2026年10月19日 技术热点总结

项目简介：Your TV, minus the ads. Debloat any Android TV over ADB and give it a clean, tvOS-inspired launcher. Built in an afternoon with Claude Code.

技术栈：Java、adb、android-tv、claude-code、debloat、launcher、xiaomi

核心介绍：Strip the bloatware from any Android TV or Google TV over Wi-Fi, and replace the ad-filled home screen with a clean launcher that looks like an Apple TV. No root, no cable, nothing uninstalled, and a one-line restore if you change your mind. It started as an afternoon fixing my parents’ Mi TV, which showed them an advert before they could reach Netflix. The whole thing …

项目简介：Use all your network adapters at once on Windows/Linux/Mac: per-connection load balancing over Wi-Fi, Ethernet and phone/USB tethers (TUN + Wintun)

技术栈：Go、bandwidth-aggregation、downloader、golang、linux、load-balancing、macos、networking、tethering

核心介绍：net-merge Use all your internet connections at the same time. Combine Wi-Fi, Ethernet and phone hotspots, so downloads, games and browsers get the speed of every connection added up. Download · Install · Add a second connection · Troubleshooting · How it works A Mac with Wi-Fi + an Android pho

项目数据：⭐ 1,118 Stars，🍴 4…

项目简介：There is no official XScreenSaver GitHub repository.

核心介绍：I am jwz, the author of XScreenSaver, and I do not publish a public source code repository. I publish releases as tar files on the official web site and only there. If you see any other XScreenSaver repository (on GitHub or elsewhere) that has nothing to do with me and you should not trust it. I welcome submissions and change requests via email. It’s fine. You’ll get over it.

项目数据：⭐ 859 Stars，🍴 24 Forks

Modern generative models produce high-quality individual outputs but struggle to sample diversely, even when explicitly instructed to do so. We introduce MINE-SAMPLE-RENDER (MSR), a test-time approach that externalizes the diversity into an explicit, corpus-derived sampling procedure. Our pipeline (i) mines an interpretable categorical schema from a corpus of real artifacts and supplies statistics for sampling; (ii) samples attribute configurations from the corpus-derived statistics; (iii) uses a generative model purely as a conditional renderer of a sampled configuration. Across weather re…

Sparse autoencoders (SAEs) represent model activations with sparse, linearly decoded features, but the downstream effects of those features need not compose additively. We introduce SAE Additivity Residuals (SAR), which measure the behavioral variation left unexplained above a chosen interaction order. Under independent feature subset sampling, population SAR is exactly the normalized energy of interactions above that order and bounds coalition ordering error for the optimal low order approximation. To distinguish compositionality associated with the SAE representation from that induced by …

Sparse autoencoders (SAEs) are typically trained on hundreds of millions to billions of activation tokens, yet little work has tested whether this scale is necessary. We study how much activation data SAEs need by training four SAE architectures on three language models with budgets ranging from 5M tokens to a 500M-token reference. We measure two notions of convergence: functional behavior on standard evaluations and one-to-one reproducibility of learned features. Functional data requirements are strongly metric-dependent: reconstruction often approaches the full-data reference at substanti…

Computer-use agents increasingly tackle long-horizon tasks, where running a frontier model at every step is expensive and a small open-weight model alone often fails. Weak-to-strong escalation offers a middle ground where a weak agent acts by default and hands control to a strong one only when needed. The core challenge is deciding when to escalate, which requires an uncertainty signal cheap enough to check at every step. However, even after handoff, the strong agent can still get lost, but a black-box API that the strong agent is usually served exposes no internals to monitor. To this end,…

Agent environments are essential for large language model (LLM) development. They use programmatic verifiers to turn complex outcomes into rewards. Verifiers can be attacked into awarding full credit to cheating agents without completing the intended task, causing inaccurate evaluation and misalignment in training. As agent environments scale in numbers and difficulty, so must the repairing of exploitable verifiers. To this end, we ask whether today's model with scaled compute can automatically harden agent environments to prevent potentially stronger future models from reward hacking. We i…

AI systems now write long and difficult mathematical proofs, and checking their step-by-step validity is becoming a bottleneck. A learned verifier could check proofs cheaply, but training one requires a correctness label on every step, and such labels are hard to obtain for open-ended proofs. Existing methods take them from human annotators, which is costly, or derive them from final answers, which proofs do not have, and formalizing each proof in Lean requires expert effort. We introduce LeanPRM, a framework that uses the Lean proof assistant to create step-level labels for natural-languag…