Chat-GPT will change THE WORLD forever

[Zerg Rush]

5 hours ago, ChronA said:

I don't want to take away from the significance of projects like this. It is amazing and an indicator of the marvels to come. But I must stand by my previous assessment. That's based on the date on the video (which predates their first demo video with GPT-3 by a month), the fact that the sequence of responses shows no processing delays at all, and the fact the description frames it as a demo of new actuators, besides my experience with how GPT LMs think (which to my knowledge are the first and only AI so far that show anything approach that level of self awareness).

If the reaction is genuinely spontaneous it is a far more significant breakthough than anything else this group has done... like why would they be wasting their time with robotics if they had those kinds of insights about machine learning? I mean the alternative hypothesis is that these guys have invented some sort of facial recognition and emotive response prediction algorithm that can:

1. Recognize its own reflection as a real-time image of itself.
2. Be surprised to see itself, implying an ability to form expectations and a sophisticated theory of world.
3. Realize it can use its reflection to test its own actuators.
4. Do all of this in real time.

That is an insane list of achievements, any one of which on its own would be able to raise millions in venture capital in a heart beat.

I would love to be proven wrong about this, really, but for extraordinary claims I need extraordinary evidence.  In the meantime, for fun here is how GPT-4 reacts to a surprise mirror: https://chat.openai.com/share/ae22e99e-ef8d-4397-9cf6-55c8f8abc526
(It's a little slower to catch on.)

Agree, as I said, already a long way to go, but experiments of this shows that bots with a real intelligence and selfaware are not longer something for a far future. For good reasons AI companies are  including killswitches.

The real risk of AI is the lack of human intelligence and the abuse they made of it for greedy, manipulative and belic reasons, instead of a usefull tool for everyone. The experiment with ChaosGPT was a good warning.

[Zerg Rush]

5 hours ago, ChronA said:

I don't want to take away from the significance of projects like this. It is amazing and an indicator of the marvels to come. But I must stand by my previous assessment. That's based on the date on the video (which predates their first demo video with GPT-3 by a month), the fact that the sequence of responses shows no processing delays at all, and the fact the description frames it as a demo of new actuators, besides my experience with how GPT LMs think (which to my knowledge are the first and only AI so far that show anything approach that level of self awareness).

If the reaction is genuinely spontaneous it is a far more significant breakthough than anything else this group has done... like why would they be wasting their time with robotics if they had those kinds of insights about machine learning? I mean the alternative hypothesis is that these guys have invented some sort of facial recognition and emotive response prediction algorithm that can:

1. Recognize its own reflection as a real-time image of itself.
2. Be surprised to see itself, implying an ability to form expectations and a sophisticated theory of world.
3. Realize it can use its reflection to test its own actuators.
4. Do all of this in real time.

That is an insane list of achievements, any one of which on its own would be able to raise millions in venture capital in a heart beat.

I would love to be proven wrong about this, really, but for extraordinary claims I need extraordinary evidence.  In the meantime, for fun here is how GPT-4 reacts to a surprise mirror: https://chat.openai.com/share/ae22e99e-ef8d-4397-9cf6-55c8f8abc526
(It's a little slower to catch on.)

Okay, like I said, there's a long way to go already, but experiments on this show that bots with real intelligence and self-awareness are no longer something for the distant future. For good reason, AI companies are including kill switches.

The real risk of AI is the lack of human intelligence and its abuse for greedy, manipulative and warlike reasons, instead of being a useful tool for everyone. The experiment with ChaosGPT was a good warning.

[Zerg Rush]

5 hours ago, ChronA said:

I don't want to take away from the significance of projects like this. It is amazing and an indicator of the marvels to come. But I must stand by my previous assessment. That's based on the date on the video (which predates their first demo video with GPT-3 by a month), the fact that the sequence of responses shows no processing delays at all, and the fact the description frames it as a demo of new actuators, besides my experience with how GPT LMs think (which to my knowledge are the first and only AI so far that show anything approach that level of self awareness).

If the reaction is genuinely spontaneous it is a far more significant breakthough than anything else this group has done... like why would they be wasting their time with robotics if they had those kinds of insights about machine learning? I mean the alternative hypothesis is that these guys have invented some sort of facial recognition and emotive response prediction algorithm that can:

1. Recognize its own reflection as a real-time image of itself.
2. Be surprised to see itself, implying an ability to form expectations and a sophisticated theory of world.
3. Realize it can use its reflection to test its own actuators.
4. Do all of this in real time.

That is an insane list of achievements, any one of which on its own would be able to raise millions in venture capital in a heart beat.

I would love to be proven wrong about this, really, but for extraordinary claims I need extraordinary evidence.  In the meantime, for fun here is how GPT-4 reacts to a surprise mirror: https://chat.openai.com/share/ae22e99e-ef8d-4397-9cf6-55c8f8abc526
(It's a little slower to catch on.)

Okay, like I said, there's a long way to go already, but experiments on this show that bots with real intelligence and self-awareness are no longer something for the distant future. For good reason, AI companies are including kill switches.

The real risk of AI is the lack of human intelligence and its abuse for greedy, manipulative and warlike reasons, instead of being a useful tool for everyone. The experiment with ChaosGPT was a good warning.

[Zerg Rush]

Nice, maybe in a near future it turns real.

 

Edited August 7, 2023 by Zerg Rush

[Arcturus]

I recently installed Comfy UI. Turned this photo into depth and normal maps using Depth Anything v2 and Marigold.

https://www.lvsdecorativearts.co.uk/en-GB/sold-archive/19thc-carved-walnut-panel-of-medieval-man/prod_11757

There's Marigold v2 now which is supposedly better.

https://huggingface.co/spaces/huawei-bayerlab/marigold-v2-web

I played with Trellis 2. At least the version that I was able to run on my machine can't compete with commercial models, but I was able to generate some simple 3d models.

Edit: Asked Chat GPT to extract albedo / color from the original image. This way highlights are removed while small details are retained.

[Arcturus]

Quick test in game:

Edit: green channel was inverted.

[Bikerdude]

On 4/22/2023 at 8:25 AM, Arcturus said:

Stanford and Google created a video game environment in which 25 bots interacted freely.

Edit: it was already posted by jaxa.

Thats quite impressive..

[Bikerdude]

On 9/15/2026 at 12:04 PM, Arcturus said:

A quick test in-game

Parallax on steroids..?

[Arcturus]

55 minutes ago, Bikerdude said:

Parallax on steroids..?

It just shows what you can get with good quality depth map + normalmap.