Re: [PATCH v2] x86/i8253: fix possible deadlock when turning off the PIT

[Fernando F. Mancera]

On 27/03/2025 23:52, Thomas Gleixner wrote:
> On Thu, Mar 27 2025 at 22:17, Thomas Gleixner wrote:
> > Though I really want people to sit down and think about the factual
> > impact of a tool based problem observation. Tools are good in detecting
> > problems, but they are patently bad in properly analysing them. And no,
> > AI is not going to fix that anytime soon, quite the contrary.
>
> May I recommend you to ask your favorite AI model of the moment the
> following question:

Just in case I gave the wrong impression, I do not like LLMs at all and 
do not use them. They strip out your personality from emails/messages 
and hallucinate a lot.

>   "Explain the discussion in the email thread starting at:
>    https://lore.kernel.org/all/20250327152258.3097-1-ffmancera@xxxxxxxxxx";
>
> I'm sure that I'm patently bad at AI prompt engineering, but that does
> not justify the utter insanities which these models threw back at me.
>
>   ChapGPT:
>
>     "....
>      This patch adds a selftest (automated test) for the nftables
>      flowtable feature using a netdevice. It targets the netfilter
>      subsystem, which is responsible for packet filtering, NAT, and other
>      packet mangling in the Linux kernel."

Here it is hallucinating using content from a different patch I sent a 
while ago. Adding some tests to nftables about flowtables. Probably it 
just spotted my email and started to generate text from it.

>   Grok:
>
>     See
>
>          https://tglx.de/~tglx/grok.html
>     
>     for the full glory of AI hallucinations.

Now, my name is Felipe. Great, good to know. Ah, and I also maintain a 
network driver too, awesome.

> At least those two were halfways reasonable:
>
>   Gemini:
>
>     "API REQUEST ERROR Reason: Unknown."
>
>   Claude:
>
>     "I don't have access to the specific URL you've provided ...
>      my knowledge cutoff was in October 2024 ..."
>
> Seriously?