Codeium AI & The Future of Programming

[xAction]

I guess this is somewhere between on topic and off topic.
Codieum AI is a free code completion AI supporting 70 languages.
They provide an extension to multiple IDE's

One of the core functionalities of Codeium is autocomplete: on every keystroke you make, Codeium tries to complete what you’re typing. Through looking at the current file, past edits, and context from relevant snippets within your codebase, it generates various plausible suggestions as “ghost text.”

I have well over ten thousands lines of xTalk code, and every time I have to re-use a function or handler I have to dig up an old file to copy/paste it or at the least find it somewhere in the current file with the find command and/or use the function index to go to the function and handler and find out what arguments it takes. So very time consuming. Been meaning to make a big old github database of everything useful, but...that's whole job and I'm not 100% sure how I should be naming things so other people can figure out what they do. The fact that xTalk uses like 90% of the English language as a keyword makes naming things tough...even as it makes other things easy, for instance "Player"...they couldn't call that MediaController?

Anyway, an LLM (large language model) that knows your code base is invaluable.

I absolutely suggest you download this for VS Code and go through the tutorial. Then take a look at Richmond's efforts and even tPerry2x assistance and my attempts in tutorial making. Notable efforts and I was thinking of pumping out applications in the same vain, but there's clearly a new paradigm to embrace.

Right now I have the tutorial open, the Codeium side bar explaning how to use the extension on the left, the tutorial running continously with video (gif?) in the center and the superfly syntax colored code window on the right, all on one screen, no shuffling off to a third party anything. And I'm learning Typescript. Here's a shrunk screen to see what I mean.

codiemtutorialexample.png (50.22 KiB) Viewed 521 times

There's more to say, but I want to do this tutorial without distraction, brb.

[xAction]

Well, that was quick. One thing I can say for sure is that GIFs for tutorial are terrible esepcially if you are looking left at gif, then right at your code and back and forth as the damn gif changes. Maybe some people have powerful visual memory and it just clicks. I'd go for a scripted, click to step through method. Seems weird to me they are explaining text based stuff with images...they already typed the damn thing and they are programmers!

Anyway on to the next relevant news.
Nvidia CEO Jensen Huang's message to kids: generative AI means you don't need to learn coding
"Everybody in the world is now a programmer. This is the miracle of artificial intelligence"

You know what this news is like?
Holding Hill #249 for decades in a battle of dumb versus dumber, and you're almost out of the trenches and going to charge the enemy in hand to hand combat and now you get the news "Wars over, we are friends with the people on the hill, come to the meet and greet, free beer."

I feel like I won, but I was really enjoying the fight and liked holding this muddy hill.

Anyway NVIDIA stock is trading at $791 a share. Jensen Huang is no dummy and he's writing the words on the wall in 8K raytraced AI generated interactive 3D.

In other news: Google's GENIE AI makes 2D platformers from a single image

Meanwhile: Unity Software Stock Tumbles. Wall Street Is Split Over Its Reset Plan.
Unity is imploding, massive rounds of lay-offs and after massive rounds of lay-offs, selling off or just shuttering whole multi-billion dollar divisions. Looks like John Riccitiello (formerly of Electronic Arts) set the company up to fail, sold over 50,000 shares and ran off with the bag. But most importantly, in a world of AI where "everyone is a programmer" the need for Unity and all it offers will likely be a memory in 5 years. Anyone with an internet connection can fire up Google or Bing and say "Give me an app/game that does x,y,z....make it more kid friendly/sci-fi/1920s/whatever"

Unity could work like that but the AI companies are beating them at their own game. Ironic.

Five years may be a stretch, it might be 10 months.

Yeah I don't know what to do with this information...it's just piling up. I figured discussing it could maybe start a web effect: you know this, something relevant comes along and makes a connection, eventually an xTalk AI solution is available anywhere anytime. I'm sure that was the dream behind someone's crowdfunding attempt. It's a good dream, that will happen...it might not look like xTalk from those other sources, but the end result will be exactly what we were all after since the 1980s (1970s?)

Interesting times ahead.

[richmond62]

The future of programming is what it was in 1970: something that every prediction will get wrong.

I would be wary of building A.I. into the IDE.

[tperry2x]

Wouldn't it just be nice to be able to say to ChatGPT (probably ChatGPT 9):

Look at this program I've just uploaded. (OpenXTalk). Look at every option the program has, see and understand the xTalk transcript language fully, and incorporate the dictionary and all the interpreter strings fully into a new IDE. This new IDE should be fully open source, so externalise all code where possible.

Also, optimise the code so that it runs as quick as it can, while updating and adding comments to all script written in transcript within the IDE.

This IDE contains references to a company that has since ceased to exist (Livecode), and as it's being fully reinterpreted, remove all references to Livecode and replace with xTalk. Rename all files mentioning the word 'Livecode' without breaking anything in the IDE or anywhere.

Now, add in native GPU support for a list of video cards and drivers going back at least 10 years until current.

Now correct various bugs in the video player implementation for Linux and the browser widget for Linux. This must also support https, tls, sFTP, 0Auth and MDToken authentication protocols.

Now give me full HTML5 export capabilities in the Standalone Builder, and fix the standalone builder so it can compile for Android 11+ and all versions of iOS above 8.

Also produce an Intel x86 build of the IDE for Linux, MacOS, and Windows.
Now produce an Arm based build of the IDE for MacOS and Raspberry pi respectively.
Also, create an instance of an AI agent that understands transcript. This AI agent should be able to compile executable binaries to multiple operating systems. Download a local copy of this AI agent to my computer so I no longer need internet access to use it.

Now give me a download link for the IDE I mentioned above, and upload everything we've talked about to my github account please.

This, it cannot do. yet.

But when the AI can do all of this, there's no point in using xTalk (or any IDE for that matter).
You don't even need a keyboard and a mouse on your computer anymore. Just an internet connection and a microphone.
My job as IT support will certainly be toast, and schools (and certainly anything involving teaching IT) will no longer exist in their current format.

[tperry2x]

Or, one step further:
Using the brain interface headset I'm wearing currently, create what I'm currently thinking about on my computer.

Or, even more scarily:
Upload a copy of my consciousness into this machine.

(yeah, that's probably a bit too off-topic)

[xAction]

Uploading consciousness isn't scary at all. It'll work just like LLM and Stable Diffusion do.
I was going through a deceased persons code with ChatGPT and I could 'feel' that person thinking/speaking/coding.
With enough data the illusion could go beyond an inner construct to a 3D holographic avatar.

Help me Obi-Wan, comment this code block.

[tperry2x]

It only becomes scary when you factor in the individual. This is why the idea of a 'hive mind', where the collective consciousness of everyone is imalgamated, requires some form of moderation. But who is 'qualified' (for want of a better word) to moderate it? Who's to say what is 'normal'.
However / whoever / whatever does the moderation undesirable elements should be filtered out. Exactly like training an ai dataset. You want to exclude certain things, to prevent history repeating such as what happened with Microsoft's Tay AI.

The other point to this I was just thinking about. If it were possible to upload consciousness, how would you then 'feel' (would you feel anymore?) about your consciousness being edited without your consent?

[tperry2x]

Returning to the original focus, before I drifted off into the future of ai speculation (I'm very prone to 'topic drift' - just like in real life), I'd at least say to an ai:

"Solve the macOS codesigning problems for me in OXT Lite"

Actually, just had an idea about that
https://www.openxtalk.org/forum/viewtop ... 7057#p7057

[xAction]

Now there's Devin the first AI software engineer

You can request access here, maybe it'll help open up the builder language to the masses (or the minutia)

[tperry2x]

On that page, a little further down:

Devin can address bugs and feature requests in open source repositories.

So, I wonder what would happen if someone was to point it toward the livecode github page?

[richmond62]

So, I wonder what would happen if someone was to point it toward the livecode github page?

Why don't you do that anyway?

Notice "the people we know and love" have NOT bothered to acknowledge my last 'love note' about Pre-Built Binaries.

[richmond62]

Codieum AI is a free code completion AI supporting 70 languages.

Is that a fact?

As it only lists 21 IDEs on it page: https://codeium.com/download

and it seems its plug-in only works with those IDEs . . .

How the "huggery-b*ggery" do you get it to work with a language that does NOT use any of those IDEs?

[tperry2x]

So, I wonder what would happen if someone was to point it toward the livecode github page?

Why don't you do that anyway?

Because as interesting a tool as it is, it's as of yet unfinished. It's one to keep an eye on.

I would be very surprised if we were ever likely to hear anything further on this. I'd love to be proven wrong though.

[OpenXTalkPaul]

Codieum AI is a free code completion AI supporting 70 languages.

Is that a fact?

As it only lists 21 IDEs on it page: https://codeium.com/download

and it seems its plug-in only works with those IDEs . . .

How the "huggery-b*ggery" do you get it to work with a language that does NOT use any of those IDEs?

That list is codium add ons for various apps. It lists Chrome in there which is not an IDE.

A lot of those, like Visual Studio, support multiple languages and more via language add-ons. There is an add-on for Visual Studio that adds support for LCS, LC (Server) and Extension Builder syntax.

[xAction]

Go to the request page an pile on the request. I'm sure we're buried under a billion requests but if any one gets approved that's better than none.

[OpenXTalkPaul]

Go to the request page an pile on the request. I'm sure we're buried under a billion requests but if any one gets approved that's better than none.

Along with spell checking, full Code-completion is a feature on my own to do list for OpenXTalk DPE.
It's not entirely clear how you can add a completion provider to the Script Editor for the hooks that are already there for adding new ones. That is, there's a mechanism in the Script Editor scripts that allowed the LC commercial version to have more full code completion while the community edition has only basic completion (some control structures, handler names, quotes, parens, etc.). I would be happy if it could just pull in words from our syntax dictionary and from words already in the script that is being edited (which is like the way Pulsar's code completion works).

A while back I started looking into adding these things, and I made some additions to Trevor DeVore's Hunspell/NSSpell wrapper library here: https://github.com/PaulMcClernan/lc-hunspell
Specifically here: https://github.com/PaulMcClernan/lc-hun ... ries/xTalk
That's an attempt at making a Spelling Dictionary with nothing but xTalk syntax, which was created with the stack in that Repo called "XTALK_Wordlist.livecode"

[xAction]

It goes way beyond simple spellcheck and command completion.
Command Completion is what we have:
type
on MouseUp,
-- hit return
-- get
end MouseUp

Code completeion is :
type
"-- a function that plots the circumference of a circle using x,y coordinates"

hit control return and the whole thing just spits right out:

Code: Select all

on drawCircle centerX, centerY, radius
   lock screen
   -- Iterate over angles from 0 to 2π (in radians)
   repeat with t = 0 to (2 * pi) step pi / 180
      -- Calculate x and y coordinates
      put radius * cos(t) into x
      put radius * sin(t) into y
      
      -- Shift the coordinates to center the circle
      add centerX to x
      add centerY to y
      
      put x*100 & comma & y*100 & cr after tCirclePoints
      --draw point x, y in graphic "MyCanvas"
   end repeat
   --put tCirclePoints
   set the points of the templateGraphic to tCirclePoints
   set the style of the templateGraphic to "polygon"
   if there is a graphic "temp" then delete graphic "temp"
   set the showName of the templateGraphic to true
   create graphic "temp"

   set the width of graphic "temp" to width of this stack -10
   set the height of graphic "temp" to height of this stack -10
   set the loc of graphic "temp" to width of this stack/2, height of this stack /2
   set the label of graphic 1 to  Radius
   set the layer of graphic 1 to bottom
   unlock screen
end drawCircle

Without you ever having to program it yourself.

That's what codieum was like. Like google finishing your search before you finishing typing it.

If we can get this thing talking builder language that'd be a big boost.
Anything to cut down the learning curve on that.

[OpenXTalkPaul]

Code completeion is :
type
"-- a function that plots the circumference of a circle using x,y coordinates"

Correct, in the community build we have only handler name, and brackets/parens/quotes completion in the script editor.
However, in the message box we DO have some form of code completion for any line that exists in the message box's cached scripts (which are stored in revPrefs file as an array). That seems a lot like the way Pulsar does code-line completion. Of course that's not going to give you multi-line code from AI.

I've used ChatGPT to help me write Extension Builder code, it writes it fairly well after teaching the language to it for a bit (maybe as well as it does other programming languages, which is of course not perfect).

[tperry2x]

I've used ChatGPT to help me write Extension Builder code, it writes it fairly well after teaching the language to it for a bit (maybe as well as it does other programming languages, which is of course not perfect).

Yes, I too have found it to be far from perfect.
I wondered if there was any way to include code completion in an offline way - a bit like how sublime text pops up suggestions while you are typing:

autocomp.png (40.95 KiB) Viewed 201 times

Actually, you've just reminded me - I created something like this a long time ago:
(link here)

Screen Shot.png (43.62 KiB) Viewed 201 times

[OpenXTalkPaul]

I've used ChatGPT to help me write Extension Builder code, it writes it fairly well after teaching the language to it for a bit (maybe as well as it does other programming languages, which is of course not perfect).

Yes, I too have found it to be far from perfect.
I wondered if there was any way to include code completion in an offline way - a bit like how sublime text pops up suggestions while you are typing:
autocomp.png

Actually, you've just reminded me - I created something like this a long time ago:
(link here)
Screen Shot.png

The message box does completion based on previous input in its cache (which is stored as Array in the revPreferences), the auto complete code in the single-line message box behavior script.