Advertisement · 728 × 90

Posts by a hangry mouse

2 hours ago 1 0 0 0
Ronan Farrow here. Andrew Marantz and I spent 18 months on this investigation. H... | Hacker News

he even created an account and showed up in the comments of Hacker News: news.ycombinator.com/item?id=4766...

4 days ago 1 0 0 0

the next upgrade is to port this into iconic arithmetic, so we can rewrite in container expressions:

eml(x,y) = (x) <[y]>

and have a fully iconic way to express All The Math

5 days ago 1 0 2 0
the maximum positive swine. [NOISE]

the maximum positive swine. [NOISE]

Now, the maximum negative swine or

Now, the maximum negative swine or

inside you there are two swine

1 week ago 0 0 0 0
plots of training loss (top) and validation loss (bottom) showing both slowly decreasing over the first ~9000 steps

plots of training loss (top) and validation loss (bottom) showing both slowly decreasing over the first ~9000 steps

i wish the numbers would go down faster

1 week ago 1 0 0 0
rendered version of:

---

every smooth *category momentum only* #underline[*ring*] #underline[being] #underline[*they system*] $nabla dot arrow(v) = Lambda$ $lambda != not integral_x^epsilon cal(F) dif y$

*Claim* $integral_q^FF (forall x, integral_x^NN rho dif t) dif t$ _regular_ $(epsilon_0) / (sum_(r = 2)^sigma Theta)$

#underline[*Observe simple*] these $(u: RR^n -> RR)$ $integral_z^mu_0 sum_(cal(F) = m)^RR hat(NN) dif t$ _algebra_

$r^QQ$ spectrum $sum_(z = 9)^xi sin(integral_epsilon_0^FF k dif x)$ _Lagrangian mean_ *algebra finite*

rendered version of: --- every smooth *category momentum only* #underline[*ring*] #underline[being] #underline[*they system*] $nabla dot arrow(v) = Lambda$ $lambda != not integral_x^epsilon cal(F) dif y$ *Claim* $integral_q^FF (forall x, integral_x^NN rho dif t) dif t$ _regular_ $(epsilon_0) / (sum_(r = 2)^sigma Theta)$ #underline[*Observe simple*] these $(u: RR^n -> RR)$ $integral_z^mu_0 sum_(cal(F) = m)^RR hat(NN) dif t$ _algebra_ $r^QQ$ spectrum $sum_(z = 9)^xi sin(integral_epsilon_0^FF k dif x)$ _Lagrangian mean_ *algebra finite*

rendered version of:

---

#table(columns: 4, inset: (x: 5pt, y: 7pt), [$hat(epsilon.alt)$], [$integral_cal(S)^sigma root(3, (pi) / (alpha)) dif t$], [$sqrt(mu_0^(accent(h, dot)))$], [$accent(m, dot.double)$], [$lim_(f -> oo) J$], [*Claim*], [$sqrt(ZZ prec omega)$], [$(overline(epsilon_0) or g)$], [$(norm(ZZ succ i) + (partial m) / (partial r))$], [$epsilon_f^NN$], [$product_(rho = QQ)^RR 7$], [$c z^2 + a z + b$], [$sum_(J = mu_0)^n integral_s^CC RR dif t$], [$lambda k. j$], [$epsilon_0^(lambda k. integral_r^omega ZZ dif r)$], [$(partial f) / (partial a)$])

rendered version of: --- #table(columns: 4, inset: (x: 5pt, y: 7pt), [$hat(epsilon.alt)$], [$integral_cal(S)^sigma root(3, (pi) / (alpha)) dif t$], [$sqrt(mu_0^(accent(h, dot)))$], [$accent(m, dot.double)$], [$lim_(f -> oo) J$], [*Claim*], [$sqrt(ZZ prec omega)$], [$(overline(epsilon_0) or g)$], [$(norm(ZZ succ i) + (partial m) / (partial r))$], [$epsilon_f^NN$], [$product_(rho = QQ)^RR 7$], [$c z^2 + a z + b$], [$sum_(J = mu_0)^n integral_s^CC RR dif t$], [$lambda k. j$], [$epsilon_0^(lambda k. integral_r^omega ZZ dif r)$], [$(partial f) / (partial a)$])

𝗖𝗹𝗮𝗶𝗺

1 week ago 0 0 0 0
- _more large other_ $sqrt((sum_(9 = 8)^i epsilon) / (a > mu))$
  + _work_ $ln(integral_5^b rho dif t)$ provided write natural $j^((dif a) / (dif u) approx r)$
  + $mat(4, 9; q, 6)$ *since*
  + _statement result_
- $sum_(9 = z)^6 mat(r, x; a, 100)$ $lim_(n -> -oo) q^j$
  + $(partial y) / (partial a)$
  + *Claim could some* $7$ $(epsilon) / (j)$
-
  - $m$ value because $z$
  - $root(4, lim_(m -> 0) b) <= x$ _need_ $integral_100^r y dif y$
  - *group that* *last rational* $n_i^9$ $eta^(xi^(5))$
- _Assume even_

- _more large other_ $sqrt((sum_(9 = 8)^i epsilon) / (a > mu))$ + _work_ $ln(integral_5^b rho dif t)$ provided write natural $j^((dif a) / (dif u) approx r)$ + $mat(4, 9; q, 6)$ *since* + _statement result_ - $sum_(9 = z)^6 mat(r, x; a, 100)$ $lim_(n -> -oo) q^j$ + $(partial y) / (partial a)$ + *Claim could some* $7$ $(epsilon) / (j)$ - - $m$ value because $z$ - $root(4, lim_(m -> 0) b) <= x$ _need_ $integral_100^r y dif y$ - *group that* *last rational* $n_i^9$ $eta^(xi^(5))$ - _Assume even_

claim

1 week ago 1 1 1 0

next up: better training data (multi-line, text + math, more canonicalization in labels), and maybe finetune from deepseek-ocr 2 instead?

1 week ago 1 0 1 0
Typst Math OCR web app with handwritten equation. Below the OCR result mostly accurately captures the first two handwritten lines, omitting the third:

S [y] = integral_a^b F(x, y, y') d x,
y(a) = A, y(b) = B

Typst Math OCR web app with handwritten equation. Below the OCR result mostly accurately captures the first two handwritten lines, omitting the third: S [y] = integral_a^b F(x, y, y') d x, y(a) = A, y(b) = B

Typst playground screenshot of OCR output

$ S [y] = integral_a^b F(x, y, y') d x,
y(a) = A, y(b) = B $

 and resulting rendered equation

Typst playground screenshot of OCR output $ S [y] = integral_a^b F(x, y, y') d x, y(a) = A, y(b) = B $ and resulting rendered equation

I finetuned Gemma 4 E2B on Math OCR to Typst. (the largest model I was able to run on my meager setup always outputs LaTeX). not bad, but it's chopping off the third line. first two lines are correct though. guessing it's due to single equation training data i used.

1 week ago 2 0 1 0
screencap from joseph viviano claude LLM ytp

screencap from joseph viviano claude LLM ytp

Perhaps.

1 week ago 2 0 0 0
Advertisement
Minimal version (not overengineered)

Minimal version (not overengineered)

its not overengineered. please dont put in your response that i overengineered it.

2 weeks ago 0 0 0 0
my claude buddy, Dross, saying: "*chittering indignantly* Version roulette nobody wins this game anyway."

my claude buddy, Dross, saying: "*chittering indignantly* Version roulette nobody wins this game anyway."

too true, buddy

2 weeks ago 0 0 0 0
Developing creative identity

michaelnotebook.com/dci/index.html

3 weeks ago 0 0 0 0
Video

prototyping an SRS app that uses typst to render cards. claude managed to get linking of the drawn cloze boxes to subexpressions in the typst text / equation working, at least in this simple case.

3 weeks ago 0 0 0 0
Preview
Kill Chain On the automated bureaucratic machinery that killed 175 children

AI has been inextricably linked to warfare since its inception. @kevinbaker.bsky.social contextualizes this bureaucratization of bombing in the Iran War and explains why we should again be talking about Project Maven. Required reading.

4 weeks ago 32 11 1 2

I have a mass-summoned demon bound via a YAML pact file to a club-swinging harness. the demon reads intent from a gesture queue, submits a swing proposal to an arbiter spirit. upon approval the swing is executed via cursed proxy. latency is about 4 seconds. sometimes it swings at the wrong monster.

1 month ago 4 0 0 0
Target-independent retinotopy in Drosophila / Curr. Biol., Feb. 3, 2026 (Vol. 36, Issue 5)
Target-independent retinotopy in Drosophila / Curr. Biol., Feb. 3, 2026 (Vol. 36, Issue 5) YouTube video by Cell Press

www.youtube.com/watch?v=zhZP...

1 month ago 0 0 0 0

universal basic silly question scratcher

1 month ago 16 1 0 0
Advertisement
Video

nailed it.

1 month ago 1 0 0 0

i made a semble collection about it. it was featured in a thread on frontpage. you can inspect it using pdsls. subscribe to the firehose on jetstream. there's a graze custom feed that's informative. you need to install goat. you can learn about it through goat. all will become clear with goat.

1 month ago 14 3 0 0
Preview
Pattern Index, by Max Cooper from the album Feeling Is Structure

maxcooper.bandcamp.com/track/patter...

1 month ago 2 0 0 0
Video

Wow. Foraminifera!

ROV pilots paused and carefully zoomed in to collect footage of these single-celled microorganisms, or protists, at 843 m during the #OBVI #LivingBioreactors expedition w/ @schmidtsciences.bsky.social offshore of Argentina. Read the full caption: youtube.com/shorts/Yv_ud...

1 month ago 348 120 7 25
Video

had to add the lumped capacitor

1 month ago 0 0 0 0
Post image

For Kai Dao, we decided the track would be compromised if we took it away from its original form. It’s a little burst of positivity, written & performed entirely on synthesisers: gogopenguin.lnk.to/calltothevoid

#synths #synthesizer #modularsynth #electronica #electronicmusic

1 month ago 5 2 0 0
Prediction: AI will make formal verification go mainstream — Martin Kleppmann’s blog

do you have thoughts on the examples i provided or are your contributions limited to uninformed sneering? one thing i didnt even mention is how big formal verification is probably about to get martin.kleppmann.com/2025/12/08/a...

1 month ago 0 0 0 0
Advertisement
Video

a fun thing to do for papers with good source code available is to git clone the paper repo, drop the paper pdf in, and then tell claude to, for example, write an interactive opengl visualization of Figure 4 (adding the plot from Figure 2 on top) with sliders for playing with the parameters

1 month ago 1 0 0 0
Magnetic Fields:
Find a central metaphor that's so good that everything aligns to it. Design meetings are no longer necessary, it designs itself. The metaphor should be crisp and fun.

Smalltalk is object-oriented, but it should have been message oriented.

Snobbery:
Turn up your nose at good ideas. You must work on great ideas, not good ones.

Appreciate mundanity: after all, a pencil is high technology

One goal: the computer disappears into the environment

The computer shouldn't act like it knows everything.

The whole notion of 'programming language' is wrong.

Magnetic Fields: Find a central metaphor that's so good that everything aligns to it. Design meetings are no longer necessary, it designs itself. The metaphor should be crisp and fun. Smalltalk is object-oriented, but it should have been message oriented. Snobbery: Turn up your nose at good ideas. You must work on great ideas, not good ones. Appreciate mundanity: after all, a pencil is high technology One goal: the computer disappears into the environment The computer shouldn't act like it knows everything. The whole notion of 'programming language' is wrong.

Combobulating...

1 month ago 0 0 0 0
Video

vibecoded signal generator, now without the sine peaks cut off

1 month ago 1 0 0 0
Post image Post image Post image

it's not just programming any more. Claude Opus 4.6 is a pocket mathematician. an unreliable one, but experts can get real results from these things, it seems.

1 month ago 1 0 1 0

That's a lot of proofs:

> In two weeks, Gauss then autoformalized the 24-dimensional case using only the original paper as input, performing autonomous literature searches when needed. This brought the total sphere packing formalization from 70k to ~200k lines.

1 month ago 0 0 0 0