The OpenAI thread

[Free]

Who wants to know some of the distasteful history of the "OpenAI Guy?"

 

[GoblinCampFollower]

Who wants to know some of the distasteful history of the "OpenAI Guy?"

Thanks for posting the GREAT video!

...I wish I could communicate this information to some people I know who really think those tech billionaires are brilliant.... (they wouldn't listen). I knew very little about Sam Altman so this was enlightening. It's not that I respected him before, but had no idea how much of a creepy weirdo he was.

As much as I dread the AI bubble bursting and taking down the whole economy, it NEEDS to happen sooner rather than later to put a stop to this nonsense. ...of course, our government would likely just bail them out anyway.

 

[CronoCloud Creeggan]

AI guys, Techbro Billionaires....a plethora of creepy weirdos among them.

 

[Noodles]

That horse farm to breed a unicorn analogy is so good.

 

[Free]

OpenAI’s own mental health experts unanimously opposed “naughty” ChatGPT launch

OpenAI draws a line between AI “smut” and porn. Experts fear it’s all unhealthy.

arstechnica.com

OpenAI cannot escape the doom cloud swirling around its rollout of a text-based “adult mode” in ChatGPT.

Late Sunday, The Wall Street Journal reported that insiders confirmed that OpenAI’s “handpicked council of advisers on well-being and AI” were “freaking out” over the company’s plans to move ahead with “adult mode,” despite their urgent warnings.

Back in January, council members unanimously warned OpenAI that “AI-powered erotica could foster unhealthy emotional dependence on ChatGPT for users and that minors could find ways to access sex chats,” sources told the WSJ. One expert suggested that without major updates to ChatGPT, OpenAI risked creating a “sexy suicide coach” for vulnerable users prone to form intense bonds with their companion bots.

Why have them if you don't listen to their input?

Oh, right. For cover.

 

[Casey Pelous]

Lord gawd amighty! These AI things are dumb as stumps!

I'm working on a project to take a rather large document and turn it into a presentation by keeping everything that follows a very well-defined pattern, then formatting it as a Beamer LaTeX presentation. I thought I'd see if this rather dumb and very boring task might be handled by one of these newfangled AI contraptions while I work on something more interesting.

I tried it on Copilot first, and it kind of did okay, but I was forced to learn a new word (against my will) in conjunction with AI: context. As in, "You can't just upload the whole book. The context of Copilot isn't big enough." So I have to split the LaTeX into chapters, turn those into text files, upload them, and then it barfs back out the LaTeX. I'd copy-paste the file into my LaTeX editor, and rinse and repeat. Rather tedious, considering there are over 40 chapters in this document. So much for "work on something more interesting." AI managed to take an already boring task, and turn it into something even more boring. That would still be kinda sorta okay, but it is like working with a toddler with fairly severe ADD. It would go along fine for a few chapters, then just spit out some totally wrong stuff for no apparent reason. Kind of like, "Oh! Are we still doing that thing where we DON'T drive into the oncoming traffic at 100 mph and die a flaming, horrible death? Huh!" I'd tell it what it was supposed to do and, sometimes, it would declare it couldn't possibly do that! "That" being the same thing it had been doing! Maybe not a toddler. Maybe more like a surly, passive-aggressive teenager who, like Dante in Clerks, isn't even supposed to be here today."

So I asked Copilot to tell me how to replace itself, which it blithely did, and now I can run a huge context locally using LM Studio, but OMG this thing is more ADD than Copilot! I've been farting around with it for two hours and it still can't seem to keep more than about four out of ten rules straight. It is also approximately as slow as a wet week. At this pace, it will be done .... *digs out calculator .... * at about 11:35 AM on July 8, 2037.

Man -- this stuff has a LONG way to go before it is even somewhat useful. These managers who are buying this crap are first-order chumps. (But, that's my general opinion of about 90% of managers, anyway.)

ETA: Now I have taken to just taunting Hermes. "Your friend Copilot is faster. Just sayin'....."

 

[CronoCloud Creeggan]

Lord gawd amighty! These AI things are dumb as stumps!

Yes. Well all the stories about AI basically are about that.

I'm working on a project to take a rather large document and turn it into a presentation by keeping everything that follows a very well-defined pattern, then formatting it as a Beamer LaTeX presentation. I thought I'd see if this rather dumb and very boring task might be handled by one of these newfangled AI contraptions while I work on something more interesting.

You'd probably have more luck searching for some shell or Perl/Python script that does it, written by some LaTeX grognard somewhere.

These managers who are buying this crap are first-order chumps. (But, that's my general opinion of about 90% of managers, anyway.)

The companies making this crap are even worse-order chumps.

 

[Casey Pelous]

You'd probably have more luck searching for some shell or Perl/Python script that does it, written by some LaTeX grognard somewhere.

WHO ARE YOU CALLING A GROGNARD, MISSY?

Also, what's a grognard?

 

[Cristiano]

Lord gawd amighty! These AI things are dumb as stumps!

I'm working on a project to take a rather large document and turn it into a presentation by keeping everything that follows a very well-defined pattern, then formatting it as a Beamer LaTeX presentation. I thought I'd see if this rather dumb and very boring task might be handled by one of these newfangled AI contraptions while I work on something more interesting.

I tried it on Copilot first, and it kind of did okay, but I was forced to learn a new word (against my will) in conjunction with AI: context. As in, "You can't just upload the whole book. The context of Copilot isn't big enough." So I have to split the LaTeX into chapters, turn those into text files, upload them, and then it barfs back out the LaTeX. I'd copy-paste the file into my LaTeX editor, and rinse and repeat. Rather tedious, considering there are over 40 chapters in this document. So much for "work on something more interesting." AI managed to take an already boring task, and turn it into something even more boring. That would still be kinda sorta okay, but it is like working with a toddler with fairly severe ADD. It would go along fine for a few chapters, then just spit out some totally wrong stuff for no apparent reason. Kind of like, "Oh! Are we still doing that thing where we DON'T drive into the oncoming traffic at 100 mph and die a flaming, horrible death? Huh!" I'd tell it what it was supposed to do and, sometimes, it would declare it couldn't possibly do that! "That" being the same thing it had been doing! Maybe not a toddler. Maybe more like a surly, passive-aggressive teenager who, like Dante in Clerks, isn't even supposed to be here today."

So I asked Copilot to tell me how to replace itself, which it blithely did, and now I can run a huge context locally using LM Studio, but OMG this thing is more ADD than Copilot! I've been farting around with it for two hours and it still can't seem to keep more than about four out of ten rules straight. It is also approximately as slow as a wet week. At this pace, it will be done .... *digs out calculator .... * at about 11:35 AM on July 8, 2037.

Man -- this stuff has a LONG way to go before it is even somewhat useful. These managers who are buying this crap are first-order chumps. (But, that's my general opinion of about 90% of managers, anyway.)

ETA: Now I have taken to just taunting Hermes. "Your friend Copilot is faster. Just sayin'....."

Gemini has a much larger context by orders of magnitude.

 

[Casey Pelous]

Gemini has a much larger context by orders of magnitude.

Interesting. Obviously, larger than Copilot, but larger than Hermes, too?

 

[Noodles]

Also for local set ups, its going to be dog ass slow depending on hardware and configuration. Without an NPU or GPU, or if the software is not set up to actually use them. Plus all the Cuda/Torxh/etc that these things use, but often that gets installed with the base software.

That said, in my few experiments running local LLMs, the memory was still shit. I had some extremely long back and forth story chats gling in ChatGPT back in 4.whatever when they nerfed it into the ground for anything "dirty", and I tried some short stuff running locally (I forget which model) and it could not even keep two characters straight during the opening prompt.

I keep meaning to go back and try a different model. I am also not sure I have the specs for it. I have like 64gb of RAM I think (maybe 48), but only an 8GB 3070 GPU.

 

[Casey Pelous]

Gemini has a much larger context by orders of magnitude.

Wow! No kidding Gemini has a much larger context! It just gobbled up the whole document and spit out the LaTeX in several sections. Whole job done in a half-hour. I had no idea there was that much difference among the commercial products. Thus far, there's none of the brattiness I had with Copilot, either. (This may have a lot to do with me learning how to write a prompt.

Okay, color me conditionally impressed, and thank you so much Cristiano!

 

[Noodles]

Wow! No kidding Gemini has a much larger context! It just gobbled up the whole document and spit out the LaTeX in several sections. Whole job done in a half-hour. I had no idea there was that much difference among the commercial products. Thus far, there's none of the brattiness I had with Copilot, either. (This may have a lot to do with me learning how to write a prompt.

Okay, color me conditionally impressed, and thank you so much Cristiano!

Copilot is kind of weird, it feels like it has the most actual marketing push and Microsoft wants everyone to use it, but also, at of the ones I have used, it seems like the worst.

ChatGPT seems like the best in a general sense.

Claude is great for Code

Perplexity is weird because it is ok but feels more like a fancy seaech engine.

I have not used Gemini or Grok because Google is Spyware and Fuck Twitter.

Then there is Copilot.

Last place is whatever the hell Facebook is doing.

 

[Casey Pelous]

Last place is whatever the hell Facebook is doing.

Mostly trying desperately to look like they didn't piss away a very large wheelbarrow of cash on their version of SL.

 

[Free]

With Sora shutting down, the partnerships are withering.

Disney cancels $1 billion OpenAI partnership amid Sora shutdown plans

Press reports suggest Disney was blindsided and that no money changed hands.

arstechnica.com

OpenAI’s recently announced plans to shutter its Sora video-generating app have also scuttled the company’s planned $1 billion licensing partnership with Disney, according to multiple press reports.

“As the nascent AI field advances rapidly, we respect OpenAI’s decision to exit the video generation business and to shift its priorities elsewhere,” Disney said in a statement provided to media outlets. “We appreciate the constructive collaboration between our teams and what we learned from it, and we will continue to engage with AI platforms to find new ways to meet fans where they are while responsibly embracing new technologies that respect IP and the rights of creators.”

Oh, I'm certain Disney will find some other way to piss away profits on the new technology frontier.

 

[Noodles]

Wait, I thought AI was the future!

"Chat GOT, make me.a marvel movie with Chris Hemsworth, Tom Holland and Chris Pratt."

Boom, billion dollar movie with no need to pay pesky people.

 

[Free]

Boom, billion dollar movie with no need to pay pesky people.

Except for Chris Hemsworth, Tom Holland and Chris Pratt, right?

 

[Noodles]

Except for Chris Hemsworth, Tom Holland and Chris Pratt, right?

We have AI versions now, we don't need to pay them.

PS, nice Pomni.

 

[Free]

We have AI versions now, we don't need to pay them.

PS, nice Pomni.

The show is far more popular than Firefly. And with less episodes, to boot.

 

[Free]

So much OpenAI news this week. And all about how they're shutting stuff down.

OpenAI “indefinitely” shelves plans for erotic ChatGPT

Some staff reportedly questioned how sexy ChatGPT benefits humanity.

arstechnica.com

Following backlash, OpenAI won’t be rolling out an erotic version of ChatGPT any time soon.

According to the Financial Times, the controversial plan has been shelved “indefinitely” as OpenAI “refocuses” its attention on “core products.”

Insiders told FT that OpenAI mulled scrapping the “adult mode” plan entirely, as even its own advisors warned that ChatGPT users could form unhealthy attachments, which might harm their mental health. One advisor chillingly suggested that the tweak risked turning ChatGPT into a “sexy suicide coach.”

Advisors weren’t the only ones seeing red flags, the report said.