<div dir="ltr"><div>I agree that actual code for ciphers must be 100% written by humans, and reviewed by humans,- but for everything else it can be a OK tool. <br></div><br></div><div dir="auto"></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">man. 30. mars 2026, 13:01 skrev Robert J. Hansen via Gnupg-users <<a href="mailto:gnupg-users@gnupg.org" target="_blank">gnupg-users@gnupg.org</a>>:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">> I am a user of tools like Cursor,- and my personal opinion is that LMM <br>
> is not perfect. But for those who cannot program because of <br>
> neurological conditions, it is a valuable tool.<br>
<br>
As a hacker who deals with mental illness, I am massively in favor of <br>
creating a community that is welcoming to people with psychological, <br>
psychiatric, and/or neurological troubles. But I draw the line at <br>
thinking we should lower our professional standards to accommodate these <br>
conditions.<br>
<br>
I do not believe LLMs should be authoring security-sensitive code, ever.<br>
<br>
> If the programming follows programming standards like PEP 8, |rustfmt|, <br>
> clippy etc..<br>
<br>
None of these verify quality code. The version of pwgen that Claude.ai <br>
created passed the normal clippy checks.<br>
<br>
> tested against Wycheproof test vectors, RFC 5639, and BSI specifications <br>
<br>
Having implemented more cryptographic algorithms in my life than I ever <br>
want to think about, "it passes the test vectors" does not create in me <br>
very much faith in the overall quality of the implementation.<br>
<br>
Circa 2008 at USENIX/EVT we gave a round of applause to a team of nerds <br>
who had implemented AES in Java, and given rigorous Floyd-Hoare proofs <br>
of correctness. It took them two and a half years of work.<br>
<br>
I once had to implement a highly reliable Galois counter mode. I yearned <br>
for the sweet release of death.<br>
<br>
> But,- who can write 100% perfect code ?<br>
<br>
It's hard but it's been done. The provably-correct AES implementation <br>
comes to mind, as does the RSRE VIPER processor. The IBM System 360 <br>
minicomputers that controlled the Space Shuttle never suffered a <br>
life-threatening bug, ever: even as _Challenger_ and _Columbia_ came <br>
apart the HAL/S software stack continued functioning correctly.<br>
<br>
<a href="https://en.wikipedia.org/wiki/VIPER_microprocessor" rel="noreferrer noreferrer" target="_blank">https://en.wikipedia.org/wiki/VIPER_microprocessor</a><br>
<a href="https://en.wikipedia.org/wiki/HAL/S" rel="noreferrer noreferrer" target="_blank">https://en.wikipedia.org/wiki/HAL/S</a><br>
<br>
_______________________________________________<br>
Gnupg-users mailing list<br>
<a href="mailto:Gnupg-users@gnupg.org" rel="noreferrer" target="_blank">Gnupg-users@gnupg.org</a><br>
<a href="https://lists.gnupg.org/mailman/listinfo/gnupg-users" rel="noreferrer noreferrer" target="_blank">https://lists.gnupg.org/mailman/listinfo/gnupg-users</a><br>
</blockquote></div>