Leo's amazing predictions

[Natasha Tori Maru]

2 hours ago, something_else said:

Intelligent behaviour is always coming from data and experience in some way or another

You most likely will need to define what you mean by 'intelligent behaviour' because Leo and yourself may be speaking about entirely different concepts. You seem to be pointing toward any organism operating like it's own LLM entrained on its own "experience" data? Wild concept (might not be your stance) to think we are all LLMs skittling around, entrained on experience  

I think a more defensible statement might be "Intelligent behaviour does not emerge from nothing; it is typically sourced from prior information, whether learned, inferred, evolved biological structures, or innate."

[Leo Gura]

2 hours ago, something_else said:

It produces a different kind of intelligence but it is still intelligence

AI has some kind of intelligence, but it ain't AGI intelligence. This is easily shown with tests. This is not even controversial at this point. There is a difference in kind here, not just degree.

Meta's former chief AI researcher explains this. Yann LeCun. LLM architecture is not capable of AGI. The mystery is, how useful can the current architecture be in the real world? That's an open question. My bet is that it will be less useful than people hope. But I could be wrong about that.

Edited 1 hour ago by Leo Gura

[Jodistrict]

There is another model of AI, contrary to the general AI model, and that is the model of AI as a personal assistant.  It’s like having your own personal secretary or employee.  It does the grunge works and frees your time for higher thinking.  It has been mentioned that the current AI can’t write significant software, but the real question is how much more efficient is it making the current programmers.   As a simple example, when I get a new product and want to know how to use a certain function I ask the AI and 9 times out of 10 it gives the correct answer.  One minute looking through a poorly written instruction manual is a minute wasted.

[Leo Gura]

6 minutes ago, Jodistrict said:

but the real question is how much more efficient is it making the current programmers.

They have actually conducted tests on this, and they found AI reduces programmer efficiency by something around 20%.

But I'm sure you can debate this and find contrary data.

The problem is you can't trust any of the authorities on this topic. Everyone is getting paid off to hype AI, and the scientists do bad science so you can't trust their AI studies.

Edited 1 hour ago by Leo Gura

[Natasha Tori Maru]

The implication that AI is "intelligent" can "decide" or "reason" is that it must experience emotions and feelings, yes?

The work of Antonio Damasio (not sure if you guys know this neuroscientist), who ran studies on patients with damage to their ventromedial prefrontal cortex, showed they struggled to make simple, everyday decisions. This part of the brain is important for linking emotion to reasoning. The patients had normal IQ's, intact memory and could reason logically. 

Emotion provides value - this is why it is critical. Reason provides "what follows from these premises?", emotion provides "why should I care". They work in tandem. So without emotion, reason has no objective function. Reason is calculating and then the emotion is supplying the goal.

I would argue that humans interaction with AI is supplying the "goal" or emotion (by providing values, preferences, motivations), and the algorithm/calculation is the AI reasoning.

The AI is not generating any feelings/emotions. So a large portion of the humans ground-up reasoning is being outsourced to it. 

A good example of the above in action:

"I'll invest in index funds because they're statistically optimal."

Seems rational. Underneath are hidden emotional/motivational values: I want financial security, I dislike unnecessary risk, I value long term planning, I prefer maximizing returns etc etc. 

The above is also why I consider feelings and emotions to important to all domains of philosophy and consciousness work.

Edited 1 hour ago by Natasha Tori Maru

[Jodistrict]

23 minutes ago, Leo Gura said:

They have actually conducted tests on this, and they found AI reduces programmer efficiency by something around 20%.

But I'm sure you can debate this and find contrary data.

The problem is you can't trust any of the authorities on this topic. Everyone is getting paid off to hype AI, and the scientists do bad science so you can't trust their AI studies.

Chatgpt told you that because it knew that is what you wanted to hear.   

[Leo Gura]

58 minutes ago, Jodistrict said:

Chatgpt told you that because it knew that is what you wanted to hear.   

I don't use AI for text.

[AerisVahnEphelia]

1 hour ago, Leo Gura said:

They have actually conducted tests on this, and they found AI reduces programmer efficiency by something around 20%.

But I'm sure you can debate this and find contrary data.

The problem is you can't trust any of the authorities on this topic. Everyone is getting paid off to hype AI, and the scientists do bad science so you can't trust their AI studies.

AI can do certain things a hundred times faster, but you need practice to understand which things, in which domains, and how to direct it properly.

People see AI as some kind of magical machine that is supposed to produce magical results.

To me, that is as fucking retarded as holding a pen and expecting it to create a masterpiece by itself.

You have to use the pen every day to learn how to draw and understand its limitations.

A good programmer might increase their productivity by 300% because they know when to code manually, when to use AI, which words to use, and which data to provide in order to guide it toward a specific goal.

A bad programmer does not know what to ask or how to ask it.

The study is stupid as fuck.

And geniuses are probably one in a hundred across every domains.

There is a few study done on geniuses/high iq/whatever you want to call a high performer.