The AI discussion thread

Page 8 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Kaido

Elite Member & Kitchen Overlord
Feb 14, 2004
48,315
5,208
136
Deep fake stuff is going to get scary. Imagine how useful it will be for extortion. Deep fake vid of you doing or saying something and if you don't send bitcoin they'll post it everywhere. It will also make legal cases interesting because people will be able to refute evidence and say it's deep fake.

That's already started:


The next few years are going to be both amazing & terrifying for video generation. The video aspect is still very much in the early-development phase, but even basic projects hold TREMENDOUS promise:

 

Kaido

Elite Member & Kitchen Overlord
Feb 14, 2004
48,315
5,208
136
They're also using digital "siblings" to merge vintage photograph in order to counteract the effects of old-time photography. For example, early photography apparently had something called the "old man wrinkle effect". More technical details here:


The AI digital-twinning technique aims to remove old-timey effects to create a more accurate representation of how people ACTUALLY looked back in the day. It's a fascinating project:


Video of the process:


Abraham Lincoln:

1697510180835.png

Henry Ford:

1697510253178.png
 
  • Wow
Reactions: igor_kavinski

[DHT]Osiris

Lifer
Dec 15, 2015
13,765
11,800
146
To clarify, AI is a "crock" as in Hollywood-style self-aware software. It's just fancy if-then software.
But at what point is actual self-aware intelligence different from 'fancy if-then software'? Our own behaviors are arguable very simple in nature, just complex when interfaced with each other. In addition, AI is already far more capable than humans at a limited subset of tasks, and is quite capable of extrapolating for itself far beyond what most people give it credit for.
 

Kaido

Elite Member & Kitchen Overlord
Feb 14, 2004
48,315
5,208
136
But at what point is actual self-aware intelligence different from 'fancy if-then software'? Our own behaviors are arguable very simple in nature, just complex when interfaced with each other. In addition, AI is already far more capable than humans at a limited subset of tasks, and is quite capable of extrapolating for itself far beyond what most people give it credit for.

Never; it's all hype. People want to believe in sentience & subsequently turn off their brains into considering the origins & structures of how the operations are executed in reality. Big data is all a sham; it's a house of cards. That doesn't mean it's not capable of wonderful (and scary!) things, or that it won't be disruptive to multiple industries. But it's all just fancy QBasic with a large data set lol.

The next few years is gonna be wild!

 
  • Like
Reactions: Ken g6

[DHT]Osiris

Lifer
Dec 15, 2015
13,765
11,800
146
Never; it's all hype. People want to believe in sentience & subsequently turn off their brains into considering the origins & structures of how the operations are executed in reality. Big data is all a sham; it's a house of cards. That doesn't mean it's not capable of wonderful (and scary!) things, or that it won't be disruptive to multiple industries. But it's all just fancy QBasic with a large data set lol.

The next few years is gonna be wild!

What I'm getting at, is at what point is actual intelligence indistinguishable from our programming? Like how do you define actual sapient intelligence, and how is that different than 'simple if-then programming'?
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,201
3,764
75
What I'm getting at, is at what point is actual intelligence indistinguishable from our programming? Like how do you define actual sapient intelligence, and how is that different than 'simple if-then programming'?
Trained AIs like neural nets (there are other methods as well) are already beyond "simple if-then programming". Data is stored in a structure that produces interesting outputs, but humans can't correlate the data with the output.

An AI doesn't usually have an inner voice. It can't think something to itself and respond to itself. Though Auto-GPT is a step in that direction.

An AI can't listen to its environment and decide when to respond or when not to. It also can't randomly have a thought, and work that though through to a point where it says something interesting out of the blue. It's a stimulus-response machine. This one I don't know how to solve.

Large language model AIs don't currently have the ability to form new long-term memories. I have some ideas along that line, but LLMs are always likely to be limited in the new information they can retain. Whether this is important for sapience is debatable, but I think it is.

I think a LLM with memory, multiple voices, and the capability to filter through great amounts of input - including those of its other voices - and decide when to respond might be in the vicinity of sapience.

Memory plus random thoughts should be designed to lead to questions, the way a toddler asks why constantly. This leads toward sapience for humans; it might for AIs as well.
 

Kaido

Elite Member & Kitchen Overlord
Feb 14, 2004
48,315
5,208
136
What I'm getting at, is at what point is actual intelligence indistinguishable from our programming? Like how do you define actual sapient intelligence, and how is that different than 'simple if-then programming'?

They run on electricity & processors, so if you shut the power off, no workie! Although I guess the same could be said for us meat-engines LOL
 

[DHT]Osiris

Lifer
Dec 15, 2015
13,765
11,800
146
Trained AIs like neural nets (there are other methods as well) are already beyond "simple if-then programming". Data is stored in a structure that produces interesting outputs, but humans can't correlate the data with the output.

An AI doesn't usually have an inner voice. It can't think something to itself and respond to itself. Though Auto-GPT is a step in that direction.

An AI can't listen to its environment and decide when to respond or when not to. It also can't randomly have a thought, and work that though through to a point where it says something interesting out of the blue. It's a stimulus-response machine. This one I don't know how to solve.

Large language model AIs don't currently have the ability to form new long-term memories. I have some ideas along that line, but LLMs are always likely to be limited in the new information they can retain. Whether this is important for sapience is debatable, but I think it is.

I think a LLM with memory, multiple voices, and the capability to filter through great amounts of input - including those of its other voices - and decide when to respond might be in the vicinity of sapience.

Memory plus random thoughts should be designed to lead to questions, the way a toddler asks why constantly. This leads toward sapience for humans; it might for AIs as well.
All fair points, I'd love to see more of this implemented in the current models. I will point out that at least with gpt4, it functionally has memory within a given conversation. I've gotten pretty far in the weeds with it before on very theoretical (not something humans have done afaik) topics. It's smarter than people give it credit for.
They run on electricity & processors, so if you shut the power off, no workie! Although I guess the same could be said for us meat-engines LOL
Was just gonna say, wait until I tell you how the brain works.
 
  • Like
Reactions: igor_kavinski

Red Squirrel

No Lifer
May 24, 2003
66,962
11,917
126
www.anyf.ca
I don't think we'll ever have to worry about not being able to turn off AI if it does things we don't want since even a highly complex supercomputer is going to have some form of single point of failure, such as power. Say it got control of the building and turned off all the door access or any form of remote access to itself or the BAS so even the IT staff can't get in the server room anymore or do anything remotely, you call Hydro to pull the fuse at the pole and the batteries will eventually run out. And say we live in a future where even the fuses are AI controlled, you just cut the triplex going to the building.

What we do need to worry about is the people that own the AI using it to do things they want but that we don't want. Like replacing jobs.
 

Kaido

Elite Member & Kitchen Overlord
Feb 14, 2004
48,315
5,208
136
Okay this is crazy...Adobe Firefly 2 has a feature called Generative Match, where it will lift the style from a source image & apply it to YOUR image!


It could be a photo, a painting, a drawing, CGI, anything:

1697636324665.png

You can use it for instant fashion design inspiration:

1697636376845.png

You can use it for theme-matching industrial design for products:

1697636411427.png


You can adjust the strength of the inspiration material:

1697636488234.png

What used to require teams of people with highly specialized skills, tools, and powerful computers is now going to be a point & click app. Nothing visual is real anymore lol.
 

JEDI

Lifer
Sep 25, 2001
30,160
3,300
126
Waiting for ai to finish grrm's winds of winter. :eek:
(Game thrones)

Or at least a fan created version:
Feed ai all of game thrones books so far plus the few pages of winds winter than grrm has released so far.
 
Last edited:
  • Haha
Reactions: Ken g6

dank69

Lifer
Oct 6, 2009
34,909
27,680
136
... you call Hydro to pull the fuse at the pole and the batteries will eventually run out. And say we live in a future where even the fuses are AI controlled, you just cut the triplex going to the building.

...
Sounds simple until an army of T-1000s is guarding the entire power chain from generation to grid to AI brain. I imagine in near the future power delivery will be wireless similar to Independence Day alien technology.
 
Jul 27, 2020
14,819
9,099
106
They run on electricity & processors, so if you shut the power off, no workie! Although I guess the same could be said for us meat-engines LOL
You just gave me a crazy idea for a movie. Gonna have to generate the screenplay on an offline GPT to keep it from getting stolen. Thanks!
 
  • Love
Reactions: Kaido

JEDI

Lifer
Sep 25, 2001
30,160
3,300
126
They're also using digital "siblings" to merge vintage photograph in order to counteract the effects of old-time photography. For example, early photography apparently had something called the "old man wrinkle effect".

The AI digital-twinning technique aims to remove old-timey effects to create a more accurate representation of how people ACTUALLY looked back in the day.

Abraham Lincoln:
View attachment 87287

Henry Ford:
View attachment 87288
i prefer the old grizzled look of Abe Lincoln.

Henry Ford: Founder of Ford modeling agency.
that henry ford? :)