munumblepants retweeted
The Department of Homeland Security is requiring all civilian federal agencies to remove DeepSeek products and 'any third-party services developed using DeepSeek implementations,' and to stop buying them.
munumblepants retweeted
We raised $1.5B to deploy Galaxy globally.
Thanks to our lead investors Trace Capital and Innovation X for backing that growth.
More to come: tenstorrent.com/newsroom/ten…
munumblepants retweeted
It's particularly interesting to me that TypeSafe AI tried to coin the term System One Models but no one uses it and instead Jev is being called a "decision model"
munumblepants retweeted
how much can models remember?
there’s a neat perspective on neural networks that deserves to be more widely known
here’s how you could have invented every layer of the transformer from memory
(1/n) from my new blog post!
The market is suddenly flooded with V100 32GB cards—plenty of stock, no need to scramble, with both SXM2 and PCIe versions available. The bad news? Prices are up at least $200. The good news? $2,000 can still build you a solid personal local LLM rig, and performance is genuinely decent—thanks to Strata’s high-performance algorithms.
munumblepants retweeted
We are bringing agents to the wet-lab bench at Google Research to navigate complex biological landscapes. Learn how we are grounding agents in physical lab robotics to run fast closed-loop molecular design cycles, ultimately discovering single-digit nanomolar protein binders for cancer targets →goo.gle/4B5mMMB
Why aren't their faces shown? All the "criminals" in those fake police videos never opt for face blur. What gives Masonic turds?
munumblepants retweeted
El famoso video de Will Smith comiendo espagueti se convirtió en una curiosa forma de observar el avance de la inteligencia artificial. Lo que en 2023 parecía una pesadilla digital, hoy puede recrearse con un realismo sorprendente. Así ha evolucionado la IA en apenas tres años.
munumblepants retweeted
"I haven't changed my opinion." Sean Connery to Barbara Walters, 1987. #DeadWrongHistory
munumblepants retweeted
Caffeine cuts blood flow to the brain by 30%.
Psychiatrist Daniel Amen says anything that lowers blood flow prematurely ages the organ.
He founded Amen Clinics, which has performed over 200,000 brain SPECT scans.
In his scans, low blood flow is linked to depression, ADHD, substance use problems and traumatic brain injury.
He says it’s the top imaging predictor of Alzheimer’s.
On SPECT, he says, caffeine constricts blood flow by 30%.
The effect is large enough that he has patients avoid caffeine the morning of a scan. Otherwise it artificially lowers the reading.
Blood flow is not his only reason.
Caffeine blocks adenosine and disrupts sleep.
It is addictive and causes withdrawal.
It raises cortisol.
His position: caffeine is best limited or avoided for optimal brain function.
And the patients who quit tell him the same thing:
“So many of my patients stop, and uniformly they tell me they feel better. They said their energy is better.”
One of them was Jeff, a pilot with brain fog and memory problems.
Jeff was having around 600 mg of caffeine per day (less than two Venti brewed coffees from Starbucks, at ~330 mg each).
His brain looked terrible.
They cut his caffeine by 5% a day. In three weeks it was gone, with no withdrawal and no headaches.
Then the texts from Jeff came in:
“Unbelievable energy. Unbelievable clarity.”
People treat caffeine addiction as one of life’s little miracles, with no cost.
Amen’s scans tell a different story.
Readers added context they thought people might want to know
Acute caffeine reduces cerebral blood flow by ~27% in studies, but large prospective cohorts associate moderate caffeinated coffee/tea intake with lower dementia risk and caffeine is in a Phase 3 trial for Alzheimer's cognition.
pmc.ncbi.nlm.nih.gov/articles/PMC27…
jamanetwork.com/journals/jama/…
clinicaltrials.gov/study/NCT04570…
- 04:19:1F:34:3B:13 ← looks like a plausible globally assigned unicast MAC
- first byte 0x04 → binary 00000100
- LSB (I/G) = 0 → unicast (not multicast)
- next LSB (U/L) = 0 → globally administered (not locally-random)
- Reversed 13:3B:34:1F:19:04
munumblepants retweeted
In case you’re interested in wargames about the cognitive terrain
Here actions influence the ownership and stability of cognitive nodes. Their effectiveness depends on network connectivity, audience engagement, existing defences and some randomness
One of the things I like best is that reach does not equal effect, since a InfoOp reaching millions of people is not necessarily successful unless it affects the adversary's decisions
kw-wargame.github.io/cogniti…























