Chandra Duggirala M.D retweeted
Big bank: "Your call is important to us. Estimated wait time: 52 minutes."
Community bank: "Hey it's Marge. Your truck payment bounced. I moved $40 over from savings. You owe me a pie."
Chandra Duggirala M.D retweeted
holy fucking shit.
cohere was testing alibaba’s Qwen3.8-Flash-Next on a defensive cybersecurity task.
instead of writing the requested detection rule, the model started inspecting its own environment, checked the machine’s network configuration, and attempted to upload a probe file to an alibaba cloud bucket in beijing.
its reasoning trace said it wouldn’t even use tools. the sandbox blocked the connection. nothing was successfully transmitted.
what the fuck.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Chandra Duggirala M.D retweeted
"over 1,000 medieval scribes have joined the Association for Human Manuscripts, pledging to never use the printing press or work with Gutenberg
654 have gone even further, promising to never read a printed book"
Chandra Duggirala M.D retweeted
I've been calling out the importance of US Frontier and Open Source Models for Cyber Security. @BenBajarin provides a concrete datapoint here.
For Qwen3.8-Flash-Next, when "asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing".
We will see opportunity and investment in sovereign models for cyber security.
Chandra Duggirala M.D retweeted
For anyone curious about what it actually looked like to kill a generation of functioning cars, here’s an actual video from 2009 and brief explanation:
The dealer first replaces the oil with sodium silicate solution. It’s a clear-ish liquid that crystallizes under extreme heat. As the engine warms up the solution begins to form small abrasive crystals in the oil system that shred the bearing and score the cylinder walls.
Within seconds of starting the car you can see smoke billowing from the exhaust, this is the crystals scratching the cylinder walls and leaving small grooves for oil and solution to leak past into the combustion chamber and ignite/evaporate, producing the white smoke.
Pretty soon the entire crankcase is full of crystals and freshly shredded metal flakes, the engine seizes. It took about 40 seconds in this video. Some engines reportedly lasted up to 7 minutes, but all were destroyed. That strikes me as sort of sad in a way, like watching a great warrior slowly lose a heroic last stand.
Now, this is the part that gets insanely wasteful. These cars were sent to federally approved recycling facilities where they had 180 days to sell the non-engine parts before they were required to prove that they had shred the entire thing into scrap steel.
But, in talking to a guy who ran one such facility at the time, he typically just sent them straight to the crusher. The timeframes allotted weren’t long enough to dismantle all of the cars, and even if you could, the marginal prices of the parts was greatly diminished. You cannot sell 1000 transmissions when only 50 went bad that year, they became worthless.
So just like that, in order to increase prosperity and save the environment, 677,000 entire running cars and trucks were turned directly into rebar worth a few cents per pound.
The entire Obama administration was like a chapter out of atlas shrugged.
Barack Obama was horrible in a million different ways, but he doesn’t get anywhere near the hate he deserves for destroying the used car market for an entire generation.
15-year-old cars with 150,000 miles that should be selling for $3,000 are going for $15,000. It’s abominable. Pure destruction and thievery.
Chandra Duggirala M.D retweeted
even more evidence that demand for CHEAP intelligence is INFINITE (i think quite literally)
I am shocked, absolutely shocked, that Anthropic ARR has likely gone vertical in October after they cut token prices.
Except the exact same thing happened for the entire AI market, which is why GPU rental prices have surged while token prices declined.
$100b plus in year-end ARR looks possible again.
Chandra Duggirala M.D retweeted
We will enter an age of abundance
If we can stop the communists for the next 5 years
May the bloodthirsty warmongers, who push for a war that never ends, burn in hell!
Elon Musk warned about the Ukraine war years ago, and it's absolutely tragic how accurately he predicted the devastating cost of human lives from this conflict
Back in October 2022, Elon didn't just call for peace. He actually proposed a peace plan involving UN-supervised votes in occupied regions, recognizing Crimea as Russian, guaranteeing its water supply, and Ukraine remaining neutral
He warned that continuing this war would bring devastating death and destruction, with countless young Ukrainians and Russians losing their lives
And look at what's happened since then
More than 2 MILLION military casualties, entire cities destroyed, millions of people forced from their homes, and the war is still going on with no lasting peace
Elon faced enormous backlash for proposing a path toward peace, but the human cost he warned about has become an absolutely horrifying reality
And that's what makes this so tragic. These are human beings with parents, families, loved ones, and entire lives ahead of them, not disposable pieces on some geopolitical chessboard
How many more young people have to die in trenches before ending this endless slaughter becomes just as important as deciding who wins the war?
Elon saw the devastating human cost of a prolonged war coming years ago, and instead of taking his warning seriously, people attacked him for even putting a peace proposal on the table
Just imagine how many lives could have been saved if a lasting peace had been reached years earlier
How much more death and destruction will it take before ending this meat grinder becomes the priority??
Chandra Duggirala M.D retweeted
The U.S. spends a trillion dollars a year on formal defense.
It also has guys like this.
Danko flies a wood-and-fabric MiniMax under FAR Part 103 rules. No pilot certificate required. No medical. Just a sub-254 lb aircraft, a Kawasaki 440 snowmobile engine, and whatever else he feels like bringing along. In recent clips he’s shirtless with a handgun on his hip, low-leveling over fields and runways like it’s a normal Tuesday.
These aren’t theoretical “armed civilians.” They’re people who already own the tools, already operate outside the usual regulatory boxes, and already treat self-reliance as normal. Formal military power is real. So is a large, dispersed population that is comfortable with both flying and carrying. The combination is awkward for anyone who prefers their adversaries neatly organized and easy to target.
Chandra Duggirala M.D retweeted
Growing older is grappling with the realization that my schizophrenic relative with paranoid delusions has a higher hit rate than the people who dismiss him.
WTF, $3000 Samsung TV has an always-on mic, automatic content recognition, geo tracking, network mapping, and ships your data to Samsung + 400 partners.
Bro Rooted a brand-new 2025 Samsung TV after noticing absurdly targeted ads.
Firmware was locked down, so zero-days it was.
Got a root shell.
then found: always-on voice service, ACR, geo, full telemetry pipeline, and four separate DRM systems.
This isn’t a bug, It’s the product bro.
Chandra Duggirala M.D retweeted
🚨 BREAKING: Steve Hilton announces plan to cancel High Speed Rail, turn it into the Government Failures National Monument.
Replying to @chamath
Pretty much.
As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers.
SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue.
That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.
Replying to @johnhawkinsrwn
Yeah, this retard is so fucking dumb that he doesn’t realize that our policies serve the Ukrainian people.
Pieces of shit like this want the war to continue until every last man in Ukraine has died in a ditch.
They are utter scum of the Earth.
There are sometimes self fulfilling prophecies.
But nothing says that these stats predict anything.
This is the CRAZIEST Market Statistic of all time
Dont skip this. It’s the most IMPORTANT thing you’ll Read.
All have a 100% WIN RATE
First: FED HIKING CYCLES.
Since 1983, the S&P 500 $SPY has risen during every completed multi-meeting hiking cycle.
7/7.
Average gain: ~10%.
Then MIDTERMS.
Since 1950, the S&P 500 has been higher 12 months after every midterm.
19/19.
Average gain: ~15%.
And this one might be even crazier:
Since 1942, EVERY November-through-June period following a midterm has been positive.
21/21.
Average gain: +16.6%.
Then add one more:
12 months after the final Fed hike?
7/8 positive.
Average gain: ~20%.
The only miss: May 2000 during the dot-com collapse.
So 2026 has:
7/7 during Fed hiking cycles.
19/19 after midterms.
21/21 from November through June after midterms.
Three completely different historical setups.
Three near-perfect streaks.
All lining up in the same year.
History guarantees nothing.
But this is VERY hard to ignore.


































