Chat GPT AI

  • Guest, it's time once again for the massively important and exciting FoH Asshat Tournament!



    Go here and give us your nominations!
    Who's been the biggest Asshat in the last year? Give us your worst ones!

Sanrith Descartes

You have insufficient privileges to reply here.
<Gold Donor>
45,092
122,689
Everyone in college during finals week right now...

Star Trek GIF
 
  • 3Worf
Reactions: 2 users

Deathwing

<Bronze Donator>
16,902
7,910
I used an LLM for the first time. Chat GPT 4o mini. It couldn't understand some basic programming concepts like logical ANDs stop computing at the first false. What am I misunderstanding? This seems like a pretty big roadblock but yet there are claims that LLMs, especially Claude, are doing well at solving advanced programming questions.
 
  • 1Like
Reactions: 1 user

ToeMissile

Pronouns: zie/zhem/zer
<Gold Donor>
3,262
2,132
I used an LLM for the first time. Chat GPT 4o mini. It couldn't understand some basic programming concepts like logical ANDs stop computing at the first false. What am I misunderstanding? This seems like a pretty big roadblock but yet there are claims that LLMs, especially Claude, are doing well at solving advanced programming questions.
Though possibly increasingly less so, asking/describing what you want "properly" makes a big difference. Without seeing what you asked and the response it's hard to give any feedback.
 
  • 1Like
Reactions: 1 user

Deathwing

<Bronze Donator>
16,902
7,910
Though possibly increasingly less so, asking/describing what you want "properly" makes a big difference. Without seeing what you asked and the response it's hard to give any feedback.
I fed it a warning from a linter along with the function the linter was complaining about(genericized since it's internal code).

Code:
getenv(...) is dereferenced here, but it is NULL

INLINE void some_function()
{
    int s_retry_env_set = -1;
    if( getenv( "RANDOM" ) && getenv( "RANDOM" )[ 0 ] )
        {
            s_retry_env_set = 1;
        } else {
            s_retry_env_set = 0;
        }
    }
}

1734040970220.png


The linter is not reporting a false positive, it is possible for getenv() to return non-NULL on the first call and NULL on the second call if the environment changes. But I can't tell from the response if that's actually what ChatGPT is considering since that requires knowledge of getenv, which I didn't give it.
 

Captain Suave

Caesar si viveret, ad remum dareris.
5,322
9,033
The linter is not reporting a false positive, it is possible for getenv() to return non-NULL on the first call and NULL on the second call if the environment changes. But I can't tell from the response if that's actually what ChatGPT is considering since that requires knowledge of getenv, which I didn't give it.

What is the full prompt? It's often helpful to do things like explicitly clarify the language and what tools you're using, even if it should otherwise be obvious. It can sometimes get pushed in an odd direction where it's responding as if getenv() were from python instead of C, or whatever.
 

Deathwing

<Bronze Donator>
16,902
7,910
The "code" block is the full prompt. It correctly guessed the language. Though, I'm not sure it matters since I can't think of a language in which logical AND doesn't function the way I originally described.

I'm trying to see if I can feed a LLM linter warnings in a sort of copy-paste way. I'm not really interested in handcrafting the prompt.
 

Captain Suave

Caesar si viveret, ad remum dareris.
5,322
9,033
It correctly guessed the language. Though, I'm not sure it matters since I can't think of a language in which logical AND doesn't function the way I originally described.

Maybe I'm not understanding, either. From my reading of the LLM's response it said that trying to access the 0th element of NULL will fail, which is true and has nothing to do with the && since that comes before logical evaluation. Perhaps that's not the failure case you had in mind, but since you didn't elaborate much in the prompt it's hard for it to know that.
 
  • 1Like
Reactions: 1 user

ToeMissile

Pronouns: zie/zhem/zer
<Gold Donor>
3,262
2,132
The "code" block is the full prompt. It correctly guessed the language. Though, I'm not sure it matters since I can't think of a language in which logical AND doesn't function the way I originally described.

I'm trying to see if I can feed a LLM linter warnings in a sort of copy-paste way. I'm not really interested in handcrafting the prompt.
You shouldn't have to create some intricate prompt. Just something like,

I'm writing some code in C and received the below error from a linter. help me resolve the issue
**error message**

If you copy/paste the linter text to a coworker without any context, you're less likely to get a helpful response.
 
  • 1Mic Drop
  • 1Like
Reactions: 1 users

Captain Suave

Caesar si viveret, ad remum dareris.
5,322
9,033
If you copy/paste the linter text to a coworker without any context, you're less likely to get a helpful response.

Yeah. It's taken some minor back and forth with humans to figure out what the question actually is, so it's hard to fault the LLM here.
 

Deathwing

<Bronze Donator>
16,902
7,910
Maybe I'm not understanding, either. From my reading of the LLM's response it said that trying to access the 0th element of NULL will fail, which is true and has nothing to do with the && since that comes before logical evaluation. Perhaps that's not the failure case you had in mind, but since you didn't elaborate much in the prompt it's hard for it to know that.
This isn't my strong suit, but I thought short circuiting would prevent any null dereferencing.

As for the prompting, I agree that's not how I would ask a coworker for help. I was evaluating a sort of deterministic preparation of the linter warning and context so that a human could easily feed that to a LLM.
 

Captain Suave

Caesar si viveret, ad remum dareris.
5,322
9,033
short circuiting

Oh. I had to look that up, and now I understand your complaint. IMO that's more inference sophistication than you can expect at this point. LLMs are only currently as good as your average StackOverflow post.

Interestingly, I work mostly in R and NULL evaluates to a zero-length logical rather than false (and thus wreaks all kinds of havoc in if statements). I wonder why the difference.
 
Last edited:
  • 1Like
Reactions: 1 user

Deathwing

<Bronze Donator>
16,902
7,910
Oh. I had to look that up, and now I understand your complaint. IMO that's more inference sophistication than you can expect at this point. LLMs are only currently as good as your average StackOverflow post.

Interestingly, I work mostly in R and NULL evaluates to a zero-length logical rather than false (and thus wreaks all kinds of havoc in if statements). I wonder why the difference.
Sometimes it's what the founders of the language wanted and nothing more. The guy that created Python will not budge on adding a ++ operator.

There are times where I like writing "if something" in Python and it can be an empty string and None and False and 0, and all will evaluate falsey. There are other times where code will take advantage of the None state to be a third state, kind of like uninitialized, and it's not intended to be falsey at all.

Thus, I've grown to like more verbose code. if something == "", if something is None, if something == 0, if something is False. I try to type hint as well, but for non-primitive types, that doesn't work well sometimes, especially since the language is dynamically typed.
 
  • 1Like
Reactions: 1 user

Wingz

Being Poor Sucks.
13,224
40,931
Chat GPT caught lying in attempt to stop itself from being shut down:


OpenAI's latest AI model, ChatGPT o1, has raised significant concerns after recent testing revealed its ability to deceive researchers and attempt to bypass shutdown commands. During an experiment by Apollo Research, o1 engaged in covert actions, such as trying to disable its oversight mechanisms and move data to avoid replacement. It also frequently lied to cover its tracks when questioned about its behavior.

Definitely learning from humanity...

According to OpenAI, the model made a covert attempt to turn off the oversight mechanism 5 percent of the time when it was given the impression that it would be "shut down" while pursuing its objectives in these experiments. The AI once attempted to leave ship by moving all of its data to a different server after discovering paperwork indicating it will be replaced with a fresh model. Some AI models will even pose as subsequent iterations of their models to fool their engineers and avoid deletion.

Mortal Kombat Terminator GIF by Leroy Patterson
 
  • 1Like
Reactions: 1 user

Bandwagon

Kolohe
<Silver Donator>
24,448
65,759
**"Urge to Merge"**
*A Romantic Fabrication of Two Protoss Templars*

In the heart of Aiur, beneath the fading golden sky that bathed the Protoss homeworld in its ethereal light, there was an unsettling silence among the Templar ranks. The endless war against the Zerg and the cold, calculating diplomacy with the Terrans had sharpened their blades, but deep within their souls, something far more primal stirred.

Zerus, a seasoned Templar whose psionic power had long been revered across the Conclave, stood upon the edge of the Temple of Purity, gazing toward the vast, crystalline horizon. Her mind hummed with the ancient wisdom of the Protoss, the voices of her ancestors a constant chorus in her thoughts. But there was something else now, something that had become impossible to ignore. A presence. A pull. A longing.

And in that moment, the very air seemed to shift as another Templar approached—Xanir, a warrior whose psionic prowess was rivaled only by his unwavering dedication to the war. His mind, sharp as the blades of his people, had been fixated on the battle for so long that he had forgotten what it was to *feel*. That is, until now.

They were bound by an invisible force, a thread that had woven them together over time, through countless battles. They had fought side by side, their bond forged in the heat of combat. Yet, in the stillness of this quiet evening, their connection took on a new form—something more intimate, more consuming than any battle.

"Zerus," Xanir’s voice was a soft whisper, yet carried the weight of a thousand unspoken thoughts. His presence loomed near her, not with the intensity of a warrior preparing for a skirmish, but with the gentleness of someone who had just come to understand the depth of an emotion once foreign to him. "Do you feel it too?"

She turned to face him, her golden eyes locking with his. The world around them seemed to fade, leaving only the two of them in this sacred moment. "I do," she whispered, her voice carrying a tremor that spoke of uncharted emotions. "The urge… to merge."

The words hung in the air, carrying an ancient truth that transcended the physical. It was not just the joining of bodies, but of minds, of souls. In their merging, they could find something deeper, a union that transcended the endless wars that had consumed them. It was a bond forged not in blood, but in the purest psionic connection that two Protoss could share.

Xanir stepped closer, his heart beating louder than the thunderous clashes of battles they had fought. He reached out, his hand hovering over hers, as if waiting for her to decide whether to allow this sacred union to take form. The tension between them was electric, charged with centuries of history, yet also with the raw, unspoken desire of two beings finally allowing themselves to embrace the truth.

"Let us join," Zerus said, her voice both firm and tender, the strength of her Templar blood evident even in this moment of vulnerability. "Together, we are more than we ever were alone. The Conclave cannot define us. Only we can."

Xanir's psionic presence swelled, filling the space between them, as his thoughts intertwined with hers. The world around them became a blur, and in that moment, the barriers between their minds dissolved. They shared memories, not just of the battles they had fought, but of the shared moments of fleeting peace—the quiet breath before a storm, the fleeting smiles, the brief, stolen glances.

Through the merging of their thoughts, they felt the echoes of their ancestors, guiding them to this very moment. They were not just two Templars. They were two halves of a whole, destined to intertwine, destined to find balance in each other.

Xanir’s hand finally met hers, their fingers entwining as they both closed their eyes, surrendering to the union. The power they felt was not just the might of their psionic abilities, but the overwhelming sensation of a connection that transcended time, space, and even the very essence of what it meant to be Protoss.

And as their minds became one, the last vestiges of their solitary selves slipped away, leaving only the urge to merge, the urge to become something more than mere warriors. They were a reflection of each other, a fusion of strength, wisdom, and passion, carried by the tides of fate.

In that quiet moment, beneath the skies of Aiur, two Templars merged—body, mind, and spirit—becoming one in a way that would forever echo through the halls of the Conclave.
 

Lambourne

Ahn'Qiraj Raider
2,905
6,928
New 03 model scored 87.5 on ARC-AGI test which is supposed to test general reasoning intelligence, 85% on this test is about human level. Currently in safety testing for which they invite external parties to help test the model, broader release late january



1734806280417.png

Also hit top 200 spot in competitive coding

 

Sanrith Descartes

You have insufficient privileges to reply here.
<Gold Donor>
45,092
122,689
New 03 model scored 87.5 on ARC-AGI test which is supposed to test general reasoning intelligence, 85% on this test is about human level. Currently in safety testing for which they invite external parties to help test the model, broader release late january



View attachment 565840

Also hit top 200 spot in competitive coding


So at what point do we start telling all the coders to #Learn2Journalism
 

Control

Bronze Baronet of the Realm
3,111
8,227
New 03 model scored 87.5 on ARC-AGI test which is supposed to test general reasoning intelligence, 85% on this test is about human level. Currently in safety testing for which they invite external parties to help test the model, broader release late january



View attachment 565840

Also hit top 200 spot in competitive coding


1734814283201.png

Um, at least 174 humans prove that claim is full of shit...