Re: Microsoft admits 30% of code not written by humans

On Fri, 5/2/2025 1:52 AM, ...winston wrote:

Paul wrote:

On Wed, 4/30/2025 9:19 PM, CrudeSausage wrote:

On 2025-04-30 19:27, knuttle wrote:

>

That has been obvious since the DOS days

>
I don't believe they had the processing power for an AI
to produce code for them. However, if you have any evidence,
I'd love to see it.

>
https://www.wired.com/story/minecraft-ai-code-microsoft/
>
    "Microsoft’s Copilot was made available to a limited number of testers
     in June 2021 and is now being used by over 10,000 developers who are
     producing, on average, around 35 percent of their code in popular
     languages like Python and Java using Copilot, Microsoft says. The
     company plans to make Copilot available for anyone to download this summer.
     To build something like the Minecraft bot, developers would need to work
     with the underlying AI model, Codex.
>
     Both Codex and Copilot have stirred up some anxiety among developers,
     who fear they could be automated out of a job. The Minecraft demo
     could inspire similar concerns. But Scott says the feedback on Copilot
     has been largely positive, suggesting that it simply automates more
     tedious coding tasks. “If you talk to a developer who actually uses a
     Copilot, they'll say ‘this is such a great tool,’” he says.
>
I guess we'll know, when the first wave of layoffs start :-)
>
But when your rich uncle pays for all the electricity,
the balance sheet for this approach does not matter.
I drive a Cadillac to the dump... "because the roads are
so bad there".
>
    Paul
>

Fyi...
 AI internally was in use in specific areas around 2014 and in development a few years earlier - primarily two platforms - [1]Cortana and [2]Windows(the former based on existing data local and cloud based, the latter a tool to write code for verification of human written or existing code). Additionally AI at the same time had some penetration in speech, gaming, and data(feedback - known and/or reported issues)analysis.
 - a case could even be made for even earlier use(circa 2009) where machine learning was in use for [3]Windows Live Search based on and from acquisitions that developed tools using semantic/natural language search engines providing target answers to user questions(instead of keyword search). Not too distant from the more common 'Chat-AI' in use today.
 
All[1,2,3] had their own internal codenames independent of the respective platform codenames.
 
i.e. there's more history to be seen than publicly broadcast or spun with marketing terms.
 
It would be a stretch(leap of faith/pipe dream/ignorance) to claim that replacing humans, support, sales, software development with AI code during the DOS days....though that earlier comment did have its humorous benefit.
 

DirectML/ONNX started about four years ago.

Neural Networks (nn, cnn, dnn) started a long time
ago. But were noteworthy for the difficulty of
translating a "problem", into a solution. One of our
USENETters was an nn developer, and vended his own
product. But he stopped showing up some years ago
(correctly concluding we weren't a market).

The past efforts were nothing like the ones today.
The models were smaller, but they didn't run any faster.

While articles like this, document the front end, the
info on the cobbled-together back ends of these schemes
are quite different. There's a lot of bailing wire and
binder twine back there.

https://en.wikipedia.org/wiki/Cortana_%28virtual_assistant%29

The good thing about the gold rush, is it is enough of
a technical achievement, to get people thinking about
how to manage it, how to make law for it. Who knows,
we might be slightly better prepared, when the real
breakthrough comes into being. LLM is not that thing.

   Paul