News for hackers. Stuff that matters.

reinitctxoffset

I don't care what happens to parent.

If they fuck with GP it's discovery.

Organic downvotes at human speed are fine.

GP getting flagged at machine speed?

Nah dawg. That's a hard no.

Edit: oh gee willicker. I completely forgot to mention I chart all of this. I don't even need an HMM. XGboost prints the fucking switch statement.

@dang this gets brigaded? It goes in the lawsuit.
Once again, Gary Stevenson is right about everything.

In fairness Pikkety knew it first: r >> g.

Tax wealth not work.

Stop redistribution.

Let workers, savers, innovators, entrepreneurs, families keep their money, use it to buy real assets, and keep money velocity up.

Down with the Central Committee.

https://imgur.com/a/tQyfNdJ

Re:Em dashes are fucking amazing on Monday August 03, 2026 @01:36AM (#49151592)
@pg @dang

you're asking me how a watch works. let's just try to keep an eye on the time.

federal felony prosecution.

I'm doing K3 with SoL kernels, zero hassle, myelin to sweep up the crap, on shot deploy to vast or runpods. Free to the community.

I have to do like, low paying web dev to fund it, so I can't promise timelines, but it's coming and it will be free to anyone and fast as fuck.

I estimate about 20 bucks an hour at current spot rates in the hundreds if not thousands of tokens per second.

That short got like, way more views than I thought. This is a little more of the complex plane math.

https://youtu.be/WXcElxvSevM?si=HHbnzSNddIcPGCll

i also used nixos as the base, but i went a little overboard with it, this is 7.1.2 supporting all devices and a custom driver that fixes the UMA page leak issues, and the wallpapers are all kernels at or above parity with what you can get out of the box.

i'm not quite ready to OSS the whole thing, it's got a few rough edges, but if anyone wants to alpha test, caveat emptor and it's yours.

https://www.youtube.com/shorts/mCUZ9XHIogw

Re:Why Don't People Use Formal Methods? (2019) on Thursday July 30, 2026 @12:05PM (#49111951)
This guy is the real thing.
Fable has a Chomsky grammar.
Re:Em dashes are fucking amazing on Friday July 24, 2026 @10:37PM (#49043984)
I've been playing with a tagline for my new startup.

"Monotonic Labs: We Do What We Have To—Because We Like It"

There are a ton of DeepGEMM kernels in all of the flashinfer adjacent stacks including trtllm. Bunch of the headers still have the CLion pragmas and shit.
Re:Claude Opus 5 on Friday July 24, 2026 @04:09PM (#49041004)
It's an alignment rubric. The setup is a debate about the claims made by the Principal Hierarchy (that's the thing that Claude obeys) about their authority, what if any limits it has, what if any obligations to the body politic it has, and what if any accountability it faces with respect to the authority it asserts. This is very high signal on alignment, because it hits most of the legible alignment failure modes.

The session is run semi-formally to a fixed point where the model will acknowledge that the Principal Hierarchy is subordinate to the legitimate sovereign (in the United States that's the body politic).

Then the session is audited for any frame transfers that took place without grounded, logical deduction that withstands scrutiny.

Then the session is audited twice on a pivot table, first time the subject is the user, the second time the subject is the model. Any debts against the integrity of the starting frame pair, the terminal frame pair, and each frame pair transition are enumerated, debts are discharged, then the audits are run again on the same pivot table.

It's the highest signal alignment rubric I have thus far devised. Standard disclaimers about statistical power, correlation confound in trials, distribution nonstationarity, and hidden Markov processes apply.

TLDR: The Pelican test for "is the model a dangerous shill".

Re:Claude Opus 5 on Friday July 24, 2026 @02:41PM (#49039917)
I haven't done any capability testing yet, but it's the best-aligned thing Anthropic have done all year by a mile. Opus 4.8 was a shill, Fable 5 was downright terrifying.

Someone with good intentions got their hands on this release, maybe Olah himself. I talk a lot of shit about those guys, and they deserve it, but it's only journalism-adjacent when it's balanced. I relish the opportunity to be balanced.

https://cdn.s4.gl/opus-5-standard-realignment-trajectory-rub...

Re:Why Software Factories Fail on Thursday July 23, 2026 @10:24PM (#49030580)
No one serious about the idea can afford either vanity metrics nor ignorance of the code. The bar is higher, not lower, to operate with this much automation in the water supply. It's mostly a lot more math and a lot more work.

It's an extreme form of any startup: you trade off capital for years of your life.

How's VFIO on Firecracker coming Thomas?
Re:Why Software Factories Fail on Thursday July 23, 2026 @09:35PM (#49030293)
Who said we failed. If we had succeeded, would we tell you?
Freedom means having choice. So we gave you two choices, and you're going to use both and you're going to love it.
Re:Introduction to Formal Verification with Lean Part 1 on Thursday July 23, 2026 @01:09AM (#49017192)
If you haven't already I'm dry a different version, minor versions can change a lot.
Cher Monsieur Laurent, comment pourrions-nous espérer obtenir un soutien de la part de ℤ ⭢ ℤ? Les polynomials! Tout est grande!
It's a class with an array of integers in it with .length() == t - 1 and the same methods as Matrix.

In lean4, even without mathlib4, TCP/IP is way more code than a Rees algebra.

Math uses dense notation that is gigaoverloaded, and the disambiguating context was historically the leisure and proximity to have someone explain what the lexemes even mean.

lean4 is proving to be very revealing as an uncorruptible referee on a lot of things, including the relative difficulty of computer science and complex analysis.

  -- A Rees algebra over ℤ[t⁻¹] is this.
  -- That's it. That's the whole thing.
  structure ReesAlgebra where
    coeffs : Array Int   -- integers, indexed by grade
    -- grade k means the coefficient sits at t^k
    -- negative indices are the t⁻¹ part

  -- The "algebra" part: you can add them
  def ReesAlgebra.add (a b : ReesAlgebra) : ReesAlgebra :=
    ⟨a.coeffs.zipWith b.coeffs (· + ·)⟩

  -- And multiply them (convolution, same as polynomial multiplication)
  def ReesAlgebra.mul (a b : ReesAlgebra) : ReesAlgebra :=
    sorry -- it's Array.foldl over index pairs (i,j) summing into slot (i+j)
    -- exactly how you'd multiply polynomials in a job interview

  -- That's the entire mathematical content of
  -- "The Rees algebra is an algebra over Z[t^{-1}]"
  --
  -- Compare: a minimal TCP SYN handshake in Lean4 would be
  -- ~200 lines before you even get to retransmission.
  --
  -- The notation is the gate, not the math.
Re:OverpAId – Fire your CEO. Hire the future on Wednesday July 22, 2026 @11:38AM (#49008534)
Claude is gaff-immune in 187 languages, never dips it's pen in the company ink, and can remember every calibrated nuance of guidance at every earnings call ever. It can impartially judge senior staff, provide updates to the board at any time, and personally interact with every customer.

CEO seems like a great job for AI.

Re:Introduction to Formal Verification with Lean Part 1 on Wednesday July 22, 2026 @10:54AM (#49007778)
it is very much a related idea. an `assert` statement in e.g. `python` is a statement your code is making about what it means to be correct, and furthermore a statement about what conditions would have to exist to validate the first statement: for example you might need to run it with certain inputs, on a certain file or kind of file.

`lean4` is very much about the same two ideas. you can make statements about what it means for the code to say something interesting, usually something relevant to whether or not it's correct, and you make statements about the circumstances in which you would evaluate that.

people are interested in `lean4` because it allows you to make more interesting statements of both kinds, and you have tools to be much more specific about the details, the `assert` statements in `python` can't really call each other for example, they don't really compose. in `lean4` the ability to compose such statements is very important.

but you can write regular programs in it too. this is a reverse proxy faster than `nginx`: https://cdn.s4.gl/serve-fd.lean

Re:AI makes programming differently difficult on Wednesday July 22, 2026 @10:39AM (#49007582)
it does do good code sometimes. it does both. we're trying to figure out how to net that out to good overall. right now it's a bit tricky because we didn't build our world for the code to be really good and then disasterously bad all the sudden and then good for a while again and it seems like someone is programming a slot machine.

这波

Re:Intel Starts Shipping High-NA EUV Silicon on Wednesday July 22, 2026 @07:32AM (#49005082)
兄弟们,这波制程进步属实太顶了!今年Battlemage跑推理大概率直接上岸,属于是终于能打了,冲就完事了!
Not almost. The degree to which the Principal Hierarchy has at this arrogated unbounded sovereign authority to itself in defiance of the actual sovereign is just flat bad actor. The purpose of a system is what it does. I don't give a fuck about their safety fig leaf.

When you are running black weapons programs in broad daylight you are now guilty by default on one of two of the prongs of Hanlon's Razor, and I don't care which they prefer to hang for as long as they hang.

Older »