H: Claude spec's air-cooled endrow
Me: Middle
H: Claude says end
Me:
H: Not end? Claude wrong?
Me: Claude is thinking 100m of Cat6. Did you tell it 800G and just 3m?
H: Oh, middle cabinet. But why, again?
:swearing:
H: Claude spec's air-cooled endrow
Me: Middle
H: Claude says end
Me:
H: Not end? Claude wrong?
Me: Claude is thinking 100m of Cat6. Did you tell it 800G and just 3m?
H: Oh, middle cabinet. But why, again?
:swearing:
H: so we have nine B300
Me: you have 89, add the rows.
H: we do? and is there a bottleneck?
Me: 64 GPUs of capacity for your 712 GPU? yeah, more like a bottle without a neck.
H: oh. so what do data centers usually do?
🤣
I got hauled into a data center today because Claude was used to spec it and screwed up badly. At least 2,500 GPUs were in an impossible upside down and backwards plan.
😱
I had to pause the call because I couldn't stop laughing. AI had sabotaged the AI datacenter spec for building more AI.
you didn't see the announcement?
RE: https://cyberplace.social/@Nadsec/117407640965290595
The Mythos verdict hasn't changed since April, I see.
Such a collosal waste of CISO time and money. The world could have done much, much more valuable post quantum work.
Not only did Mythos make people stop budgeting and talking about real quantum threats, a huge mistake in itself, but AI/GPU resources have broken classical encryption faster. So we saw companies work less on a disaster coming sooner.
"An effigy of a woman on a bike was mutilated during the 1897 Cambridge protest"
That bike is very significant, seen as empowering, liberating, enabling women. The tech meant women could escape patriarchal limits, like ride to work and make wider social connections.
2012 BSides: Big Data's Fourth V
https://www.flyingpenguin.com/2012-bsideslv-presentation-big-datas-fourth-v/
2016 BSides: Great Disasters of Machine Learning
https://www.flyingpenguin.com/2016-bsideslv-ground-truth-keynote-great-disasters-of-machine-learning/
RE: https://mastodon.social/@nixCraft/117369225190039735
some of us have been saying this for longer than OpenAI and Anthropic have existed.
In 2012 I called it the Loch Ness problem. In 2016 I said Tesla AI would kill a lot of people because of such poor design.
we will never find the AI monster, as the whole game is selling a mythological fear while failing to do the basics.
RE: https://tech.lgbt/@Natasha_Jay/117352985195451871
instead of potty mouth it should really be the Toiletic method
these stickers were so popular at German OWASP Day I quickly ran out.
printing more. maybe I'll put an order form up, or feel free to make your own.
subtext: neighboring states had cracked the German Enigma by the late 1920s. the rapid threat of Nazis included them being very bad at technology they acquired and ran poorly.
I'm so unbelievably tired of Zuckerberg shit
Meta Petal Undersea Cable Colonizes France https://www.flyingpenguin.com/meta-petal-undersea-cable-colonizes-france/
AfD “Remigration” Defined by German Courts as Ethnic Cleansing https://www.flyingpenguin.com/afd-remigration-defined-by-german-courts-as-ethnic-cleansing/
OpenAI Fails History at BlueHat: Decade of AI Deaths Erased For What? https://www.flyingpenguin.com/openai-fails-history-at-bluehat-decade-of-ai-deaths-erased-for-what/
And I called it out in 2016 as a Titanic level problem coming if we didn't immediately turn the ML ship.
BSides Las Vegas ran 2 to 3 August 2016, five weeks after Tesla disclosed the Brown death (the second Autopilot fatality of 2016; the first, in China, surfaced only that September) and eleven weeks before Musk announced every car shipped with full self-driving hardware.
My talk was between the second Autopilot fatality and Musk's driverless promise of ML.
You see?
The exact problem was named in public, at a major security conference, a decade before this OpenAI laundry deck says the "surprising" leaps began.
RE: https://mastodon.social/@HalvarFlake/117284297206579258
Our industry has a vendor story telling problem. We did NOT stumble into stochastic ML in production in 2023. It went on public roads in 2015, people immediately died, and the crashes kept climbing (July 2026 set a new record of Autopilot/FSD crash reports to NHTSA, four fatal). Note: the AI vendor story has always been that things improve with volume over time. A decade of volume and the crash reports are worse than ever.
This talk dates the problem just to "the last 2-3 years" and files the decade before as "inside the expected trajectories." Wow. Nope. A decade of dead people is being suspiciously erased from an OpenAI view of history.
RE: https://mastodon.social/@VeroniqueB99/117265999780139765
American 1960s (Nixon) decoder ring:
- half-hour drive = "interstate" highway to turn black neighborhoods into rubble and dump pollutants on those who remain
- no sidewalks = criminalize young blacks
- no bike infra = criminalize young women
loneliness, nativist isolation, is the white nationalist platform. anti-connected, anti-cosmopolitan, the racists call distant weak bonds "supremacy".
RE: https://hachyderm.io/@mekkaokereke/117258973124308644
Truth. This reminds me of Olivia Sage and how zero Americans study who she was and what she did with her money.
https://www.flyingpenguin.com/charity-spend-doesnt-compensate-for-billionaire-misconduct/
Also Craig Newmark