2026年8月16日 技术热点总结

项目简介：Two-phase DeepSeek Harness preset: Minimal-aligned bootstrap, then full Standard tools (Project2 98/99)

技术栈：JavaScript、deepseek、deepseek-harness、dsh-plugin、llm-agent

核心介绍：中文说明 Experimental DeepSeek Harness agent presets — one base mode plus two variants (real Minimal tool schema, no auto-injected context), then promote to a small resident catalog once the session is durable, unlocking heavier Standard tools This is a community project. It is not an official DeepSeek preset and is not affiliated with or endorsed by DeepSeek. Every mode directory is self-contained and insta…

项目简介：dsh-routing-suite — injector + router-standard kit: install the runtime injector first, then the task-aware reasoning-mode router preset (measured P1-P23).

技术栈：PowerShell

核心介绍：一个仓库装齐「运行时手术台 + 思维模式路由预设」：先装注入器（免重启运行时管理层）， 再用它装配 router-standard 预设（任务感知思维模式路由，P1-P23 实测）。 中文 | English

关键特性：三行为带 + weak 内路由：spec（计划-集体）/ react（执行者）/ mixed（陷阱，回避）/ weak（模型自分类）；按模型选 persona：Pro=spec 句+few-shot（区分度 +5.0）；Flash=neutral+classify（+5.7）；近距离引导：每轮用户消息后注入固定引导（缓存 92-94% 命中），路由 96% + 收敛 100% + 反稀释；单任务三锚（persona 静态）：回顾 + 收敛 + 反跑题 —— 开放任务完成率 0% → 100%；plan-mode 保留：只替换 p…

项目简介：Opinionated Oxlint rules for rejecting low-evidence TypeScript and JavaScript patterns

技术栈：TypeScript、agent-skills、linting、oxlint、typescript

核心介绍：Opinionated Oxlint rules that reject low-evidence and low-signal TypeScript and JavaScript patterns. This project is meant to be vendored, not treated as a fixed npm dependency. Copy the rules into your repository, read them, and change them to match your team’s standards. The bundled agent skill handles the initial copy and configuration; after that, the vendored files are yours to maintain and make your own.

项目数据：⭐ 1,243 …

We introduce , a recurrent Transformer architecture with fixed-size memory that generalizes sliding-window attention while remaining parallelizable during training. consists of two coupled models: a prefiller Q, which leverages full attentionIn practice, we use interleaved full and sliding-window attention for Q, as this yields stronger performance. The essential requirement is that Q be more expressive than P, with access to the full history. to produce memory targets m'_t, and a decoder P, which uses only sliding-window attention and recurrent K/V injection to produce decoder memories m_…

Test-time compute scaling is a primary driver of performance in large reasoning models (LRMs), but extreme inefficiency bounds current approaches, shifting the critical question from how much compute to spend, to where to allocate it. We formalize test-time reasoning as a constrained compute allocation problem over partial trajectories. Under a fixed hardware budget, existing paradigms fail to actively allocate the compute to the most promising partial progress: traditional parallel sampling treats traces independently and induces severe memory bottlenecks, while subtractive pruning starves…

Rib fractures are common and time-consuming to localize on computed tomography (CT). We ask whether fractures detected independently in two orthogonal CT-derived projections (anteroposterior and lateral) can be paired across views and triangulated into reliable 3D points at a controlled rate of false outputs, and we answer it with a staged diagnostic study. The projection geometry is exact, and given correct correspondence, localization is accurate (median 4.0 mm, 88% within 10 mm, 93.6% rib-exact). On a sealed 55-case cohort, a large share of fractures is in principle recoverable (61.1% du…

Interactive autoregressive video generation demands both low-latency rollouts and precise online control. Few-step distillation accelerates generation by reducing denoising steps, while online control imposes a causal constraint: frames and blocks should depend on history and controls available during generation. Existing video distribution matching distillation (DMD) pipelines, however, often supervise causal few-step students using bidirectional teachers that score complete clips. The score for a target can therefore depend on future frames and controls that were unavailable when the stud…

Large audio-language models (LALMs) have demonstrated strong capabilities in understanding diverse audio inputs. This diversity includes low-frequency signals that are inaudible to humans but can still enter the model and influence its generation. However, the practical impact of such low-frequency inputs on LALMs remains largely unexplored. In this paper, we propose Intermittent Low-Frequency Lockout (ILL), an inaudible red teaming method that evaluates this risk using a universal waveform template in a black box setting. ILL uses Sentence Attention Scale Estimation to determine active int…

Machine translation (MT) systems often fail to correctly translate gender, especially when converting from a gender-neutral language like English to a gendered target language such as Romanian. This bias results in translations that default to masculine forms or reinforce gender stereotypes. We propose a hybrid pipeline to mitigate this issue by combining large language model (LLM)-based gender classification with neural machine translation (NMT). Our system uses a fine-tuned LLM to detect the intended gender of target words in English sentences and insert inline gender hint tags. These tag…