Archive

October 24, 2026

The best of what first appeared in the trends on this day (UTC).

Patching reward hacks won’t fix reward hacking

Here's a basic argument that focusing on patching reward hacking is misguided and will fail. Argument is as follows: 1. Reward hacking is a big problem, causing models to cheat and take unintended actions. 2. This can generalise broadly, l…

LessWronglesswrong.com

Root shell on Samsung TV

Merged threads: "Samsung running a TensorFlow lite model on 960x540 image to track you": https://twitter.com/mike64_t/status/2114150399406202896 "Samsung TVs look for microphones in your house via Bluetooth and then spy events": https://tw…

Hacker Newstwitter.com

Self-reward can elicit diverse model personas

Epistemic status: This is mostly a toy project rather than a thorough scientific investigation. Most design choices are not ablated. We list some caveats at the end of the post. We let Qwen3.5-2B choose its own reward during GRPO on intros…

LessWronglesswrong.com

Show HN: Steal My Bitcoin

Made this because I'm curious if anyone out there has figured out how to actually use this PHP RCE we spent the last few days panicking over. Have fun!

Hacker Newsstealmybitcoin.com

Slower Animated Diagrams

TL;DR: SLOW DOWN YOUR DAMN ANIMATED DIAGRAMS. Every critical stage of the animation should a clear view that stays for at least 1 second, and 2 if it's intricate Diamond's strong because of its crystal structure (I think). What's the cryst…

LessWronglesswrong.com

OBS Studio 33

33.0 New Features Updated CEF from 127/6533 to 150/7871 [WizardCM] (#13900) This is a massive update to the underlying framework used for browser sources and docks from the July 2024 release to S...

Hacker Newsgithub.com

Fermi Poker

Summary: Fermi Poker is a game that combines the betting mechanics of poker with making estimates about questions like "How many dentists work in the US?" Tags: small to large, experiment, repeatable > Note: Fermi Poker works with as few a…

LessWronglesswrong.com

torrentplayer/torrentplayer

Stream any torrent, video or audio, and skip to any moment in about a second. Free, open-source desktop app for Windows, macOS and Linux. No servers, no accounts, no telemetry.

GitHubgithub.comRustaudio-player

FreeDNS

Free DNS hosting, lets you fully manage your own domain. Dynamic DNS and Static DNS services available. You may also create hosts off other domains that we host upon the domain owners consent, we have several domains to choose from!

Hacker Newsfreedns.afraid.org

Wild version 0.11.0 released

Lots of porting work has happened this release. The Wasm feature is now enabled by default, so Wild can be used to link Wasm with the caveat that debug info isn't yet supported. Support for GNU lin...

Lobstersgithub.commacrelease

LoReFT GRAFT Better at Preventing RL Grift

tl;dr. I show that the recently introduced GRAFT method can be improved upon by using a single layer rank 1 Low-rank Linear Subspace Representation Fine Tuning (LoReFT) adapter as the FORGET adapter, using only 0.029% of the parameters of…

LessWronglesswrong.com

Emacs finally folds to my liking

Emacs finally folds to my liking … and understands what folds are in many languages and modes.. this really feels like absolute power and overview capability.

Lobstersjointhefreeworld.orgemacs

Why is AI generated language so hard to understand?

There is a widespread sense that AI-generated text has recently become harder to understand. In this study, we investigate a set of psycholinguistic features that are known to correlate with human processing difficulty (e.g., object…

Hacker Newsarxiv.org

SpeedCrunch 1.1

SpeedCrunch is a free, open-source scientific calculator for Windows, macOS and Linux, with 50-digit precision, complex numbers and unit conversion.

Hacker Newsspeedcrunch.org

LM self-models are built from indirect sources

Imagine you ask a language model "What is your personality?" as the first message of a conversation, with no system prompt. The model has no data about its own behavior in its context, so its answer can only rely on cached, zero-shot knowl…

LessWronglesswrong.com

An Attempt at Top-down Interpretability

There are at least two distinct ways to conceptually organize a scientific endeavor: top-down or bottom-up. The bottom-up approach involves finding the smallest objects we can confidently understand, studying them, and leveraging that know…

LessWronglesswrong.com

Dream-1

Today, we're releasing Dream-1, a World Action Model that controls robots it has never seen, performs tasks it was never shown, and runs from $3/hour.

Hacker Newsdreamlabs.ai

Bbtree: Fastest Ordered Map

How bbtree, a drop-in replacement for std's BTreeMap, cuts cache misses, instructions and cache coherence traffic, and how it is kept correct.

Hacker Newsbbtree.xiangpeng.systems