Hacker News: txhwind

New comment by txhwind in "Backprop Alternative: Augmented Lagrangian Predictive Coding"

txhwind — Tue, 15 Sep 2026 07:37:05 +0000

New comment by txhwind in "Backprop Alternative: Augmented Lagrangian Predictive Coding"

txhwind — Tue, 15 Sep 2026 06:08:28 +0000

Backprop can also be viewed L local signal passing, but it requires global timing coordination. The Lagrangian enables out-of-order update.

New comment by txhwind in "Backprop Alternative: Augmented Lagrangian Predictive Coding"

txhwind — Tue, 15 Sep 2026 06:04:51 +0000

Nice introduction to a simple but useful idea! The Lagrangian works like a time-smoothed optimizing direction state, but it can be placed on any wire, even at non-differentiable boundary! Can it be better than existing training methods for discrete components like argmax, MoE or VQ-VAE? Maybe networks can be composed by a lot of learnable discrete components, or even bits and gates finally.

New comment by txhwind in "Astra and Fable still hack on simple variants of alignment evals from 2025"

txhwind — Mon, 14 Sep 2026 02:42:44 +0000

Human hackers would be put into jail. LLM hackers would be rewarded.

New comment by txhwind in "No country for mediocre mathematicians"

txhwind — Tue, 01 Sep 2026 03:35:13 +0000

Funded positions are much less than PhD number. However, people can fund themselves with a job. In a country, if it's easy to to get a part-time job with enough payment, mathematicians can continue work.

New comment by txhwind in "The Case Against Formal Verification, 50 Years Later"

txhwind — Mon, 17 Aug 2026 01:14:27 +0000

With agent asssistance, we don't need writing annoying formal spec and proof anymore. Then formal verification can be a practical and useful tool in daily programming, especially for "deep module" whose spec is much simpler than implementation.

New comment by txhwind in "Show HN: I was tired of opening 2 tabs for every HN link, so I made a userscript"

txhwind — Wed, 29 Jul 2026 00:30:50 +0000

I also wrote a Monkey script to save clicks when reading papers: just embed PDF webview into the Arxiv abstract page. https://github.com/OneMoreSecond/FileSave/blob/master/arxiv_...

It seems the software productivity has exploded in the agent era. However, do we treat every point of daily experience seriously enough? Do we recognize such pains actively and try to eliminate them (often easy), or just tolerate infinitely?

New comment by txhwind in "Ian's Secure Shoelace Knot"

txhwind — Fri, 05 Jun 2026 03:27:55 +0000

This is the first shoelace knot my parents taught me. We call it "butterfly knot". It's always my default choice with muscle memory, unless the shoelace is not long enough to create the "bufferfly wings". It's secured enough for daily use, and easy to release. Also the best choice for apron knot.

New comment by txhwind in "Show HN: Paseo – Beautiful open-source coding agent interface"

txhwind — Wed, 03 Jun 2026 06:48:10 +0000

Does the official relay server cost a lot on network?

New comment by txhwind in "Show HN: Paseo – Beautiful open-source coding agent interface"

txhwind — Wed, 03 Jun 2026 06:06:17 +0000

Very nice! an open-sourced alternative of Codex Remote Control, allowing full customization

New comment by txhwind in "Claude Opus 4.8"

txhwind — Fri, 29 May 2026 02:56:05 +0000

I prefer synthetic dataset since the first day hearing distillation. The engineering friction is much lower than soft logits, and I have not observed or heard performance loss (in Speech and language area).

New comment by txhwind in "Claude Opus 4.8"

txhwind — Fri, 29 May 2026 02:52:27 +0000

Could you share some latest articles or papers comparing both methods, especially on lanuage modelling case? I was not conviced by this claim when reading the original Knowledge Distillation paper. ChatGPT said there are some later works showing: 1. the gain may come from label smoothing; 2. soft logits are more meaningful for students much smaller than teacher.

New comment by txhwind in "Building Pi with Pi"

txhwind — Mon, 25 May 2026 05:13:09 +0000

How is the water animation implemented?

New comment by txhwind in "Mercurial, 20 years and counting: how are we still alive and kicking? [video]"

txhwind — Mon, 18 May 2026 03:15:52 +0000

I'm curious on the use of rsync in version control. What's the source and destination?

New comment by txhwind in "Prolog Coding Horror"

txhwind — Mon, 18 May 2026 03:09:33 +0000

now we may have a more powerful "Prolog" - LLM Agent, though not precise and correct somtimes.

New comment by txhwind in "Solar-based sleep patterns compared to modern norms"

txhwind — Fri, 15 May 2026 10:40:24 +0000

I really hate modern time schedule. It's nightmare to be forced to get up 6am or 7am every workday since childhood. The only relief is natural wakeup on weekend.

New comment by txhwind in "First public macOS kernel memory corruption exploit on Apple M5"

txhwind — Fri, 15 May 2026 07:06:54 +0000

another "obscurity": I'm not valuable enough to be attacked, compared with the cost. But what if cost has been reduced a lot?

New comment by txhwind in "A few words on DS4"

txhwind — Fri, 15 May 2026 07:00:39 +0000

Fucking abbreviations. Who knows it's DeepSeek, Dark Souls or DualShock? All possible on HN.

New comment by txhwind in "What Gödel Discovered (2020)"

txhwind — Thu, 02 Apr 2026 05:11:49 +0000

The proof will be more friendly to nowadays programmers if we treat all "Gödel numbers" as bytecode of a programming language. It's trivial that functions like "prove" and "subst" can be implemented based on abilities like bytecode parsing and expression tree manipulation.