me2beats

[me2beats @me2beats1]

For AI, it's more important to distinguish, compare, find connections and patterns, interpret than to calculate.
This means there will be more strings, dictionaries, and graphs than numbers and arrays

[me2beats @me2beats1]

A tool must be shaped by the problem it solves.

This means, for example, that the most used data types and structures must respect the values ​​of the problems being solved.

You can't write an efficient calculator in SKI

[me2beats @me2beats1]

Reflection > Homoiconicity

Human thinking is not homoiconic. Its algorithms do not live as data; they live in processes that we represent as data through other processes

Having tools for analyzing algorithms execution, getting info about what they do and why is more important

[me2beats @me2beats1]

When you create a symbolic system, you also create a programming language.

1. The language must be easily modified by the system – for learning and adaptability.
2. An ability to stop programs – to avoid infinite loops.
3. Code execution analysis tools – for self-optimization

[me2beats @me2beats1]

This combination must be simple.

Yea, the core must be simple.
If the system is complex or uses multiple entities interacting in complex ways, this increases the cognitive barrier for the system, which will inevitably have to understand itself

[me2beats @me2beats1]

This combination must be bold.

It must partially break with the accepted design patterns of symbolic systems. Otherwise, we would have long ago enjoyed a symbolic system capable of solving problems at least as well as the first version of ChatGPT

[me2beats @me2beats1]

But is it worth it then to seek completely unexpected solutions that are almost impossible to predict?
No.
Fundamentally new problems require fundamentally new combinations of fundamentally old solutions.

[me2beats @me2beats1]

Give the most advanced neural network the task of creating a powerful AI that isn't an NN.
It will fail.
But why?
It's a statistical problem. NNs choose the statistically most probable. But there are no statistics on powerful non-NN AIs because such AIs simply don't exist.

[me2beats @me2beats1]

LLMs have great and flexible algorithms, but any algorithm is limited. Except for an algorithm that can change its own algorithm.

So we don't want a new NN.
We want an AI that can "smartly" modify itself. Any other AIs are not about intellect

[me2beats @me2beats1]

The issue is that any system that cannot learn (adaptively change its internal algorithms of interpreting things and get new ways to act) cannot choose the best actions, not only because "best" is highly subjective but also because knowledge is always incomplete

[me2beats @me2beats1]

Spoiler: probably the deepest idea I came up with is that any intellect is just a tool for choosing the best next action.

Sounds familiar? Yeah – LLMs (like Grok/GPT) also try to choose the best next... Token! Tokens encode meanings, meanings can encode actions. So that's wrong?

[me2beats @me2beats1]

My name is Albert, and I've been dreaming about creating my own symbolic AI for about a year.
I have ideas, so I'm going to use this channel to share them.
Disclaimer: These are just ideas at the moment, and many of them may sound chaotic and may be rejected later

[me2beats @me2beats1]

@reduzio It's not a failure. Godot simply can't get many advantages from VS now, in its current form. Making it a plugin first and making it mature is a great decision imo. In our Godot server (Discord, almost 2k members) no one asks any questions about VS. GDScript wins.

[me2beats @me2beats1]

@Sketchfab 50 results per page would be fine

[me2beats @me2beats1]

@Sketchfab you are wonderful. But please make search better — add pages! Maybe optional. 50 results per page wold be

[me2beats @me2beats1]

@reduzio @g_de_gabriel but not low-end mobile :(

[me2beats @me2beats1]

@ThisWeekInGodot @godotengine I've asked it in issues, the project author didn't use Godot https://t.co/Xe0GFSjlPV

[me2beats @me2beats1]

@ThisWeekInGodot @godotengine Hi! Thanks, lots of interesting stuff. But is FreeHL Godot related? I looked at the source code and couldn't find anything like project.godot file etc

[me2beats @me2beats1]

@pycbouh won't that clutter up the main view?
now the menu popup always closes if the mouse is pressed anywhere (inside or outside the menu), I find this fine for small screens

[me2beats @me2beats1]

@pycbouh That's cool but what if you have a smaller screen?:)