I posted this question a few minutes ago (Thursday 10:50 am) on Google (Gemini)



I'm sick of this level of incompetence, it happens a lot when asking it about more complex subjects.

 

I posted this question a few minutes ago (Thursday 10:50 am) on Google (Gemini)

View attachment 368987

I'm sick of this level of incompetence, it happens a lot when asking it about more complex subjects.

I think I posted here about a recent experience in which I was looking for whether a particular comedian was going to be performing in the area anytime soon and it came up with a performance that was several months in the past and when I asked if I needed a time machine to book the tickets, told me that I didn't since today was (correct date) and the show (correct date several months earlier) was nearly a year in the future.

A few days ago I was asking it about the a specific event in the story arc of a TV series I started watching a couple weeks ago (the show is from the 2004 time frame) and it went into detail about the central event, describing not only what had happened by why it had happened and how that fit in with the other subplots. The only problem was that, when I finally got to the episode where the event happened, virtually none of what it described was there at all -- it was worse than all of the "inspired by true events" movies in which the "true event" is that a person with that name once lived in that state.

While I continue to be impressed with some of the legitimate results that it produces, the fact that these LLMs still get the simple stuff glaringly wrong on a regular basis should make everyone question everything they spout -- but it doesn't. And I'm not immune to it, either. If I hadn't eventually watched that episode, I would have likely accepted that ChatGPTs description of the events and how it interacted with the overall story arc was accurate.

 

I think I posted here about a recent experience in which I was looking for whether a particular comedian was going to be performing in the area anytime soon and it came up with a performance that was several months in the past and when I asked if I needed a time machine to book the tickets, told me that I didn't since today was (correct date) and the show (correct date several months earlier) was nearly a year in the future.

A few days ago I was asking it about the a specific event in the story arc of a TV series I started watching a couple weeks ago (the show is from the 2004 time frame) and it went into detail about the central event, describing not only what had happened by why it had happened and how that fit in with the other subplots. The only problem was that, when I finally got to the episode where the event happened, virtually none of what it described was there at all -- it was worse than all of the "inspired by true events" movies in which the "true event" is that a person with that name once lived in that state.

While I continue to be impressed with some of the legitimate results that it produces, the fact that these LLMs still get the simple stuff glaringly wrong on a regular basis should make everyone question everything they spout -- but it doesn't. And I'm not immune to it, either. If I hadn't eventually watched that episode, I would have likely accepted that ChatGPTs description of the events and how it interacted with the overall story arc was accurate.

I agree, they get stuff wrong that a child would never get wrong, every kid knows tomorrow is Friday for God's sake. I asked it why it gave me the wrong day of the week and got more gibberish, world salad - there's a word for all this and it isn't "intelligence".

Given this trait one really worries about other "knowledge" these things might impart, at least with software source code the stuff compiles and is - by and large - generally pretty good, but other stuff? like what chemicals to put in a pool or what to give a pet dog if it has X or whether Y can be eaten raw and so on.

I can imagine a bunch of teens having a booze party at an airbnb and relying on AI for advice one the unfamiliar air conditioner or hot tub, and all hell breaking loose.

 

I've heard of one where a woman asked if she should drive or walk to the carwash to get her car washed. (Admittedly that's a setup to see if the AI stumbles.) It checked the weather, found a nice walking route, touted the health advantages of a walk in the fresh air, and answered that she should walk to the carwash.

"What about the purpose of my trip, to get my car washed?"

Even better! Walk back and get the car.

It took several more questions and challenges before the AI gave up and admitted she should drive to the carwash.

 

 

 

I could do without comments from its peanut gallery! After asking a few questions about logic chips...
" You’re really building yourself a little digital lab tonight—I like it. " Huh?!?!?!

 

https://www.theregister.com/ai-and-...s-it-could-lose-the-farm-it-bet-on-ai/5265438

“Our business is, and may continue to be, exposed to risks of customer non-payment and non-performance,” the company wrote. Well, yes.

And even if they pay up, there’s no guarantee its customers will renew their leases. “If customers do not renew their contracts, we may be unable to re-lease, repurpose or assign such capacity on acceptable terms, if at all,” the filing reads.

Customers' ability to pay their bills may not be the only risk factor facing Oracle’s AI gamble. As the company notes, it is already having trouble securing enough power at fair prices to fuel its datacenter buildout.

 

https://www.theregister.com/ai-and-...s-it-could-lose-the-farm-it-bet-on-ai/5265438

“Our business is, and may continue to be, exposed to risks of customer non-payment and non-performance,” the company wrote. Well, yes.

And even if they pay up, there’s no guarantee its customers will renew their leases. “If customers do not renew their contracts, we may be unable to re-lease, repurpose or assign such capacity on acceptable terms, if at all,” the filing reads.

Customers' ability to pay their bills may not be the only risk factor facing Oracle’s AI gamble. As the company notes, it is already having trouble securing enough power at fair prices to fuel its datacenter buildout.

It's hard to really glean much useful information from these filings, as they generally "disclose" every conceivable way that things might go bad so that, no matter what actually happens, they can't be accused of not disclosing it. There have even been filings in which they disclosed the negative impact the revelation of an extraterrestrial presence could have on their stock performance.

 

https://www.foxnews.com/politics/on...rers-says-ai-creating-jobs-not-replacing-them

[Corning] says AI is creating jobs — not replacing them

 

​

 

I know this thread is titled ChatGPT but is largely covering all sorts of AI. In my limited experience with freely available AI tools, they are nowhere near the same.

Going forward I think it would be useful to specify if possible which "AI" is being discussed.

Personally, I've settled on Grok as my go-to AI tool because one of its design goals is truth, not just regurgitation. It can't always deliver an answer but I can't say it's ever lied to me, which I cannot say about the others.

 

View attachment 369030​

That could've been me at one point in my former life.

 

I know this thread is titled ChatGPT but is largely covering all sorts of AI. In my limited experience with freely available AI tools, they are nowhere near the same.

Going forward I think it would be useful to specify if possible which "AI" is being discussed.

Personally, I've settled on Grok as my go-to AI tool because one of its design goals is truth, not just regurgitation. It can't always deliver an answer but I can't say it's ever lied to me, which I cannot say about the others.

I tried to use Grok to test it out, but it insists that I register to use it. No thanks.

 

View attachment 369030​

AI don’t do Excel.

 

AI don’t do Excel.

Actually, it does. At least Grok does.

 

Personally I've avoided using any of them, it's far to early to decide what's truthful and what's BS from these statistical synoptic algorithmics.

 

LLMs are just another tool in its early stages. I reminds me a little of the first days of the home computer, full of clumsy and buggy programs.

 

LLMs are just another tool in its early stages. I reminds me a little of the first days of the home computer, full of clumsy and buggy programs.

So... what's changed?

 

Personally I've avoided using any of them, it's far to early to decide what's truthful and what's BS from these statistical synoptic algorithmics.

That's why their use should be limited only to humans with useful brains.