[fpc-devel] [RFC] Modernising the FPC Release Process -- Proposal for Review

[robert rozee]

> Sent: Thursday, April 16, 2026 at 9:08 PM
> From: "Michael Van Canneyt via fpc-devel" <fpc-devel at lists.freepascal.org>
>
> Back in 2022 that was correct reasoning. Today [2026] I will ask Claude to fix it.
> This will seriously reduce the need to wait.
>

how certain can you be that Claude has not used source code it found on
Microsoft's (or IBM's) website to "fix it"? or to be more precise, how sure can
you be that of ALL the material used to train Claude, absolute NONE of it came
from a copyrighted source? and if Claude was trained by lesser AIs, as now
seems to be the fashion, how sure can you be that those lesser AIs were trained
on 100% copyright free source material? and in the case of AI begets AI, begets
AI, begets AI (ad infinitum) how far back should you go? there is a school of
thought that it is turtles all the way down!

to be safe, you would need to ask Claude to provide source references for any
solution it gives you, and then extensively verify that the those sources are
(a) unencumbered by copyright, and (b) sufficient for a 'reasonable programmer'
to have arrived at Claude's solution independently. in which case it may be
more economic to simply employ said 'reasonable programmer' in the first place
and do without Claude muddying the waters.

i'm not wanting to be negative, i'm just saying that Claude has no sense of
morality, honesty, or indeed any understanding of, well, ANYTHING. it just
blindly predicts the next word to emit - and is rather good at it!


cheers,
rob   :-)