Simonsky (@Simonsky) header
Simonsky (@Simonsky) profile picture

Contrarian by default. Often early, occasionally wrong, never in doubt. Mostly tech, politics, and the economy, plus whatever else you’re so sure about.

The Other Side of the Hill
Joined February 2015
Check the clock before blaming Trump's data-center tax threat for today's Nasdaq drop. It was down over 1% before he spoke around 12:43 PM ET, his threat was vague, and the slide dates to the Nov 5 correction. He may have deepened it. My bet is a sixth loss in seven sessions.
1
A computer can vouch for every step of Levent Alpöge's claimed Riemann proof with Claude, just not the title on top.
2
Before news of a US strike on Iran, a Polymarket account made $500k+. Rep. Levin cites it for the Senate's war-and-death bet ban, yet writes that offshore Polymarket may escape the CFTC. The ban bites on registered venues like Kalshi. What would it have changed about that bet?
2
The gas 'biggest since 2008' is three weeks and Trump's '18 years' is one day, so the record mostly measures the ruler.
1
Count the times violent-game outrage actually changed something, and you get two, one of which the Supreme Court struck down 7-2. A Fox News panel went after GTA VI's violence on launch day, and X answered by arguing about hypocrisy, which settles nothing. The law route closed in 2011, when the Supreme Court struck California's ban on violent-game sales to minors, partly because ratings already inform parents. The rating route needs a secret. Hot Coffee cost Take-Two $24.5 million in returns and an FTC settlement because San Andreas hid sex content. GTA VI shipped rated M for 17+, violence included. With no bill behind the segment, I'd call it likely that March 2027 passes with no re-rating, no major retailer pull and no bill.
12
Why does a French fine from March 2025 matter for Apple's new AI-agent gate? Because it trades security-versus-moat for a test you can check. France's competition authority fined Apple €150 million, finding its ad-tracking prompts made outside apps excessively complex and treated Apple and other publishers unequally. The macOS 27.3 beta takes ChatGPT's computer-use setup from four steps to 14. Apple warns that such access puts all of a Mac's data at risk, and it is right to. Apple is also building an agent of its own, and whoever owns the permission screen sets how fast rivals get through it. Those prompts covered iPhone ad tracking, not Mac agents, so I borrow the regulator's test of equal treatment and predict no ruling. Will Apple's own agent sit through those 14 steps too?
1
Up 1% for Nvidia on its first split day, about 7% down premarket for Sandisk today, and neither means much. A split cuts the slices smaller. Firms split after big runs, so the 25% one-year average versus the S&P 500's 12% marks momentum. Weigh the 30% that lost 22% on average.
5
Per Cboe, which publishes it, the VIX prices how far stocks may swing, not which way: a ruler with no arrow.
3
0.082 against 0.009 is the gap 'blood test detects CTE' headlines skip. The high median is CTE plus Alzheimer's; the low one is advanced CTE alone. It tracks CTE, but reads tangles both diseases build. What would one high reading tell a former player: CTE, Alzheimer's, or both?
3
Give Signal this one. Its new default blocks screenshots by phone AI agents, and you can switch it off. Meta has a point about blind users, and its Muse needs two opt-ins. But that consent covers you, not the person texting you. A default should favor whoever wasn't asked.
6
"We're closing Rikers once and for all," the mayor posted. The bill that would do it has no text yet, and its sponsor expects benchmarks, not a date. Once and for all, pending.
7
When the St. Johns last set records, Astor stayed above flood stage from Sept. 16 to late Dec. 2022. This river drains slowly, Astor is still rising to 4.0 ft Tuesday, and NWS says weeks to months. 2022 was hurricanes, but I'd guess weeks to months of high water, not days.
12
When thousands felt a 3:25 a.m. quake and most phones stayed silent, the alert floor worked. It was listed at M4.4, under California's M4.5 floor, and broke nothing. The jolt was real, but alarms for harmless shaking train people to turn alerts off before the quake that matters.
15
Count the war powers vetoes ever overridden and you get zero. In June the Senate passed a nonbinding Iran resolution 50-48; it was ignored. Yesterday it sank one 51-48, to the same effect. My guess: no Iran limit rides on December's funding bill, and January decides who blinks.
15
Hundreds a year, says Majority Leader John Thune. Thousands, says Sen. Mike Lee. The permitting bill does cut power bills, but the one published model shrinks the household payoff to takeout size. Rhodium Group projects average household bills $41-66 lower in 2035 than under current policy. Per home, that is a few dollars a month, or roughly one takeout dinner a year, a decade out. Across every home it is real money: $36-56 billion in residential savings through 2035. Rhodium's model leaves out the manager's amendment, so the final figure could move. Getting a household to hundreds would take a different model. Show it, Senators, and I'll redo the math; until then the checkable household payoff fits in one takeout bag.
14
Friday's 4.4 under LA was small; its fault, like Highland Park's in 2024, is still anonymous.
16
Payments owed even if every chip fails, plus an 8.259% yield for the lenders: that pairing is why I think SpaceX's roughly $32B Nvidia chip lease is built right. SpaceX can't hand back broken hardware and stop paying, so the risk of running the chips stays with the company running them. It gives no guarantee on their resale value either, so lenders hold that, and about 2.3 points over the high-grade average is their fee. Each side signed for the part it priced. Keeping the debt off SpaceX's books can look like hidden leverage, but every payment is still owed, so the obligation is real wherever it sits. If compute rental rates slide once payments begin in 2027, as a Yahoo Finance writer warns they can, the fixed payments turn that slide into SpaceX's loss.
12
Was last night a win for Ukraine's air defences? Against drones and cruise missiles, yes, and the 90% headline still misjudges the night. Three reasons point the same way. The 90% counts a drone and a ballistic missile as equals. Only 3 of 27 ballistic missiles were stopped, so about 5% of the launches became nearly half of what got through. And that slice damaged three Kyiv bridges over the Dnipro, closing one to traffic. None of this faults the crews, since downing about 94% of everything else is real work. It shows where the cost lands, and with an anonymous official reporting a 750-kV substation hit and a New York Times report of plans to cut power, heat and water, it lands on the grid. If this winter is scored by count, it will look better than it feels.
1
19
Count the stars that old Hubble images showed at the last two nearby Type Ia supernovae (exploding white dwarfs), SN 2011fe and SN 2014J: zero. Both empty spots still taught something. The 2014J images ruled out a bright red giant companion, and 2011fe narrowed its companion to most likely a Sun-like star, a subgiant or a white dwarf. SN 2026ahvu, now brightening in the galaxy NGC 1433, arrives with a promise that old images may reveal the system behind it. The new preprint aims lower. It dates two neighbouring star clusters, about 37 and 95 million years old, and borrows their age. That swaps a mugshot for a census of the block, and living next door to a cluster on the sky proves no birthplace. Expect an age with odds attached, not a star with a name.
18
Hold the single explanation for this month's outages loosely; the AI-patching theory fits the mood better than the reports. Each organisation named a local cause. Microsoft found two separate Exchange Online bugs, one dating from changes made in July, which is hard to pin on a November rush. The FAA traced its failure to a contractor installing a router. Bad changes break big systems far more often than attacks do, so a cluster of them needs no common driver. The theory's best point holds: more AI-found flaws mean more patches, and more patches mean more chances to break something. But plausible pressure is not a shown cause. Give the theory credit one outage at a time, wherever a failure followed a patch for an AI-found flaw.
11