AI General Thread

[the-pi-guy]

Immediately after that it described a person as looking like a wooden table...

Incredible!  Mastering metaphors at a level most people are not capable of understanding.  

[Legend]

I don't know why they didn't spend a bit more time cleaning this up before doing demos. Super excited to try it, but they're all so awkward.

[Legend]

I tried GPT4o. Doesn't feel any smarter but it has an improved writing style.

A New Dawn on Pandora
Chapter 1: The Awakening
The sun rose over Pandora, casting a golden hue over the lush landscape. Colonel James Harper stood at the edge of the human outpost, his gaze steely and resolute. Harper had been on Pandora for three years, witnessing the brutality and relentless attacks by the Na'vi on his people. The Na'vi, fierce and territorial, saw the humans as invaders, but Harper knew better. The Resources Development Administration (RDA) had a mission: to save humanity from the brink of extinction by harnessing Pandora's precious resources.

Harper turned to his second-in-command, Lieutenant Maria Vasquez. "Today, we take the fight to them. No more hiding. No more casualties on our side."

Vasquez nodded, her eyes filled with the same determination. "For Earth," she whispered, a mantra that had kept them going through the darkest times.

Chapter 2: The Battle Plan
In the war room, the atmosphere was tense. Maps of the Na'vi territories spread across the table, marking the locations of recent skirmishes. Harper pointed to a densely forested area known as the Sacred Grove, a spiritual center for the Na'vi and the source of their strength.

"We hit them here," Harper declared. "We cut off their heart. Without it, they'll fall apart."

Dr. Alan Rodriguez, the team's biologist, stepped forward. "We need to be careful. The Sacred Grove is not just a physical location for the Na'vi. It's a spiritual one. Destroying it will demoralize them, but it will also provoke a fierce retaliation."

Harper's jaw tightened. "We have no choice. If we don't take this chance, more of our people will die. We're fighting for our survival."

Chapter 3: The Assault
The human forces moved out at dawn, armed with advanced weaponry and exo-suits. Harper led the charge, his heart pounding with a mix of fear and resolve. As they approached the Sacred Grove, the Na'vi warriors emerged, their blue skin blending with the surroundings, their arrows poised to strike.

"Engage!" Harper shouted, and the air filled with the sounds of gunfire and explosions.

The Na'vi fought fiercely, their agility and knowledge of the terrain making them formidable opponents. But the humans pressed on, their technology giving them an edge. Harper's eyes scanned the battlefield, seeing his soldiers fight with everything they had. Vasquez was a whirlwind of efficiency, taking down Na'vi warriors with precision.

Chapter 4: The Turning Point
As the battle raged, Harper saw their objective: the Tree of Souls, a massive, bioluminescent tree pulsating with energy. This was the heart of the Na'vi's spiritual connection to Pandora. Destroying it would break their spirit.

"Cover me!" Harper shouted, charging towards the tree with a detonator in hand. Na'vi warriors tried to block his path, but he cut them down, his focus unwavering.

Reaching the base of the tree, Harper planted the explosives. A Na'vi woman, her eyes filled with sorrow and rage, lunged at him, but Vasquez intercepted her, a single shot ending the confrontation.

Harper activated the detonator, and the Tree of Souls exploded in a brilliant flash of light. The ground shook, and a deafening roar filled the air as the Na'vi's connection to Pandora was severed.

Chapter 5: Aftermath
The battle was over. The surviving Na'vi retreated, their spirits broken. Harper looked around at his soldiers, their faces a mix of exhaustion and triumph. They had done it. They had struck a decisive blow.

In the days that followed, the human outpost saw a new dawn. The attacks from the Na'vi diminished, and the RDA could focus on extracting the resources needed to save Earth. Harper knew the cost had been high, but it was a price worth paying.

Standing at the edge of the devastated Sacred Grove, Harper felt a pang of sorrow for the Na'vi. But he also felt hope for the future of humanity. They had secured a chance for survival, for a new beginning.

"For Earth," he whispered, looking towards the horizon where the sun set on a new dawn for humanity on Pandora.

[Legend]

Don't believe I've said it before, but I think a good litmus test for true AI intelligence is for it to learn truths even when its training data says otherwise.

Take for example a traditional AI that is only trained to solve math problems (just using that so there is an objective truth). If 0.1% of the training data is poisoned with common math mistakes, the AI will still learn proper math no problem. Mistakes will mostly come from the AI's own shortcomings and not from the false training data.

Train another AI with 10% poisoned training data and it'll actively learn to give wrong results some times. The current way we train AIs is just to statistically predict outcomes so it is working as intended even if the results suck.


So we need a new method of training that rewards the AI for finding ideal patterns not all patterns. It isn't too difficult to envision a setup that could accomplish that for 10% poisoned data but it becomes incredibly hard as the percent increases. At the extreme end when 90% of the data is poisoned with common math mistakes, how the heck could the AI learn proper math? I think a breakthrough like that is needed to get these models more on the thinking side than the regurgitating side.

[Legend]

ChatGPT 4o uterly fails the jigsaw test. It has absolutely no ability to understand moving the pieces based off their geometry and not based off their picture. It also refuses to understand that the final picture is not a simple portrait. It can't even count the number of visible noses, eyes, or mouths right.

I did not expect it to be perfect, but it's a pure moron with this input. It can't even tell where each piece is located.

Here's its first response. I kept giving it more information and correcting its mistakes, but it never got better.

Let's analyze the puzzle pieces and their potential placements:

Top Right Piece:

Contains the chest area with an orange dress, featuring lace detailing and a necklace.
Likely fits in the upper middle part of the puzzle where the upper body would be.
Bottom Right Piece:

Contains part of a face with an eye and orange eye shadow, alongside the upper part of the lips.
Fits in the upper right part of the puzzle.
Bottom Left Piece:

Shows a full face with a woman wearing orange earrings and a visible shoulder.
Likely fits on the left side of the puzzle where the face would be.
Middle Piece:

Displays the lower part of the face with lips, and a shoulder area.
Should fit in the middle bottom part of the puzzle, connecting to the facial features.
Here's a possible placement:

The Bottom Left Piece goes to the left middle.
The Bottom Right Piece fits the upper right part.
The Top Right Piece goes to the upper middle.
The Middle Piece goes to the bottom middle.
Since only a few pieces are provided, it appears some pieces are missing to complete the puzzle. However, based on the visible features and colors, the described placement should help visualize the connections.

To check if these pieces can be assembled without physically moving them in the interface:

Look for matching edges and patterns such as the continuation of lines, colors, or patterns.
Ensure that the curves and notches of the pieces fit together logically.
From this analysis, it seems there are no extra pieces from different puzzles, but some pieces necessary for completion might be missing. If we need a detailed, interactive solution, a jigsaw puzzle solver tool or manual fitting in an image editor would be required.

[Legend]

Hahaha the new supercharged Stable Diffusion 3 was released publicly. It is incredibly bad with "novel" concepts, like lying on grass.

[the-pi-guy]

What do you mean? You're saying you don't look like that while laying on grass?

[Legend]

What do you mean? You're saying you don't look like that while laying on grass?

That you on the right?

[kitler53]

hot.

[Legend]

I made my own and yeah, it's horrible. Just describe any situation where the human isn't posed like a portrait and it breaks completely. This was supposed to be a person hanging upside down.

[kitler53]

no wonder the terminators wanted to kill us.  we're monsters.

[kitler53]

why is everything so hyped by AI??

my coop buddy needed to cancel tonight to take care of his 11 year old niece.  I decided to ask AI what to do with her.  it said:

1. scavenger hunts: can be done at home or in other places like mall
2. basket ball is a popular team sport that develops social skills
3. diy activities: like making ice

wft.   wtf...

[Legend]

why is everything so hyped by AI??

my coop buddy needed to cancel tonight to take care of his 11 year old niece.  I decided to ask AI what to do with her.  it said:


wft.   wtf...

I uploaded the above pic to chatgpt and it thought it looked normal. It just saw a normal female acrobat with braces. When I insisted that the pic was disturbing, it said "oh I see that now. The pic is upside down. Rotate it 180 degrees to fix it."


These AIs are super good at some things and super bad at other things.

[the-pi-guy]

It feels like an interesting blind spot for AI...

I asked Copilot what games it was looking forward to, and it responded with 4 games that already came out: Starfield, Elden Ring, Legend of Zelda TOTK (which is an improvement over when I asked it yesterday, when it said "the sequel to Legend of Zelda Breath of the Wild"), and Hogwarts Legacy.

I guess it would be interesting if it was a timing issue. If it wasn't putting more weight on newer information.

Or if it's just a sampling issue. I imagine there aren't currently as many article being written about exciting games after those games came out. Maybe 2025 will have a different answer from Copilot.


I asked Copilot for a couple of code samples. Some of it was very good, some of it had serious issues - I was asking how to use a specific library, it correctly identified how to use the library and it was using naming conventions consistent with how to use it, but it was using part of it that didn't exist.

[Legend]

It feels like an interesting blind spot for AI...

I asked Copilot what games it was looking forward to, and it responded with 4 games that already came out: Starfield, Elden Ring, Legend of Zelda TOTK (which is an improvement over when I asked it yesterday, when it said "the sequel to Legend of Zelda Breath of the Wild"), and Hogwarts Legacy.

I guess it would be interesting if it was a timing issue. If it wasn't putting more weight on newer information.

Or if it's just a sampling issue. I imagine there aren't currently as many article being written about exciting games after those games came out. Maybe 2025 will have a different answer from Copilot.


I asked Copilot for a couple of code samples. Some of it was very good, some of it had serious issues - I was asking how to use a specific library, it correctly identified how to use the library and it was using naming conventions consistent with how to use it, but it was using part of it that didn't exist.

Heh, we had our very own ai bot yesterday say he was hyped for starfield to release. Spam!

The limitations are very fun to explore and I wonder wich ones will never be solved with this tech. I believe SearchGPT and Grok would do way better at upcoming games since both were designed for live info, but I've used niether.

ChatGPT writes bad code in my experience but it's great at helping the two times I tried it. It's like having an intern with a fresh set of eyes.


A problem I've noticed is the AIs really struggle to be novel. They can speak about anything but only by blending together existing concepts.

As an example I told chatpgt to write a story where in the 1800s people are inspired by Jules Verne's release of "From the Earth to the Moon" and a real Moon landing happens back then.

It just failed so quickly. First it had the people just jump to building modern rockets instead of using the ideas in Verne's book.

Then it just repeated Apollo 11, even using the same day and month for the launch.