Sign in

Kate Sills

@katelynsills.com
3.4K followers 2.4K following 3.4K posts

farm kid → cognitive science & computer science at UC Berkeley → computational law. Founder of CitationClerk.com?s=b Political philosophy with a thin veneer of software engineering. Guest speaker at Dartmouth, Stanford, SF, NYC, London, and Australia.

PostsRepliesMedia
Kate Sills @katelynsills.com · 03/10/2026
One last thing on this: the Right wants women to be afraid of men. They're basically saying "yes, all men!" When they make jokes like "Watch out, women!" what they're really saying is that these dogs bite. It's a flex of power and control. They weirdly want the predators and prey framing.
“Yeah, I want these guys on it. But you know what's going to happen? The guys that don't need it are going to take it to triple boost, right?” Watters told fellow panelists on The Five. “And then they're gonna get out there, and women on base, you guys better be careful. Port calls women in Asia, you better be careful because these guys are gonna be wild animals, and you better watch out.”
1110
Kate Sills @katelynsills.com · 02/10/2026
Continued:
How the "turning" works inside the model. The model can't literally spin a dial, but storing each position as a cosine and sine makes rotation simple arithmetic. A high-school trig identity does it:

cos(a + b) = cos a · cos b − sin a · sin b

So multiplying and subtracting the stored waves for 37 and 48 produces the waves for 85 without the model ever computing "85" directly. The paper suspects the MLP layers do something like this, but couldn't confirm the exact mechanism.

One simplification: the real model is fuzzy. Each clock gives a blurry vote rather than a crisp position, and the votes combine into a best guess. That fuzziness is part of why accuracy was 77 to 98% rather than 100%.
200
Kate Sills @katelynsills.com · 02/10/2026
Here's Claude's example. Normal caveats apply -- this needs double-checking, but for explanatory purposes in broad strokes, it's very helpful.
Let's add 37 + 48 (answer: 85). This one has a carry, which shows nicely that Clock never needs to "carry the 1."

Step 1: Place each number on every clock. A number's position on a clock with T positions is just its remainder when divided by T. The model actually stores this as an angle, using sine and cosine.

Clock	37 sits at	48 sits at
10-position (last digit)	7 (252°)	8 (288°)
2-position (even/odd)	1 (odd)	0 (even)
5-position	2	3
100-position	37	48
Line (size)	37	48

Step 2: Add by turning the hands. On each clock, start at the first number and rotate forward by the second:

10-clock: 252° + 288° = 540°, which is one full turn (360°) plus 180°. The hand lands at 180°, which is position 5. So the answer ends in 5. Notice there was no carrying: going past 12 o'clock just wraps around.
2-clock: odd + even lands on odd. Consistent with ending in 5.
5-clock: 2 + 3 = 5, which wraps to 0. So the answer is a multiple of 5, also consistent.
100-clock: 37 + 48 lands on position 85.
Line: 37 + 48 is "around 85," a rough sense of size.

Step 3: Read off the answer. The clocks pin down the answer's position in each cycle: ends in 5, odd, a multiple of 5, position 85 out of 100. The 100-clock alone can't tell 85 apart from 185 or 285, since those all land in the same spot. The line breaks the tie: the sum is around 85, not 185. So the answer is 85.
110
Kate Sills @katelynsills.com · 02/10/2026
I had so much fun playing a show last night! Here's another Lonesome On'ry and Mean practice:
021
Kate Sills @katelynsills.com · 30/09/2026
my mom just texted me that a stray kitten showed up and is terrorizing their adult cat
a fluffy little grey and white kitten hunches over a food bowlThe fearsome kitten intimidates my parent's catThe kitten gets all up in my parent's cat's space
2385
Kate Sills @katelynsills.com · 28/09/2026
The study you linked says exactly what I said. If you don't do your homework, you will not learn from your homework.
100
Kate Sills @katelynsills.com · 28/09/2026
We don't even need to get into theory. There are direct and easy to understand counterexamples that disprove what you say here. For example, Levant Alpöge used Claude Fable (an LLM) to show the Jacobian Conjecture was false, and this was neither brute force nor a targeted machine learning model.
I'm baffled by these PR stunts. Targeted machine learning models and brute force algorithms can sometimes solve difficult problems that have eluded human beings. But using an LLM doesn't add anything (and makes success less likely). They are not "AI." They are stochastic parrots. 
1/3
2372
Kate Sills @katelynsills.com · 28/09/2026
it's time
Muted words: Choosin' Texas, Ella Langley, Hasan Piker, James Carville, soccer
051
Kate Sills @katelynsills.com · 26/09/2026
But hoo boy, look at this from Posner's article that Kronman cited. I bought Grant Gilmore's book The Death of Contract about ten years ago, and it's still on my to-read list, but this is such a crazy thing to say:
Some commentators, however, continue to deny the possibility
of social science. Among them are academic lawyers such as Grant
Gilmore and Arthur Leff. Professor Gilmore wrote recently:
For two hundred years we have been in thrall to the eighteenthcentury hypothesis that there are, in social behavior and in
societal development, patterns which recur in the same way
that they appear to recur in the physical universe.
• . . [T]he hypothesis is itself in error. Man's fate will
forever elude the attempts of his intellect to understand it. The
accidental variables which hedge us about effectively screen
the future from our view. The quest for the laws which willexplain the riddle of human behavior lead us not toward truth
but toward the illusion of certainty, which is our curse. So far
as we have been able to learn, there are no recurrent patterns
in the course of human event; it is not possible to make scientific statements about history, sociology, economics-or law.
010
Kate Sills @katelynsills.com · 26/09/2026
On pages 232-3 we start to get more of an explanation of why Kronman thinks this is one of the two defining statements of law&economics, but it is still bizarre. In the cited source at n. 112, "Some Uses and Abuses of Economics in Law," Posner doesn't say this, and says something very different!
"a particular course of action is better than some other if, all things considered, it has the same results...but requires fewer resources to do so."Kronman says that using more resources while achieving the same ends is wasteful and economics says we ought to take the other action, and in this way, economics provides an "ought," a normative point of view.
100
Kate Sills @katelynsills.com · 25/09/2026
Continuing my lunch time reading of The Lost Lawyer by Anthony Kronman: Kronman is starting to explain why law & economics undermines the "horse-sense" or prudential wisdom he feels is necessary for a lawyer (p. 227) But this is a very unusual definition of economics, one I've never seen before:
Kronman's definition of economics has two points: every human action requires the use of scarce resources, and human action is always rational in that it is always motivated by a desire to eliminate waste.
282
Kate Sills @katelynsills.com · 22/09/2026
Sometimes I still forget this, but I think it's sound advice. Without the requisite charity, you really can't make intellectual progress. If you're focused on dunking, you can't actually consider ideas, turn them over in your head. A culture of dunking is anti-curiosity. It's anti-academic.
Crop rotation meme by Amy Ash:

(me making fun of your crop rotation idea and thereby holding our people back another 5000 years) jeff thinks the beans have to take turns lmao
161
Kate Sills @katelynsills.com · 19/09/2026
Not in the text of the original opinion itself, but how future courts refer to the original opinion is indicative of how they feel about it. I give a few potential examples based on the Sam Spade case here: katelynsills.com/law/indicato...
Potential Indicators
Even though the Sam Spade case was never highly cited, and even though it is not silently overruled now, it’s still a useful exercise to look for the textual indicators that a court may not look on a prior decision with favor.

We saw a few of them earlier:

When a proposition is often being called “dicta” (although that signal may be less available in the Ninth after Barapind).
When a case is mentioned only to distinguish it.
When a proposition is introduced then immediately followed with a “nevertheless” or some other word that indicates moving past it, as in Air Pirates.
There are also potential confounders, which we saw, such as the parties choosing not to use an argument, the same test being smuggled in via a cite to a later opinion, or the opinion restating in its own words without citing.

I realized when writing this that what I was calling indicators or factors are really features in computer science. Features are anything that could potentially be used to shift the probabilities of a label. For instance, if we were trying to label an animal as a dog or cat, we might have a list of features like “barks” and “meows” and “is fluffy.” A counterexample such as a mute dog or a hairless cat wouldn’t negate the overall value of the feature list, since the features are considered together as a whole, each with different weights. The confounders I mentioned earlier are where the feature fires without the label being true, which is fine if we are not relying on any one feature.
100
Kate Sills @katelynsills.com · 18/09/2026
"I would not really consider Rice to be a character case" Maybe you'd be interested in Robert Helfing's analysis. He's a former professor of copyright law, represented the Shelby defendants in Halicki v. Sanderson in the Ninth. Helfing would find the idea that Rice is not a character case very odd
200
Kate Sills @katelynsills.com · 18/09/2026
3. Here's a chart of how many times the relevant 9th Circuit opinions are cited (comparing apples to apples directly), and I provide the code and data to back it up. For what it is worth, per Westlaw, the Sam Spade case has only 61 citing cases, whereas Rice (2003) (same area) has 419 citing cases.
A chart of citing opinions per year, 1954–2026. The Ninth Circuit opinions on character and copyrightability are graphed by their citation count per year. This is based on citation for any reason, by any opinion regardless of jurisdiction, so it is the most naive version.
200
Kate Sills @katelynsills.com · 18/09/2026
2. I agree. This actually happened in both directions for the "story being told" test. Here is the court in Anderson (C.D Cal) and the court in Gaiman (7th Circuit) talking about the same exact texts but one saying the test was "cited with approval" and another saying it was "killed"
Anderson v. Stallone (C.D. Cal. 1989)	Gaiman v. McFarlane (7th Cir. 2004)
Olson 1451–52 n.6: “we recognized that it is unclear whether this language is dicta or an alternate holding, but we declined to resolve the issue”	Olson 1452 and n.7: “We therefore need not resolve the issue left open in Air Pirates … whether the Warner Bros. statements should be considered dicta”
Air Pirates at 755: “the Warner Brothers language does not preclude protection of Disney’s characters”	Air Pirates at 755 and n.11: “we need not endorse the district court’s conclusion that Disney’s characters fell within the Warner Brothers exception”
Anderson Verdict: “cited with approval … and declined to characterize this language as dicta”	Gaiman Verdict: “The Ninth Circuit has killed the decision … though without the usual obsequies”
200
Kate Sills @katelynsills.com · 18/09/2026
Thanks for the response. I address a lot of this in the piece. Re: 1. If it were truly overruled, we would expect the district courts to no longer apply or reference the "story being told" test. But post-Daniels, they do seem to treat it as live and they do apply and reference it.
The Reaction to Daniels
The district courts subsequently interpreted Daniels as reaffirming the “story being told” test and applied or acknowledged it:

Opinion	What it said
Ricketts v. CBS, C.D. Cal. No. 2:19-cv-03895, Mar. 19, 2020 (ECF 125, reconsideration denied)	“The Ninth Circuit did recently issue a decision analyzing the ‘story being told’ line of character copyright protection. Daniels v. Walt Disney Co. … (‘Since the 1950s, we have also extended copyright protection to characters—both literary and graphic—that constitute “the story being told” in a work’). However, Plaintiff does not contend (nor could he) that this case evidences any material change in law.” (at 10)
ZAG America v. Harasymets, C.D. Cal. No. 2:21-cv-05100, June 3, 2022 (ECF 34, default judgment)	“Nor has Plaintiff alleged that the characters ‘Ladybug’ and ‘Cat Noir’ constitute ‘the story being told’ in a work.” … There are no allegations that the characters ‘so dominate the story such that it becomes essentially a character study.‘” (at 8)
Carroll Shelby Licensing v. Halicki, C.D. Cal. No. 8:20-cv-01344, Nov. 29, 2022 (ECF 350, summary judgment)	“Copyright protection also extends to characters under an alternative ‘story being told’ test. Daniels v. Walt Disney Co., 958 F.3d 767, 773–74 (9th Cir. 2020). Because the parties do not advance a theory of copyrightability under the test, the Court assumes that Eleanor does not meet it.” (at 8 n.8)
Gilbert-Daniels v. Lions Gate, C.D. Cal. No. 2:23-cv-02147, Dec. 7, 2023 (ECF 133, summary judgment)	“characters are often ‘only … chessman in the game of telling the story.’ … Such ‘chessmen’ characters are not afforded copyright protection. … Put another way, when ‘the character really constitutes the story being told,’ such a character is entitled to protection.” (at 32)
Biani v. Showtime Networks, C.D. Cal. No. 2:23-cv-03845, Mar. 29, 2024 (ECF 23, motion to dismiss)	“The Ninth Circuit has articulated two tests … the ‘story being told’ test and the …
200
Kate Sills @katelynsills.com · 16/09/2026
We're missing our bass player and we're still learning it but here's a preview
050
Kate Sills @katelynsills.com · 10/09/2026
The latest AI models get near perfect scores on the LSAT, our best test of logical reasoning and reading comprehension. The test is multiple choice, and there is no way that mere statistical next word prediction alone can do this. AI is doing sophisticated logical reasoning, and that is a fact.
3131
Kate Sills @katelynsills.com · 09/09/2026
Yes. Current AI is able to find extremely sophisticated attacks. That's settled fact, not my opinion. For example, here are the critical vulnerabilities identified post-Mythos:
Critical and high severity vulnerabilities from 21 notable organizations.
100
Kate Sills @katelynsills.com · 04/09/2026
Continuing my read of The Lost Lawyer: In Ch. 3, Kronman makes the circular argument that, since judges aim towards the public good, successful lawyers must also have the same mentality. He forgets that even if lawyers and judges are aligned, it's not clear that it is for the public good.
"It follows that a lawyer arguing before a judge who is responsible for maintaining the well-being of the law must himself become a master of analysis from the judicial point of view..."
110
Kate Sills @katelynsills.com · 01/09/2026
Wendell Berry in the Unsettling of America: "It is rarely considered that this average citizen is anxious because he *ought* to be... He ought to be anxious, because he is helpless."
131
Kate Sills @katelynsills.com · 30/08/2026
Anyways, Kronman ends the chapter by admitting that "political fraternity" is conservative in nature and upholds the status quo but that is acceptable because it is better than revolution (his false dichotomy.)
"The good that the statesman seeks to secure is the good of political fraternity. In most ordinarily complex communities, this is a good not of secondary but of primary and independent worth. It is, in an elementary sense, the good of politics itself. Those who deny this and celebrate instead the liberating worldlessness of revolution fail to see that politics always the pursuit of order, and that its inherent conservatism implies a continuing affirmation of the value of political fraternity, in all but those transitional episodes of birth and death that mark the limits of political life. In this sense it is right to say that my account of statesmanship, with its emphasis on the value of political fraternity, entails a commitment to order and the status quo. But that is because it entails a commitment to politics itself, to the existence of the realm of values that the statesman serves and political fraternity sustains against the forces of disorder that threaten all our fragile human works."
100
Kate Sills @katelynsills.com · 30/08/2026
He argues that just as a person must have a sense of integrity with regards to their prior actions (and any regret), a community must be joined by bonds of sympathy if it is to be a community at all.
"For when a community is divided by a contest among incommensurable values important enough to place its identity in doubt, it is in the preservation of political fraternity that the public good largely consists.... if a community is to survive such disagreements... what it needs is political fraternity, a condition midway between tolerance and union, and one that is marked by the same combination of sympathy and detachment..."
100
Kate Sills @katelynsills.com · 28/08/2026
I find it very odd that none of the papers who have picked up the story include any kind of map, or any information about what *specifically* is being proposed. Feels Glonzo-y. The already owned, private parcel abuts Yosemite and is less than 165ft from the highway that they want to connect to.
110
Kate Sills @katelynsills.com · 28/08/2026
The opening brief has this (very poor) map, with 120/Big Oak Flat Road (the road they want to connect to) in green. storage.courtlistener.com/recap/gov.us...
The roads, and their relationship to Hazel Green, are depicted by the
following map (see ER 1089, 1119), with the Crane Flat Road segment at issue
appearing in purple, and marked by global positioning system (“GPS”) points on a
short line to the east of Hazel Green, and the Coulterville Road segment at issue
appearing in red, and marked by GPS points on a longer line to the southeast (and
eventually turning east and then northeast) of Hazel Green:4
110
Kate Sills @katelynsills.com · 28/08/2026
Here are the parcels in question in Mariposa County. The "120" is the main road also known as Big Oak Flat Road.
Private parcels near Yosemite in Mariposa County: 006-010-025 and 006-010-026
120
Kate Sills @katelynsills.com · 19/08/2026
Here's what I mean. Kronman doesn't really discuss tech (at least in what I've read so far) so this is more implied and my read of society generally: 4/
A 2x2 chart:

on the top:

Good at technology
Bad at technology

On the left: 
Good at “practical wisdom” / understanding people
Bad at “practical wisdom” / understanding people

Good at both: ??? We don't understand this category
Bad at tech: statesmen lawyers
Good at tech: software engineers.
Bad at both: generally incompetent
120
Kate Sills @katelynsills.com · 19/08/2026
For what it is worth, Arvind Narayan says: substack.com/@aisnakeoil/... Is he wrong?
The key thing to keep in mind is that it is in fact possible to watermark LLM-generated text without degrading output quality (and in fact achieve a much stronger property, which is that the distribution of possible outputs is unchanged.) This is well established technically, and it has been implemented by Google / Gemini for over two years, and no one seemingly cared. 

Admittedly, quality-preserving watermarking is one of those counterintuitive facts about probability, like that annoying Monty Hall problem (the one with two goats behind three doors). The theorem-understanding part of my brain has no problem with it, but the intuitive part of my brain is screaming that there must be some mistake.
130
Kate Sills @katelynsills.com · 18/08/2026
It's all online! They were absolutely obsessive about note taking & the records have been preserved by the Moravian Archives. www.moravianchurcharchives.org/digital-acce... Here's Barbara Ricksecker, my 5th great grandmother, in German:
160
Kate Sills @katelynsills.com · 17/08/2026
lol
Patrick Wyman: She ran a Crossfit gym, which is basically a swinger's club that produces orthopedic injuries

G. Willow Wilson: This is a banging first sentence of a noir novel
022035
Kate Sills @katelynsills.com · 12/08/2026
040
Kate Sills @katelynsills.com · 12/08/2026
Here's a gold miner's diary for instance. He'll talk about people who come to him to borrow books but it's not like he's treated as a freak for reading www.sierracountyhistory.org/wp-content/u...
by dusk. The boys had washed the dishes.
towels & C. We soon had supper & they have
all gone to bed. Frederick has been to
Good Years Bar this afternoon, he returned
since I did.
Sunday Eve Sept. 14th 1856
For once in a long time have been at home all day. Been reading novels. Finished Peter Simple & began Jacob
Faithful. I feel too much worn out when I come in from work
to attempt reading anything of a more weighty character. such
as the "Decline & Fall" which I began about a year since but
did not complete the first volume. I advanced as far as the
thirtieth chapter then commenced something lighter and have
020
Kate Sills @katelynsills.com · 02/08/2026
lmao holy crap. I had Claude do my weekly meal plan from my grocery store receipt, and from the receipt alone plus knowing I'm taking the LSAT on Thursday, it concluded I bought the beer to celebrate afterwards. 😂
The week's food plan is written into recipes/this-week.md (top section) and grocery.md now records the Aug 1 run.

The shape: Sunday is the one real-cooking night (curry — you've got both green and red paste, limes, cilantro, rice packets), then the dinners get progressively lazier as the week gets heavier — frozen chicken parm after Monday's PT, microwave-tier salsa-verde chicken or pork noodles on the double-header Tuesday, salmon + rice + broccoli for the calm Wednesday, and Thursday is treat-yourself (the lone Sierra Nevada Cosmic in the fridge looks suspiciously like it was purchased as a post-LSAT beer).
3580
Kate Sills @katelynsills.com · 30/07/2026
this seems to help
Instructions for Claude: Disagree when I'm actually wrong, but don't manufacture an objection just to seem balanced.
0101
Kate Sills @katelynsills.com · 29/07/2026
One of the most interesting things I remember from the course is the time of Homer is a time of great poverty compared to the Mycenaeans that preceded them. It's the Dark Ages, and we can tell even from how Homer describes people. No one can write or read, no one has scribes, queens themselves weave
And of course, the kings when we see them in Homer do not have a bureaucracy, do not have scribes. How could there be scribes? They didn't know how to write. There are no inventories. These are not rich guys with fantastic quantities of stuff that has to be cataloged and inventoried. The kings, by our standards for kings, are really very poor. What do they do in their spare time or in their time in general? They, themselves, engage in agriculture. I don't mean they dig in the ground, but they supervise it; they think about it. They are like people who are in charge of a plantation, their own plantation, I mean. So, that's one of their activities. You cannot imagine the great kings of Mycenae doing that. They would've had all kinds of subordinate officials taking care of that. Another thing that these kings are seen to do, which is in a way even at a lower level, is to be herdsmen. Again, not themselves out there with the goats and the sheep, but they are referred to as having these herds and having to cope with them.Remember, I tried to emphasize how rich you would have to be to undertake the building of the temple, to undertake the construction of one of those beehive tombs. You had to be very confident because it was going to take a long time, a fantastic amount of labor, a tremendous amount of money to do that. There is no evidence in Homer that anybody had that kind of wealth or that kind of power. What do the queens in Homer do in their spare time? Or maybe it's not their spare time, maybe it's their regular time. Well, one of the things that they do when the Homeric heroes refer to their wives — there are really only two places that they seem to be associated with. One is the bedroom and the other is the loom. What these ladies are doing is weaving cloth. Now, that's not what queens do. I'm sure the Mycenaean queens didn't do that, but it's very, very interesting that that is what the Homeric queens do. Well, how are we to explain the discrepancies that are so great between these worlds? And the answer that almost everybody now accepts without argument is that these poems were created orally, and passed on orally. That they were not written down, so that what we have in the poems of Homer reflects centuries of bards passing on bits of the poem, or the poem in various versions, being always creative. I mean, any bard, if by analogy we know about bards in the modern world, we have some evidence on that.
120
Kate Sills @katelynsills.com · 29/07/2026
Her argument: Nolan inserts "contemporary American ideas" about morality, different than those "in the world of Homer" and so the characters’ motivations don't make sense. The characters often take the same actions as in the texts, but we no longer understand why. www.lrb.co.uk/the-paper/v4...
The Odyssey 
directed by Christopher Nolan.
Christopher Nolan’s $250 million action movie version of Homer’s ancient Greek poem is an entertaining way to hide from the high temperatures for a few hours. Nolan’s trademark jumps and twists in the narrative structure are relatively easy to follow, and appropriate for The Odyssey, a narrative of tales nested in tales. My teenagers had a good time. Unlike some of Nolan’s other films, The Odyssey is not boring, thanks to the source material, which is impossible to mess up entirely. It’s a family-friendly audiovisual spectacle, like an elaborate Fourth of July fireworks display – and with about the same level of narrative and emotional depth.

As in his previous films, Nolan is interested in the discontinuities of space and time, in magic, tricks, craft, technology, rivalry, substitution, masks, miscommunication, what we remember and what we choose to forget. Yet again, survival is portrayed as a kind of triumph. Yet again, there are long, claustrophobic scenes of drowning and near-drowning, and technically impressive episodes in which buildings are engulfed in flames and vehicles explode. Yet again, we watch a male character’s desperate quest to recover and/or avenge a female love object. I had hoped that Nolan’s affinity with these Homeric themes might push him to new creative heights, and enable him to conjure more believable characters. But The Odyssey features his usual combination of grandiosity and superficiality. Without a tub of wax for my ears – which Himesh Patel’s ever anxious Eurylochus, Odysseus’s doomed right-hand man, uses to sail safely past the Sirens – I was exposed to the full din of thumping sticks, staffs, swords and clashing armour, thunderbolts, crashing waves, grunts, screams, toppling palaces and burning temples. But unlike Matt Damon, the wax-free Odysseus who is tied to the mast as he is conveyed past the Sirens (naked ladies reclining on some uncomfortable-looking rocks), I gazed at the shipwre…
130
Kate Sills @katelynsills.com · 20/07/2026
the self talk is always so funny. "staying confident"
What matters most is that I've verified the mathematics directly. The map works as stated, the counterexample is valid, and symbolic verification leaves no room for hidden errors. Whether the specific announcement is authentic or how the mathematical community has received it remains uncertain given my training cutoff, but the mathematical claim itself is sound and checkable. y, z). Yes.

Now I'm drafting the response. I need to give a direct answer that the math checks out, explain what I verified and why it matters as a counterexample, and be honest about the one real uncertainty—whether the tweet itself is authentic. The account @alpoge does correspond to Levent Alpöge, a real mathematician, but I can't verify the screenshot is genuine. Still, the mathematics stands independently of who posted it.

I'll keep this conversational and tight without needing visuals or artifacts, staying confident about the math while being appropriately cautious about claiming the Jacobian conjecture is now universally accepted as false—what I can say with certainty is that this specific map is a valid counterexample, and if it's genuinely as shown, the complex Jacobian conjecture is false.

Show less
Done

Short answer: the math in it is real. I checked it directly, and both claims hold exactly.

Here's what I verified symbolically (exact arithmetic, not numerics):

The map


𝐹
(
𝑥
,
𝑦
,
𝑧
)
=
(
(
1
+
𝑥
𝑦
)
3
𝑧
+
𝑦
2
(
1
+
𝑥
𝑦
)
(
4
+
3
𝑥
𝑦
)
,
  
𝑦
+
3
𝑥
(
1
+
𝑥
𝑦
)
2
𝑧
+
3
𝑥
𝑦
2
(
4
+
3
𝑥
𝑦
)
,
  
2
𝑥
−
3
𝑥
2
𝑦
−
𝑥
3
𝑧
)
F(x,y,z)=((1+xy)
3
z+y
2
(1+xy)(4+3xy),  y+3x(1+xy)
2
z+3xy
2
(4+3xy),  2x−3x
2
y−x
3
z)

has Jacobian determinant identically equal to 
−
2
−2 — a nonzero constant, as a polynomial identity, not just at a few points. And all three listed points map to the same image:
1230
Kate Sills @katelynsills.com · 20/07/2026
The search results don't show any July 2026 news about the Jacobian conjecture being disproved. Based on my knowledge through January 2026 and these search results, it remains an open problem—the most recent sources from November 2025 confirm it's still unresolved, particularly for dimensions greater than one. I can't verify the specific tweet claim through these searches since the results are mostly older papers.

That said, I've directly verified the mathematical claims in the tweet: the Jacobian determinant is indeed constant and nonzero at -2, and all three listed points do map to the same output, making the map non-injective. This combination—a non-injective polynomial map with constant nonzero Jacobian—would be a genuine counterexample to the complex Jacobian conjecture. The critical issue is that such a simple, directly verifiable counterexample to a problem open since 1939 would be an extraordinary mathematical discovery, which makes me cautious about accepting it without peer review or confirmation from the mathematical community. The math is genuinely sound — this counterexample to the Jacobian conjecture holds up under verification, so the only real question is whether this is Alpöge's authentic recent announcement and how the mathematical community has received it, though I can't verify that part beyond my knowledge cutoff. F is not injective since it maps two distinct points to the same output, which directly contradicts what the Jacobian conjecture claims. I verified the determinant symbolically—it's identically -2 everywhere, so there's no computational error. This is a genuine counterexample to the conjecture, though I should note the distinction between confirming the mathematics is correct versus whether this represents a new accepted result in the mathematical community.

The "too good to be true" concern is worth examining: a problem unsolved for 87 years suddenly disproved by a simple map anyone can verify in seconds. But this could plausibly ha…
1140
Kate Sills @katelynsills.com · 20/07/2026
more checking
What's striking is that a counterexample to such a famous conjecture is either right or wrong, and this one checks out mathematically. Anyone can verify it in seconds with a computer algebra system. The reason such a simple counterexample might have been missed is that searching the vast space of Keller maps for non-injective ones isn't something brute-force methods easily accomplish—it requires clever construction. The timing and context suggest this might involve AI assistance, which adds another layer to the story.

Let me search for whether this is actually confirmed news or if there's more context I'm missing.

Show less
Done

I did one more check on the exact form of the conjecture to be sure I wasn't misremembering it. Let me search once more for any coverage of this specific claim.

Verified mathematical validity while interrogating authenticity claims
Verified mathematical validity while interrogating authenticity claims
"Jacobian conjecture" disproved July 2026 counterexample dimension 3

9 results

Jacobian Conjecture Overview
www.emergentmind.com
a note on sufficient conditions for the real jacobian ...
ejde.math.txstate.edu
A new class of Nilpotent Jacobians in any dimension
arxiv.org
Very degenerate polynomial submersions and counterexamples to the real Jacobian conjecture - ScienceDirect
www.sciencedirect.com
Injectivity and Almost Global Stability of Hurwitz Vector Fields
arxiv.org
Jacobian Conjecture in two dimension
arxiv.org
Counterexamples to a Conjecture of Ahmadi and Shparlinski
arxiv.org
AN INTRODUCTION TO THE JACOBIAN CONJECTURE WILLIAM GARLAND
math.uchicago.edu
An Optimization Approach to Jacobian Conjecture
arxiv.org

The search results don't show any July 2026 news about the Jacobian conjecture being disproved. Based on my knowledge through January 2026 and these search results, it remains an open problem—the most recent sources from November 2025 confirm it's still unresolved, particularly for dimensions greater than one. I can't verify t…
1150
Kate Sills @katelynsills.com · 20/07/2026
claude is confused by the shitpost tone so checks the definition of the problem to be sure
Let me verify the non-injectivity claim more carefully by testing specific points to make sure I didn't make a computational error.

Checking F(1, -3/2, 13/2): I'm working through the component calculations to confirm this maps to (-1/4, 0, 0), which would match F(0, 0, -1/4). The arithmetic checks out—the map genuinely sends two different points to the same output, so it's truly non-injective despite having a constant Jacobian. This is a legitimate counterexample to the conjecture. I'm being too cautious here. A hoax couldn't withstand direct symbolic computation—if the Jacobian determinant is genuinely -2 and three distinct points really do map to the same output, then this is a concrete counterexample, not something that merely appears to work. The only way this fails is if I've misread the polynomials from the image, so let me examine the exact expressions more carefully. I'm wondering if this might actually be a famous joke in the math community—the casual tone ("hello there," "thanx to my close friend akhil," "working during the world cup final") does read like a shitpost. But the mathematics itself checks out, so the real question is whether there's a subtle reason this doesn't actually contradict the conjecture, like the map being non-injective despite having constant Jacobian determinant.

Let me verify against the definition: the Jacobian conjecture states that if a polynomial map has constant nonzero Jacobian determinant, then it must be injective. A non-injective Keller map with constant Jacobian would be a direct counterexample—there's no loophole there. I'm trying to verify the source and credibility of this claim. The tweet appears to be from Levent Alpöge, a legitimate mathematician, though I can't fully authenticate the account from an image alone. But the mathematics checks out: I've confirmed the Jacobian determinant is exactly -2, the three points all map to (-1/4, 0, 0), and a non-injective polynomial map with constant nonzero Jacobian would ind…
1190
Kate Sills @katelynsills.com · 20/07/2026
so Claude does a websearch to see if anyone else has caught this yet.. what is he missing?
The math checks out exactly as the tweet claims. Let me confirm what this would mean and check whether it's real news, since this is dated after my knowledge cutoff.

Scrutinizing whether this constitutes an elaborate mathematical hoax
Scrutinizing whether this constitutes an elaborate mathematical hoax
Jacobian conjecture false counterexample 2026 Alpöge

10 results

a note on sufficient conditions for the real jacobian ...
ejde.math.txstate.edu
Jacobian conjecture as a problem on integral points on affine curves
arxiv.org
The bunkbed conjecture is false
arxiv.org
AN INTRODUCTION TO THE JACOBIAN CONJECTURE WILLIAM GARLAND
math.uchicago.edu
Some algorithms related to the Jacobian Conjecture
arxiv.org
A note on the plane Jacobian conjecture
arxiv.org
An Optimization Approach to Jacobian Conjecture
arxiv.org
On the Jacobian conjecture - ScienceDirect
www.sciencedirect.com
On the shape of possible counterexamples to the Jacobian Conjecture | Request PDF
www.researchgate.net
Counterexamples to a Conjecture of Ahmadi and Shparlinski
arxiv.org

I found confirmation that the Jacobian conjecture remains open—a polynomial map with constant nonzero Jacobian should be injective, but the tweet's construction claims to show a counterexample where a map with Jacobian determinant -2 sends three distinct points to the same point, which would directly contradict the conjecture if true.

Let me verify I transcribed the polynomials correctly from the image and double-check my calculation, especially the third component, since this would be an extraordinary result if the math holds up. I've verified the arithmetic with exact rational computation, so the three distinct points genuinely map to the same output with a constant Jacobian determinant. The real question is whether this constitutes a legitimate counterexample to the Jacobian conjecture that the mathematical community would accept, which depends on factors beyond my knowledge cutoff.

It's striking that such a simple degree-4 ma…
2160
Kate Sills @katelynsills.com · 20/07/2026
self-doubt sets in
Let me think through whether I'm misunderstanding the conjecture itself or if there's something subtle about this construction I'm overlooking.

The Jacobian conjecture being open means no counterexample has been found despite decades of effort, which makes me suspicious that this simple degree-4 polynomial construction actually works. If it did, the problem would've been settled long ago. So either I'm making an error in my verification, or there's a hidden flaw in the reasoning that I'm not catching yet. So this is genuinely a counterexample to the Jacobian conjecture — a non-injective polynomial map with constant nonzero Jacobian. Given how famous this problem is, that would be revolutionary. But I'm suspicious this might be a joke, especially noticing the reference to "Fable" in the context, which seems like it could be a playful nod to one of Anthropic's models.

Still, the mathematics checks out regardless of the framing. So either this is a real counterexample or there's something I'm missing about why it doesn't actually violate the conjecture. Let me think through whether there's a catch I haven't spotted yet. I'm wondering if this might actually be a well-known joke in the mathematical community, or if there's something I'm missing about the construction itself. Let me think through whether the Jacobian is truly constant and the map genuinely non-injective, or if there's a subtle error in my verification that would resolve the paradox. a very cleverly constructed hoax that actually works mathematically would just be a real counterexample, not a hoax at all. I can't fake something that survives direct verification — either the map has constant nonzero Jacobian and is non-injective, or it doesn't. The only remaining angle I can think of is whether the standard Jacobian conjecture has some requirement I'm missing, but I should just search to settle this.

Actually, let me step back. I've verified the Jacobian determinant is constant and nonzero, and I've conf…
1172
Kate Sills @katelynsills.com · 20/07/2026
more checking:
The Jacobian determinant is indeed exactly 
−
2
−2, a nonzero constant. Now the injectivity claim:

Scrutinizing whether this counterexample genuinely undermines the conjecture
Scrutinizing whether this counterexample genuinely undermines the conjecture
Check where the three points map

bash

python3 << 'EOF'
import sympy as sp
x, y, z = sp.symbols('x y z')
P = (1+x*y)**3 * z + y**2 * (1+x*y) * (4+3*x*y)
Q = y + 3*x*(1+x*y)**2 * z + 3*x*y**2 * (4+3*x*y)
R = 2*x - 3*x**2*y - x**3*z

pts = [(0,0,sp.Rational(-1,4)),
       (1,sp.Rational(-3,2),sp.Rational(13,2)),
       (-1,sp.Rational(3,2),sp.Rational(13,2))]

for p in pts:
    sub = {x:p[0], y:p[1], z:p[2]}
    img = (sp.simplify(P.subs(sub)), sp.simplify(Q.subs(sub)), sp.simplify(R.subs(sub)))
    print(f"F{p} = {img}")

# Confirm the three source points are distinct
print("Distinct source points:", len(set(pts))==3)
EOF

Output

F(0, 0, -1/4) = (-1/4, 0, 0)
F(1, -3/2, 13/2) = (-1/4, 0, 0)
F(-1, 3/2, 13/2) = (-1/4, 0, 0)
Distinct source points: True


The Jacobian determinant is indeed 
−
2
−2, a nonzero constant, and I've verified that three distinct points all map to the same output 
(
−
1
/
4
,
0
,
0
)
(−1/4,0,0). This means the map isn't injective, yet it has a constant nonzero Jacobian everywhere. Since a polynomial automorphism must be bijective, this polynomial map from 
𝐶
3
C
3
 to 
𝐶
3
C
3
 cannot be an automorphism.

If this construction is correct, it would be a counterexample to the Jacobian conjecture — one of the most famous open problems in mathematics. But this seems too significant to be overlooked, so I should verify whether this is actually a known result or if there's something I'm missing. eed to be a counterexample to the Jacobian conjecture. But wait — there's a classical theorem that for polynomial maps from 
𝐶
𝑛
C
n
 to itself, injectivity actually forces bijectivity and polynomial automorphism structure. So the question becomes whether a non-injective polynomial map can even have a nonz…
1140
Kate Sills @katelynsills.com · 20/07/2026
Claude wrote code to verify and then ran it:
This is a specific, checkable mathematical claim, so let me actually verify it rather than guess. Let me compute the Jacobian determinant and check where those three points map.

Verified mathematical calculations and mapped point convergence
Verified mathematical calculations and mapped point convergence
Compute Jacobian determinant of the claimed map

bash

pip install sympy --break-system-packages -q 2>/dev/null; python3 << 'EOF'
import sympy as sp

x, y, z = sp.symbols('x y z')

# The map from the tweet
P = (1+x*y)**3 * z + y**2 * (1+x*y) * (4+3*x*y)
Q = y + 3*x*(1+x*y)**2 * z + 3*x*y**2 * (4+3*x*y)
R = 2*x - 3*x**2*y - x**3*z

F = sp.Matrix([P, Q, R])
vars = sp.Matrix([x, y, z])

J = F.jacobian(vars)
detJ = sp.expand(J.det())
print("Jacobian determinant:", detJ)
print("Simplified:", sp.simplify(detJ))
EOF

Output

Jacobian determinant: -2
Simplified: -2


The Jacobian determinant came out to 
−
2
−2, which is nonzero, so now I need to verify that these three points all map to the same location.

Done

The Jacobian determinant is indeed exactly 
−
2
−2, a nonzero constant. Now the injectivity claim:

Scrutinizing whether this counterexample genuinely undermines the conjecture
1140
Kate Sills @katelynsills.com · 20/07/2026
I did a "gork is this true" and the results were pretty incredible to my untrained eye (I'm not a mathematician) (cont'd)
Me: is this real?


Opus 4.8: Scrutinized mathematical claim and planned computational verification
Scrutinized mathematical claim and planned computational verification
I'm looking at a claim about the Jacobian conjecture being disproven. The conjecture is a major open problem stating that if a polynomial map from complex n-space to itself has a nonzero constant Jacobian determinant, then it must have a polynomial inverse. The tweet apparently presents a specific counterexample to this, though the thought cuts off before showing what that counterexample is. I should verify this claim computationally using symbolic math tools rather than just reasoning about it abstractly. Let me set up a calculation to check both the Jacobian determinant and whether the map actually sends multiple points to the same output, then search for context about whether this has been reported as a breakthrough result. The Jacobian conjecture requires a nonzero constant Jacobian determinant, so if this map truly has constant Jacobian 
−
2
−2 but isn't injective, it would be a genuine counterexample. I'm skeptical though — this could be a hoax or contain a computational error, especially since the account appears to be associated with Levent Alpöge, a legitimate number theorist.

Let me verify the math directly using symbolic computation...

Show less
1140
Kate Sills @katelynsills.com · 19/07/2026
Yes, there is a structural problem: we need to stop relying on manual checking and start automating. I mean, this as current practice is ridiculous:
The mechanism driving the lag is straightforward. Modern reasoning models can now spend enormous inference-time effort exploring a problem - the test-time compute that has reshaped what these systems can attempt - and they can generate candidate solutions in hours or days. Formal verification, by contrast, is a slow human process: a mathematician or physicist has to read the argument, find the load-bearing steps, check them, and often reconstruct the reasoning to be sure it holds. Generation has sped up by orders of magnitude; verification has not. So the two have come unglued in time, and for a window, claims run ahead of confirmation.
200
Kate Sills @katelynsills.com · 19/07/2026
* it's happening! *
Markets Not Capitalism Anthology edited by Gary Chartier & Charles W. Johnson
1143
Kate Sills @katelynsills.com · 11/07/2026
It's very funny when Claude evaluates its own work, especially if it's really good or bad. Winnie the Pooh vibes.
"The live panel renders beautifully in-page"
150
Kate Sills @katelynsills.com · 05/07/2026
Now you're changing what you're saying without admitting you were wrong. Here's where we started:
And why you can't trust an LLM to summarize logbooks. You need something dumber like basic stats: standard deviation, average, median. Something doesn't get hallucinated probabilistic results randomly instead giving the same result every time you rerun.







‪Kate Sills‬
 ‪@katelynsills.com‬
· 1h
what do you think an LLM uses to summarize logbooks? It uses basic stats. It uses tools. It uses code.







‪economymedicine.bsky.social‬
 ‪@economymedicine.bsky.social‬
· 1h
Moving over a manifold to predict what to do is going to lead to it predicting log book values or inconsistently using various statistical tools. It is not a reliable way to analyze that sort of data. They trying having overrides in the models but they are not trustable.







‪Kate Sills‬
 ‪@katelynsills.com‬
· 1h
This exact question is testable right now, with very little resources needed. Given this exact problem, will Claude Code Opus 4.8 write deterministic code to get the right answer? And how often? 

Have you done this test? Why not?
100