discuss and sometimes cash payers do not like it because in the past they got a 2 for 1. They talked to the doc about the primary visit and also something else, typically minor. Now they get billed for that too.
rjhtamu
How long do you want to ignore this user?
AG
It's still the physician's responsibility to make sure the bill reflects the encounter's complexity.

The reality is, according to CMS (the govt) it doesn't take much at all for a medical encounter to be complex with more than one complaint and a handful of chronic conditions.

If anything, AI scribing may be more accurately capturing the complexity.
No material on this site is intended to be a substitute for professional medical advice, diagnosis or treatment. See full Medical Disclaimer.
EclipseAg
How long do you want to ignore this user?
AG
I'm rewatching "Mad Men," and the other day I was trying to remember which Broadway musical was mentioned in a particular episode.

I asked Gemini, and it gave me an answer I knew was wrong. I said that was incorrect, and it came back with another wrong answer. And another. And another.

All sandwiched in an overly wordy, adjective-laded plot summary that are the hallmarks of AI writing.

Basically, it was searching for Broadway musicals that were popular in the early '60s and just inserting them one after the other as a guess.

ETA: The correct answer was "A Funny Thing Happened on the Way to the Forum."
rjhtamu
How long do you want to ignore this user?
AG
I still very much think this thing is in the toddler stage.

I like testing it from time to time with the Wordle puzzle.
Chat GPT 3 couldn't solve it.
Kept forgetting where I had said correct and incorrect letters were.
Gave me nonsense words.
Gave me 4 or 6 letter words.
Acted like a true LLM just predicting words without logic, reasoning or memory.

Chat GPT 5.2 can now solve a Wordle with logic, reasoning and skill comparable to a person in my opinion.

Waiting for it to get better at chess though...
No material on this site is intended to be a substitute for professional medical advice, diagnosis or treatment. See full Medical Disclaimer.
AustinAg2K
How long do you want to ignore this user?
rjhtamu said:

Waiting for it to get better at chess though...

Chess was solved by AI 30 years ago...

What's funny is that since ChatGPT came out, LLMs have pretty much dominated the AI conversation. There have been different forms of AI basically since computers have been around. A lot of them are more useful than LLMs for specialized tasks. However, LLMs have dominated the landscape for the past three years. Personally, I don't think we should be looking for a single AI to do everything. We should be using different forms of AI which are designed for specific tasks (like playing Chess).
rjhtamu
How long do you want to ignore this user?
AG
AustinAg2K said:

rjhtamu said:

Waiting for it to get better at chess though...

Chess was solved by AI 30 years ago...

What's funny is that since ChatGPT came out, LLMs have pretty much dominated the AI conversation. There have been different forms of AI basically since computers have been around. A lot of them are more useful than LLMs for specialized tasks. However, LLMs have dominated the landscape for the past three years. Personally, I don't think we should be looking for a single AI to do everything. We should be using different forms of AI which are designed for specific tasks (like playing Chess).


Agree with this, and yes, I was speaking to LLMs, not chess engines.
No material on this site is intended to be a substitute for professional medical advice, diagnosis or treatment. See full Medical Disclaimer.
AustinAg2K
How long do you want to ignore this user?
EclipseAg said:

I'm rewatching "Mad Men," and the other day I was trying to remember which Broadway musical was mentioned in a particular episode.

I asked Gemini, and it gave me an answer I knew was wrong. I said that was incorrect, and it came back with another wrong answer. And another. And another.

All sandwiched in an overly wordy, adjective-laded plot summary that are the hallmarks of AI writing.

Basically, it was searching for Broadway musicals that were popular in the early '60s and just inserting them one after the other as a guess.

ETA: The correct answer was "A Funny Thing Happened on the Way to the Forum."

I've never watched Mad Men, but a quick google (without AI) tells me that Broadway musicals have been mentioned many times on Mad Men. This was the very first result:
https://playbill.com/article/10-times-mad-men-met-broadway-what-happened-and-what-couldve-happened-com-345149

Which mentions "A Funny Thing Happened on the Way to the Forum" in Season 2, Episode 5.

Anyways, plain ol' Google for the win.
EclipseAg
How long do you want to ignore this user?
AG
AustinAg2K said:

EclipseAg said:

I'm rewatching "Mad Men," and the other day I was trying to remember which Broadway musical was mentioned in a particular episode.

I asked Gemini, and it gave me an answer I knew was wrong. I said that was incorrect, and it came back with another wrong answer. And another. And another.

All sandwiched in an overly wordy, adjective-laded plot summary that are the hallmarks of AI writing.

Basically, it was searching for Broadway musicals that were popular in the early '60s and just inserting them one after the other as a guess.

ETA: The correct answer was "A Funny Thing Happened on the Way to the Forum."

I've never watched Mad Men, but a quick google (without AI) tells me that Broadway musicals have been mentioned many times on Mad Men. This was the very first result:
https://playbill.com/article/10-times-mad-men-met-broadway-what-happened-and-what-couldve-happened-com-345149

Which mentions "A Funny Thing Happened on the Way to the Forum" in Season 2, Episode 5.

Anyways, plain ol' Google for the win.

Interesting!
hph6203
How long do you want to ignore this user?
AG
Gonna guess there's a massive overlap between people that think LLM's are stupid and can't answer a simple question and people that don't pay for LLM's best capabilities.

ChatGPT


Grok
satexas
How long do you want to ignore this user?
AG
You would be correct. There are bad ai engines out there being silly and ruining their reputation but the rest is pure ignorance on how to best use AI and how life changing it is.

As I said in an earlier post, after I've taken the time to teach people how to use and talk to it, it's forever changed them. People just need help with it and don't realize it - and the super cheap subscription goes a long, long way too.
Reading this forum sometimes is like people that can't speak English well trying to differentiate between a "booty call" and a "butt dial".
KidDoc
How long do you want to ignore this user?
AG
Lots of them really are dumb. My current corporate overlords will only allow access to Insightli which is pretty terrible. I was asking a very basic question about Arc Raiders (video game) and it got it wrong.

Yes, ARC Raiders is designed primarily as a co-op PvE (Player vs. Environment) experience.
The core gameplay loop centers around you and your squad
working together to survive against the ARC's robotic forces and collect resources. There isn't a dedicated competitive PvP mode where players fight
against other players.
While the developers have occasionally hinted at the possibility of some form of player interaction or even light competition
in certain elements, the overwhelming focus, as described and shown in trailers and developer interviews, is on:

Squad-based co-op: Teaming up with
friends or other players.
Fighting AI enemies: The ARC's robotic forces are your main adversaries.
Resource gathering and crafting: Building up your
gear to fight tougher threats.
Surviving the ARC's attacks: The environment itself is a threat.

So, if you're looking for a game where you primarily
fight AI and cooperate with other players, ARC Raiders definitely fits that bill. You won't be constantly worried about getting ambushed by another
player squad.

---- Then I told it to fact check itself as that seemed inaccurate--

You are absolutely right to fact-check that! My apologies. My information was based on earlier announcements and assumptions about the game's initial
design philosophy.
The most recent and accurate information from developer Embark Studios unequivocally states that ARC Raiders will now feature PvP
(Player vs. Player) combat.
Here's the corrected and verified information:

ARC Raiders is officially confirmed to have PvP combat.
It was originally
announced as a purely PvE co-op shooter.
Embark Studios later announced a pivot in design, integrating PvP elements.

Why the confusion and
change?
Embark Studios decided to shift the game's direction to include PvP after a period of internal development and reflection. They shared a
development update in late 2023 / early 2024 (specifically around the Game Awards 2023 timeframe and subsequent communications) confirming this
change. They stated they found a stronger, more engaging game by integrating player-versus-player conflict into the existing PvE elements.
How does
PvP work in ARC Raiders?
While full details are still emerging, the current understanding is that it's not a separate "mode" but rather a core part of
the experience within a larger map. Players will still be fighting the ARC machines (PvE), but they will also encounter and compete (or fight) with
other player squads for resources, objectives, and supremacy on the map. This suggests a "PvPvE" or "extraction shooter" style, similar to games like
Escape From Tarkov, The Cycle: Frontier, or potentially even elements of Dark and Darker, where PvE threats are present alongside other hostile player
teams.
Summary of the corrected information:
ARC Raiders WILL feature PvP combat, integrated into its core gameplay loop, alongside its original PvE
elements.
Thank you for catching my error and prompting the correction! It's crucial to have the most up-to-date information, especially with games in
active development.
No material on this site is intended to be a substitute for professional medical advice, diagnosis or treatment. See full Medical Disclaimer.
rjhtamu
How long do you want to ignore this user?
AG
Chat gpt got the wordle in 3 today, btw.
No material on this site is intended to be a substitute for professional medical advice, diagnosis or treatment. See full Medical Disclaimer.
CDub06
How long do you want to ignore this user?
AG
I've found my uses for it. It's never going to be my final draft of anything.

But it's been great at creating complex macros for me to filter and format excel files and running command line stuff. I'm not a programmer, but with these tools, I can create stuff our programmers are impressed with.

I also use LLMs for basic research and to pull together stuff to explore deeper.

It's not dummy proof, but it's a good enough tool for daily use and is getting better.

I've got a ChatGPT subscription, but have been using Gemini's new model more. Claude is in rotation. People are raving about Claude's new model's ability in coding.
EclipseAg
How long do you want to ignore this user?
AG
hph6203 said:

Gonna guess there's a massive overlap between people that think LLM's are stupid and can't answer a simple question and people that don't pay for LLM's best capabilities.

ChatGPT


Grok


I normally use ChatGPT or Copilot on my desktop, but Gemini is easier to access on my phone so I used that one quickly.

I wasn't suggesting that LLMs are "stupid;" just giving an example of how you can't always trust what they tell you.
Koko Chingo
How long do you want to ignore this user?
AG
I primarily do coding and technical stuff with AI, I also have it proof and summarize contracts, purchasing agreements and all that longwinded stuff before I review it all.

The way you ask questions or prompt makes a big difference. For coding and technical things; it also helps a lot if you have some background knowledge.

I have also found the paid versions to be much better. With Gemini Pro it really speeds up reviewing contracts and purchase agreements.

I have a google drive folder with al the company policies and procedures along with a Google Doc of my own notes that I update. My Google doc has both statements and questions for the AI review. i then upload the contract and ask it to review the contract for any conflicts with my company policies and refer to the file names in my Google Drive and also ask it to follow the instructions of my notes in the specific Google Doc.

I then have it populate a Google Sheet with a a summary of all possible discrepancies and it lists the page and paragraph number of the potential discrepancy - cites the section of policy where there is a possible discrepancy and color codes it red - yellow - green based on the severity.

Extra tabs in the sheet are populated with W-9 and COI information including dollar limits and issue & expiration dates.

I just started also having it auto generate an email asking for corrections to be made citing the issues and corrective actions. That part is a work in progress.

All of that happens in seconds and saves me a lot of time. I still read it but its nice having all this stuff identified before hand. So far it has over identified things that are not issues versus skip something.

Most people do not realize you can customize a document as a reference in your prompts. As I update the Google Doc, my results keep getting better and better. Your reference doc can contain anything. Mine is just bulleted sentences of various statements, questions, and instructions with the first paragraph saying please complete the following requesting and instructions. I also will paste in instructions from our contracting people.
satexas
How long do you want to ignore this user?
AG
Yes sir, you are exactly my case in point. Using it correctly in your profession with respects to our current state of advancements.
Reading this forum sometimes is like people that can't speak English well trying to differentiate between a "booty call" and a "butt dial".
Lathspell
How long do you want to ignore this user?
AG
From my experience, Gemini is by far the worst of the major LLM's out there. It's trained on garbage and all you get out of it is garbage.
Refresh
Page 2 of 2
 
×
subscribe Verify your student status
See Subscription Benefits
Trial only available to users who have never subscribed or participated in a previous trial.