Something to ponder about Machine Learning...

[JackKirk]

I am in the middle of a quite intense discussion with my ML model of choice (Copilot) making some quite fancy modifications to a big chunk of HTML and javascript we co-wrote a while ago.

As usual I have to do it in little increments, one step at a time....

It just occurred to me that these tools are not going to be writing complete apps on any platform anytime soon.

The reason is what they have access to - they basically scrape the web - Github, B4X forums etc.

This content is all basically titbits - not complete apps.

If they don't have access to the code of lots of full function apps they can not mimic it.

So they will remain real clever titbit masters.

Until they work out how to disassemble stuff

 

[aeric]

You mean LLM (large language model)? I think it is not accurate to call it machine learning but learned model.
Well, If you give it a working code, ask it to refactor it, it will give you back a broken code that doesn't work.
My experience is, it worked first time but subsequently, most of the time, it suddenly doesn't know to do it anymore. Hallucinating all the way.

 

[JackKirk]

My experience is, it worked first time but subsequently, most of the time, it suddenly doesn't know to do it anymore. Hallucinating all the way.

Not my experience

 

[rabbitBUSH]

Until they work out how to disassemble stuff

Right - or - until they "learn"/are taught the skill of Analyst Programmer - that being a/the major part of specifying a rounded Code+Input+UI+Output application.

Incidentally, Turing, Von Neumann and contemporaries discussed a lot of this (the prospects of) in the '40s. That's roughly a 90 year track.....

 

[LucaMs]

https://www.b4x.com/android/forum/t...liable-as-if-we-didnt-know.155226/post-961180