Yall ready for this?

[dorndawg]

I know we've had a bunch of ai threads lately, but GOODNESS. The speed, size, and scope are already just about overwhelming already and we're just getting started.







https://www.nytimes.com/interactive...e_code=1.1VA.q6Dc.MM_lCP1pUb2_&smid=url-share (free link)

 

[Augustus McCrae]

 

[JackReacherDawg]

My most recent AI thoughts, as I've been using it for high level technical work:

I feel like AI is in its subsidized era, like when Uber was brand new or when streaming was cheap.

The error rate is still insanely high. I've even corrected it, had it acknowledge the error, then go right back to spitting out the same error on the very next query.

It is not just trying to answer your query. It is trying to tell you what you want to hear, based on how you phrased the query and your past inputs (and probably data mining).

Putting this together, all "AI" is is an upgraded search tool. Its a Google search result listing that is instantly analyzed and summarized by a language parser, taught to parse that language with false confidence and slanted to what you want to hear. With a heavy amount of personification from the user/society.

This can still be very useful! For my current role I am using it to analyze hundreds of technical papers at a time. The time savings for me, for free!, are astounding. But i have the sufficient experience and skills to use it without falling for its errors.

Its sort of like having a team of new, green employees that you can give a task to, and a time machine to instantly see their results. Just because they can do tasks fast doesnt mean theyll do them right.

Its not going to get better, other than marginally. It can't be smarter than its inputs. Its already trained now on the entirety of the internet. This is it.

What you see now is what you'll get years from now, just not free or probably even cheap. All the players are just fighting to be the interface that replaces Google search.

 

[Xenomorph]

Data center construction is currently propping up the entire US economy.

That is not hyperbole.

 

[Darryl Steight]

I know we've had a bunch of ai threads lately, but GOODNESS. The speed, size, and scope are already just about overwhelming already and we're just getting started.

View attachment 1376897

View attachment 1376902

View attachment 1376903

https://www.nytimes.com/interactive...e_code=1.1VA.q6Dc.MM_lCP1pUb2_&smid=url-share (free link)

I had no idea one of the AI had disproven planar unit distance conjecture. WTH? How did I miss that? I was thisss close to solving it myself damnit, but I guess I'll stop now.

 

[HailStout]

I can’t be the only one who heard this in their head reading the title

 

[Maroon Eagle]

The error rate is still insanely high. I've even corrected it, had it acknowledge the error, then go right back to spitting out the same error on the very next query.

It is not just trying to answer your query. It is trying to tell you what you want to hear, based on how you phrased the query and your past inputs (and probably data mining).

TLDR — AI is a Six Packer

 

[Bulldog Bruce]

I liked this story from Alcorn State:

Dr. Jason Gibson, a history professor at Alcorn State University in Mississippi, caught 32 out of 35 students cheating on a midterm exam by embedding a hidden prompt in white text. The invisible instruction directed AI chatbots to insert nonsensical references to "Madagascar" into their responses, which students submitted without proofreading.

Gibson’s trap revealed that students were blindly copy-pasting AI-generated answers for an essay on the Industrial Revolution, resulting in bizarre phrases like "Madagascar purple bicycle whispers to the ceiling." He stated his goal was to enforce academic integrity rather than embarrass students, and only two students appealed their failing grades for that section. This incident highlights a broader crisis in higher education, with similar cheating scandals recently reported at Brown University and Princeton University.

It was my theory during High School That the reason some kids were in Advanced classes and others were in the standard classes was the Advanced kids were much better cheaters.

 

[thatsbaseball]

Trump's avid support of data centers makes some a bit "dubious" about them . LOL

 

[GloryDawg]

I think they have approved five new Nuclear Plants. Located in TX and Ohio.

 

[Bulldog Bruce]

AI Data Centers are kinda like Atomic Weapons. They could be very dangerous, but you can't ignore or avoid them because the can has been opened.

 

[JackReacherDawg]

TLDR — AI is a Six Packer

Yes...

"The error rate is still insanely high. I've even corrected it, had it acknowledge the error, then go right back to spitting out the same error on the very next query."

And no...

"It is not just trying to answer your query. It is trying to tell you what you want to hear, based on how you phrased the query and your past inputs (and probably data mining)."

 

[ckDOG]

My most recent AI thoughts, as I've been using it for high level technical work:

I feel like AI is in its subsidized era, like when Uber was brand new or when streaming was cheap.

The error rate is still insanely high. I've even corrected it, had it acknowledge the error, then go right back to spitting out the same error on the very next query.

It is not just trying to answer your query. It is trying to tell you what you want to hear, based on how you phrased the query and your past inputs (and probably data mining).

Putting this together, all "AI" is is an upgraded search tool. Its a Google search result listing that is instantly analyzed and summarized by a language parser, taught to parse that language with false confidence and slanted to what you want to hear. With a heavy amount of personification from the user/society.

This can still be very useful! For my current role I am using it to analyze hundreds of technical papers at a time. The time savings for me, for free!, are astounding. But i have the sufficient experience and skills to use it without falling for its errors.

Its sort of like having a team of new, green employees that you can give a task to, and a time machine to instantly see their results. Just because they can do tasks fast doesnt mean theyll do them right.

Its not going to get better, other than marginally. It can't be smarter than its inputs. Its already trained now on the entirety of the internet. This is it.

What you see now is what you'll get years from now, just not free or probably even cheap. All the players are just fighting to be the interface that replaces Google search.

I have similar experience. It's outright amazing for contract and document summary. I can ask questions about certain aspects and it will answer it and link me to the clause. Countless hours saved in reading, CTRL+F zombie work, and summary documentation. I would pay for a service myself to use in my job if it needed to if for nothing else but freeing my brain up on more thought provoking things. If this tool went away tomorrow I don't know what I would do. It'd be depressing reverting back a more manual world knowing how much more productive I am now with it.

I've also used it for summary work and making certain decisions and responses. That's where you get burned bc sometimes it just craps out and feeds you what you want to hear or won't acknowledge that it's wrong. I feel like I've trained it to think about certain things more accurately but I still fear it's feeing me something that sounds good rather than reality - so I don't rely on it for everything.

My real worry is long term for humans in general. What I'm using it for is making me faster at things I've done thousands of times before because I'm seasoned and know what I'm doing. I can sniff out bullsh!t with actual experience. Someone new in their career is either going 1) consistently get burned by bad analysis or 2) never learn to think critically in their professional area and turn themselves into someone that just feeds a bot all day. Neither of those are good outcomes.

 

[615dawg]

I liked this story from Alcorn State:

Dr. Jason Gibson, a history professor at Alcorn State University in Mississippi, caught 32 out of 35 students cheating on a midterm exam by embedding a hidden prompt in white text. The invisible instruction directed AI chatbots to insert nonsensical references to "Madagascar" into their responses, which students submitted without proofreading.

Gibson’s trap revealed that students were blindly copy-pasting AI-generated answers for an essay on the Industrial Revolution, resulting in bizarre phrases like "Madagascar purple bicycle whispers to the ceiling." He stated his goal was to enforce academic integrity rather than embarrass students, and only two students appealed their failing grades for that section. This incident highlights a broader crisis in higher education, with similar cheating scandals recently reported at Brown University and Princeton University.

It was my theory during High School That the reason some kids were in Advanced classes and others were in the standard classes was the Advanced kids were much better cheaters.

When I was in high school, the idea of a home PC was new and very few of us had one.

We would literally turn in the Encyclopedia Britannica article on papers. I am quite certain that our valedictorian was valedictorian because his family owned a Packard Bell 75MHz with a CDROM.

 

[Maroon Eagle]

We would literally turn in the Encyclopedia Britannica article on papers. I am quite certain that our valedictorian was valedictorian because his family owned a Packard Bell 75MHz with a CDROM.

Back in the pre-Internet days, there was an assignment where I had the best score because we had an Encyclopedia Britannica at home

Fortunately for me an uncle sold encyclopedias as a side hustle

Apropos of the topic, the subject was the scientific concept of mass. One of my classmates wrote about the Christian worship service instead…

 

[dorndawg]

AI Data Centers are kinda like Atomic Weapons. They could be very dangerous, but you can't ignore or avoid them because the can has been opened.

I think you're right, but we and the rest of the world regulate nuclear energy HEAVILY.

 

[thedoubledeuce]

I think they have approved five new Nuclear Plants. Located in TX and Ohio.

May be the best thing to come from all of this.

 

[Negative Waves]

It'll be interesting to see the winners and losers in the upcoming AI crash. At some point investors are going to start checking out once the promised returns don't materialize.

 

[MSUDOG24]

I can’t be the only one who heard this in their head reading the title

View attachment 1377074

If you were a father attending cheerleading competitions in the 90's you certainly did.

 

[Chesusdog]

Let me just say that all the data centers and AI **** are absolutely screwing people who want to buy computer components. SSDs and RAM prices have gone through the roof, hell even HHD prices are crazy, much like GPUs with the crypto mining push a few years back. I've been piecemealing a build for the last year because the cost of parts has tripled if not quadrupled.

 

[DesotoCountyDawg]

Let me just say that all the data centers and AI **** are absolutely screwing people who want to buy computer components. SSDs and RAM prices have gone through the roof, hell even HHD prices are crazy, much like GPUs with the crypto mining push a few years back. I've been piecemealing a build for the last year because the cost of parts has tripled if not quadrupled.

Yes it’s outrageous to build a PC now, especially a gaming PC.

Whenever Sony comes out with the PS6 in a few years there’s no telling how much it will cost.

 

[TilloDwg]

Wait for the AGI and the ASI....fun times.

Also, I saw where Trump has started the process of a new oil refinery. First in over 50 years.

 

[JackReacherDawg]

If you were a father attending cheerleading competitions in the 90's you certainly did.

Lol, this made it click for me. Must've heard it in high school pep.squad dances or something.

 

[patdog]

I think you're right, but we and the rest of the world regulate nuclear energy HEAVILY.

I’d much rather live near a nuclear power plant than a coal or even natural gas plant. Of course, I’d rather not live near anybody of them. But if given a choice …

 

[Chesusdog]

Yes it’s outrageous to build a PC now, especially a gaming PC.

Whenever Sony comes out with the PS6 in a few years there’s no telling how much it will cost.

I expect it to be in the $900 range, and though I would like to think their recent anti-consumer behavior would deter people from paying that, I know they will cave to whatever price Sony puts on it.

 

[Yeti]

Just think how it works in the hands of Lebby…. “I think that although we are up 3 scores with a quarter to go I need to go fast and be aggressive ?”

“ that’s a good thought use as little of the play clock as possible and air it out..want me to show you a few go routes to run?”

 

[mcdawg22]

I can’t be the only one who heard this in their head reading the title

View attachment 1377074

I immediately thought of Jock Jams.

 

[Dawgzilla2]

A shocking number of lawyers have filed court documents citing cases that do mot exist because of AI. Judges are furious about it.

In a recent case in MIssissippi, the attorneys for BOTH SIDES were busted. All four attorneys were removed from the case, and two of them were suspended from appearing in the Northern District for two years.

I know a lot of attorneys don't have a paralegal who can check every case citation in their brief for them. But is proof reading really that hard?

I use AI to help me out with first drafts, but I always double and triple check my documents for accuracy even if I wrote all of it myself. I cited the wrong case in a brief once when I was first starting out, and I'm still not over it. I can't imagine citing a case that doesnt even exist, or having sentences that don't make sense.

And what do you say to your client? "Hey, I know I charged you 40 hours for that brief, but I didn't actually write it and now you have to find a new lawyer."

 

[jethreauxdawg]

Data center construction is currently propping up the entire US economy.

That is not hyperbole.

Data center groups are buying up every empty warehouse as well.

 

[1msucub]

 

[horshack.sixpack]

I know we've had a bunch of ai threads lately, but GOODNESS. The speed, size, and scope are already just about overwhelming already and we're just getting started.

View attachment 1376897

View attachment 1376902

View attachment 1376903

https://www.nytimes.com/interactive...e_code=1.1VA.q6Dc.MM_lCP1pUb2_&smid=url-share (free link)

Imagine AI killing is because we have no natural resources left and not because of a robotic army

 

[horshack.sixpack]

My most recent AI thoughts, as I've been using it for high level technical work:

I feel like AI is in its subsidized era, like when Uber was brand new or when streaming was cheap.

The error rate is still insanely high. I've even corrected it, had it acknowledge the error, then go right back to spitting out the same error on the very next query.

It is not just trying to answer your query. It is trying to tell you what you want to hear, based on how you phrased the query and your past inputs (and probably data mining).

Putting this together, all "AI" is is an upgraded search tool. Its a Google search result listing that is instantly analyzed and summarized by a language parser, taught to parse that language with false confidence and slanted to what you want to hear. With a heavy amount of personification from the user/society.

This can still be very useful! For my current role I am using it to analyze hundreds of technical papers at a time. The time savings for me, for free!, are astounding. But i have the sufficient experience and skills to use it without falling for its errors.

Its sort of like having a team of new, green employees that you can give a task to, and a time machine to instantly see their results. Just because they can do tasks fast doesnt mean theyll do them right.

Its not going to get better, other than marginally. It can't be smarter than its inputs. Its already trained now on the entirety of the internet. This is it.

What you see now is what you'll get years from now, just not free or probably even cheap. All the players are just fighting to be the interface that replaces Google search.

It has already trained on all the available data, both legally acquired and stolen. It’s pretty terrible for accuracy but for agents with closed training data sets and closed query sets. Meanwhile singularity alarmists will persist.

 

[horshack.sixpack]

I had no idea one of the AI had disproven planar unit distance conjecture. WTH? How did I miss that? I was thisss close to solving it myself damnit, but I guess I'll stop now.

Carry on. We need you to check AI. It isn’t very smart.

 

[horshack.sixpack]

Yes it’s outrageous to build a PC now, especially a gaming PC.

Whenever Sony comes out with the PS6 in a few years there’s no telling how much it will cost.

I’ll be riding this PS5 Pro for a while.

 

[horshack.sixpack]

Just think how it works in the hands of Lebby…. “I think that although we are up 3 scores with a quarter to go I need to go fast and be aggressive ?”

“ that’s a good thought use as little of the play clock as possible and air it out..want me to show you a few go routes to run?”

AI doesn’t hallucinate enough to make worse decisions than we saw last year.

 

[horshack.sixpack]

A shocking number of lawyers have filed court documents citing cases that do mot exist because of AI. Judges are furious about it.

In a recent case in MIssissippi, the attorneys for BOTH SIDES were busted. All four attorneys were removed from the case, and two of them were suspended from appearing in the Northern District for two years.

I know a lot of attorneys don't have a paralegal who can check every case citation in their brief for them. But is proof reading really that hard?

I use AI to help me out with first drafts, but I always double and triple check my documents for accuracy even if I wrote all of it myself. I cited the wrong case in a brief once when I was first starting out, and I'm still not over it. I can't imagine citing a case that doesnt even exist, or having sentences that don't make sense.

And what do you say to your client? "Hey, I know I charged you 40 hours for that brief, but I didn't actually write it and now you have to find a new lawyer."

If AI results are going to be used for anything you aren’t the expert in you’d better check it’s references because even when it gets a citation correct it often does not interpret and present the data correctly.

 

[DesotoCountyDawg]

I’ll be riding this PS5 Pro for a while.

I think we are still at least three years from the 6.

 

[NWADawg]

My most recent AI thoughts, as I've been using it for high level technical work:

I feel like AI is in its subsidized era, like when Uber was brand new or when streaming was cheap.

The error rate is still insanely high. I've even corrected it, had it acknowledge the error, then go right back to spitting out the same error on the very next query.

It is not just trying to answer your query. It is trying to tell you what you want to hear, based on how you phrased the query and your past inputs (and probably data mining).

Putting this together, all "AI" is is an upgraded search tool. Its a Google search result listing that is instantly analyzed and summarized by a language parser, taught to parse that language with false confidence and slanted to what you want to hear. With a heavy amount of personification from the user/society.

This can still be very useful! For my current role I am using it to analyze hundreds of technical papers at a time. The time savings for me, for free!, are astounding. But i have the sufficient experience and skills to use it without falling for its errors.

Its sort of like having a team of new, green employees that you can give a task to, and a time machine to instantly see their results. Just because they can do tasks fast doesnt mean theyll do them right.

Its not going to get better, other than marginally. It can't be smarter than its inputs. Its already trained now on the entirety of the internet. This is it.

What you see now is what you'll get years from now, just not free or probably even cheap. All the players are just fighting to be the interface that replaces Google search.

We are using it more and more at work and it definitely has strengths and weaknesses. AI is great at things like - "look at these large data sets across numerous locations and show me where there are inconsistencies."
We've used it for statistical analysis and it gets you very close but has what we call "AI rounding error". Run the same data multiple times and always get almost the exact same answer- hardly ever the exact same answer. Sometimes tenths don't matter, sometimes they do.
Our concern is that (sorta like the students in the story) is that people will not just fail to verify. They will lose the background knowledge that comes from being able do it the old fashioned way and no longer know what or how to verify. How do you know if data is correct if you don't understand enough to know what right should look like?

 

[Darryl Steight]

I expect it to be in the $900 range, and though I would like to think their recent anti-consumer behavior would deter people from paying that, I know they will cave to whatever price Sony puts on it.

Same consumerist sheep who will pay $1,000 for a pair of tennis shoes, or sleep overnight on the sidewalk outside the Apple store before the big new phone drop.

"The things you own, end up owning you"- Tyler Durden

- Darryl Steight