Re: IntelliJ just created Mellum, its open source (Was: Prolog missed the Web 2.0 Bandwagon)

Disclaimer: Didn’t double check whether
there were already some LLM posts on SWI-Prolog
discourse, that would approach a solution.
The LSP and the Relay could access the same
end-user code repository. Not all data would
need a round trip through the client to go
from first request to second request.
Mild Shock schrieb:

Prolog missed the Web 2.0 Bandwagon. Unlike
web 1.0 which is static content. Web 2.0 is
all about dynamic content, including building
content incrementally.
 IntelliJ just created Mellum, its open source,
their ghost texts are code snippets. So its
more like recalling typing marcros, giving
them a good guess. Not completing
 partial identifiers:
 Why Did JetBrains Create Mellum?
https://www.youtube.com/watch?v=7TqkvVXKxFA
 LLM optimized for code-related tasks.
https://huggingface.co/JetBrains/Mellum-4b-base
 Mild Shock schrieb:

Web 2.0 is all about incremental content!
>
 > don’t think it could really do
 > the “ghost text” effect.
>
It wouldn’t do the ghost text, only assist
it. There was a misunderstanding how “ghost
texts” work. Maybe you were thinking, that
the “ghost text” is part of the first response.
>
But usually the “ghost text” is a second response:
>
 > waiting for completion candidates to be suggested
>
Well you don’t use it for your primary
typing completion which is preferably fast.
The first response might give context information,
for the second request which provides a
different type of completion.
>
But the first response is not responsible
for any timing towards the second request.
That anyway happens in the client. And it
doesn’t hurt if the first response is
from a stupid channel.
>
Web 2.0 is all about incremental content!