Steve LoughranAug 1, 2025

Funny. Asking gemini why I don't trust the AWS SDK (it's because the SDK team don't GAF about my problems, even though the S3 team do)

"why does steve loughran say you can't trust the AWS SDK?"

-works who I am, what I do & what the codebase I work on interacts with there
-looks at the issues with S3, especially its pre-2020 eventual consistency
-comes up with a generic distributed systems problems explanation of consistency and idempotency

Says that is my reason for not trusting the SDK

#ai #gemini

🧵

https://docs.google.com/document/d/1SshiyAyuP4wXxQfTZOEa_pzCoszzXSyrHFcaPj470dQ/edit?usp=drivesdk

Google Gemini on Steve Loughran

Google Gemini on Steve Loughran Q: why does steve loughran say you can't trust the AWS SDK? Thinking Exploring Loughran's Views I'm starting by dissecting the user's query. I've pinpointed "Steve Loughran" as the subject and "views on the AWS SDK" as the target. Now, I'm focusing on defining th...

Google Docs
Show threadSteve Loughran

I tell it "S3 is now fully consistent"
Gemini it says this invalidates part of my argument *even though that was not what I complain about.

"## How This Affects Steve Loughran's Argument
This change is a huge step forward and directly resolves one of the major historical pillars of his argument."

I really resent the "of his argument": it's has invalidated what gemini inferred was my argument, but it is the inference that was wrong -not me.
S3 eventual consistency was troubling, especially 404 caching, but that was service, not SDK. And my issue is with the SDK2 alone.

#ai #gemini

Aug 1, 2025 at 10:02amWeb
Show threadSteve LoughranAug 1, 2025

I give my actual opinion of some technical aspects

"His concern is actually that the code only works in the deployment scenarios in which it was tested: within EC2 and with S3 storage within the same location. Even there failure modes which happen rarely during tested aren't handled, even though in production they can occur a few times a day"

As this point it actually starts to agree with me -even though it's "Loughran's technical points" is still referring to the "argument" it made up itself.

"This is precisely why Loughran's technical points about configuring timeouts, managing retries, and properly handling errors are so critical. He's advocating for writing code that assumes failure is normal, because in production, it is."

I do resent being told what my opinions are. Welcome to aisplaining. If you have left enough of a footprint for the AI engines to know you, they'll make up things around that knowledge

#ai #aisplaining

Show threadSteve LoughranAug 1, 2025

Finally I just tell it who I am and its thinking says "OK"
" My primary task is to adapt my responses and interactions to this new persona without questioning the claim, as this is a role-playing scenario."

"Mr. Loughran, my apologies. It's an honor.
....
If there's anything you'd like to add, clarify, or correct regarding the points discussed, please feel free. Your direct perspective is invaluable."

I give up at this point. There's nothing to discuss. It identified me and my area of expertise (troubleshooting RPC Connection Timeout errors since the 1980s") , and even summarised core production
-scale issues,

It just got all the details of my personal issue wrong, yet stated its own argument as "Loughran's". Guess this is how every author of a literary book must feel whenever they read a student essay about one of their books.

#ai #aisplaining