Leo's Blog Discussion Mega-Thread

[Jannes]

1 hour ago, integral said:

Claude 4 opus has nearly solved codding, its so good i dont write any code anymore, 0.

Do you still need to know how to code to give the proper prompts for AI and check for errors ? 

[ZenRocker]

22 hours ago, Leo Gura said:

I will add 40 books about such topics to my book list soon.

Ive been on the edge of my seat for years waiting on the book list update --- if the bottle neck is recording all the summaries, couldn't you just release the list and then add those as you complete them?

[Leo Gura]

3 minutes ago, ZenRocker said:

Ive been on the edge of my seat for years waiting on the book list update

Sorry about that. But during that time I have collected even more books. So the new list has doubled from last year. This will be my biggest, juiciest update ever.

Edited June 6 by Leo Gura

[integral]

2 hours ago, Jannes said:

Do you still need to know how to code to give the proper prompts for AI and check for errors ? 

No I just explained everything in English to the AI. 

NO: For small apps if you never programmed before you could probably figure it out along the way and get things done and even deploy something.

YES: For larger apps you need to understand the technology language and many software terms, there's so many technologies that you have to use that you never heard of and you have no idea how anything works that it's just going to be very complicated for a complete beginner. 

Edited June 7 by integral

[integral]

2 hours ago, Leo Gura said:

That is a valid point and one I'm concerned about.

However, I don't think we can reach 30%+ unemployment long-term without general intelligence.

As long as AI is not generally intelligent it will be hard to replace humans, even if the AI is very capable at narrow things like coding.

How do you define general intelligence? Because the current chatbots are technically generally intelligent? They just can't use tools or use computer or navigate anything or complete any task on their own. They're basically stuck in a chat box.

Sequential tool usage means it: creates a plan, then it makes a phone call, then it writes an email based on that phone call, then it opens up a spreadsheet and inputs whatever it needs to, then it does some kind of tax document on some kind of website, then it's creates an invoice and emails it, then it stores everything it learned in a database for future use, because it can learn on the job, then it follows the script which tells it to ask itself how it could improve for the next day.

If it could get good at each of those individual tools and tasks and then and if it could run for a week straight without any input needed.

Then I think it basically replaces a lot of people?

Edited June 7 by integral

[Leo Gura]

6 minutes ago, integral said:

Because the current chatbots are technically generally intelligent?

They are not even close.

They create an illusion of general intelligence by piggybacking off the intelligence in the collective symbolic output of mankind.

Memorizing the entire internet is not the same thing as intelligence. But it produces pretty intelligent-looking output.

[Eskilon]

25 minutes ago, Leo Gura said:

Memorizing the entire internet is not the same thing as intelligence

Then how do you get intelligence? If a human could calculate every single chess move, have it stored in it`s brain and become ubeatable - is that intelligence? 

In the case of A.I, if it is not conscious it will always exist within some parameters, it cannot rewrite itself, nor desobey orders. How can it ever be intelligent？

 

Edited June 7 by Eskilon

[Natasha Tori Maru]

4 minutes ago, Eskilon said:

Then how do you get intelligence? If a human could calculate every single chess move, have it stored in it`s brain and become ubeatable - is that intelligence? 

In the case of A.I, if it is not conscious it will always exist within some parameters, it cannot rewrite itself, nor desobey orders. How can it ever be intelligent？

 

Can AI intuite something?

[Leo Gura]

5 minutes ago, Eskilon said:

If a human could calculate every single chess move, have it stored in it`s brain and become ubeatable - is that intelligence? 

Nope.

Intelligence is not a set of rules.

[Eskilon]

11 minutes ago, Natasha Tori Maru said:

Can AI intuite something?

No.

That requires a subjective experience, which an AI does not have.

[Eskilon]

14 minutes ago, Leo Gura said:

Intelligence is not a set of rules.

Then we won't have an intelligent AI in anytime soon I think.  

AI as it is now and even AGI is just a very complex set of rules. You can have AGI and it doesn't need to be conscious.

Edited June 7 by Eskilon

[integral]

58 minutes ago, Leo Gura said:

They are not even close.

They create an illusion of general intelligence by piggybacking off the intelligence in the collective symbolic output of mankind.

Memorizing the entire internet is not the same thing as intelligence. But it produces pretty intelligent-looking output.

Yes but you're also describing the average person and the work they do for 90% of jobs! lmao

The first AI that replaces humans will be a fake general intelligence.

 

Edited June 7 by integral

[Leo Gura]

15 minutes ago, Eskilon said:

Then we won't have an intelligent AI in anytime soon I think.  

I think AGI is possible, but would require a new structural breakthrough, not just a bigger LLM.

[Natasha Tori Maru]

20 minutes ago, Eskilon said:

No.

That requires a subjective experience, which an AI does not have.

I consistently return to intuition regarding the limits of AI.

[Eskilon]

2 hours ago, Natasha Tori Maru said:

I consistently return to intuition regarding the limits of AI.

Yeah, it is one major limit. Right now, AI is purely logical and calculative -- it lacks the mystery of intuition.

Spontaneity and qualia is fundamental to intuition in my view.

From what I currently understand those 2 things can only exist in a conscious system.

Edited June 7 by Eskilon

[Leo Gura]

LLMs are doing intuition. They just lack general intelligence.

[Natasha Tori Maru]

1 hour ago, Leo Gura said:

LLMs are doing intuition. They just lack general intelligence.

How are you defining intuition, as applied here or in general? If there indeed is a difference for you, when reduced for this application

[Rafael Thundercat]

@Leo Gura your last posts are fire,pure fire. No more running around circles, just plain Truth. 

[Vynce]

5 hours ago, Leo Gura said:

I think AGI is possible, but would require a new structural breakthrough, not just a bigger LLM.

Also to engineer AGI you would have to understand how general intelligence fundamentally works. And since we don't even know how human neurons became intelligent, its impossible to copy.

Most likely its gonna be happening like Penicillin, a mysterious byproduct under the right circumstances. 

[zurew]

https://www.instagram.com/reel/DKPu4MppDnj/