CO2 in rooms impairs decision-making
室内二氧化碳影响决策能力 ⭐️ 8.0/10

Elevated CO2 levels in indoor spaces can significantly impair cognitive function and decision-making, and simple CO2 monitors can increase awareness and prompt better ventilation. This matters because many people spend time in poorly ventilated offices, classrooms, and homes, often attributing mental fatigue to other causes; widespread use of CO2 monitors could lead to behavioral changes that boost productivity and health. CO2 levels can quickly rise to 2000 ppm in classrooms, and cognitive effects start around 1000 ppm; historical examples like Cornelis Drebbel's 1610 submarine CO2 management are also referenced.

hackernews · gslin · Jul 4, 06:32 · Discussion

Background: Carbon dioxide (CO2) is a natural byproduct of human respiration. In indoor spaces without adequate ventilation, CO2 accumulates. Research shows that concentrations above 1000 ppm can impair decision-making and cognitive performance. Affordable CO2 monitors now allow people to measure and respond to air quality in real time.

Discussion: Commenters share personal experiences: a teacher found classroom CO2 hit 2000 ppm, and some desire wearable monitors. However, one commenter questions the scientific rigor, noting that submarines operate at high ppm without reported cognitive issues.

Tags: #CO2, #indoor air quality, #cognitive performance, #ventilation, #health

Open Source AI Gap Map Launched
开源 AI 差距图发布 ⭐️ 8.0/10

Current AI, a non-profit founded at the AI Action Summit in Paris, launched the Open Source AI Gap Map v0.1, indexing 421 open source AI products across models, tools, datasets, and hardware. This map provides a structured, comprehensive overview of the open source AI ecosystem, helping developers and organizations identify gaps and opportunities. It is backed by $400 million in committed capital and released under an MIT license, making it a valuable community resource. The Gap Map includes 266 software tools, 85 models, 50 datasets, and 20 hardware projects from 228 organizations, organized into 14 categories across 3 stack layers. The underlying data (1,184 YAML files) and scripts are available on GitHub under an MIT license.

rss · Simon Willison · Jul 3, 22:04

Background: Current AI is a global non-profit partnership launched in February 2025 with $400 million in committed capital, aiming to build a public option for AI. A gap map is a tool that visually identifies areas where resources are lacking or abundant, helping guide investment and development efforts in a specific domain.

Tags: #open source, #AI, #ecosystem, #mapping, #resources

Cognitive Framework for Claude Fable 5: Navigating Unknowns
Claude Fable 5 认知框架：驾驭未知 ⭐️ 8.0/10

A cognitive framework based on the 'map is not territory' metaphor has been introduced for collaborating with Claude Fable 5, categorizing user knowledge into four types: known knowns, known unknowns, unknown knowns, and unknown unknowns. This framework shifts the bottleneck from model capability to the user's ability to clarify unknowns, providing a practical methodology that enhances the effectiveness of advanced LLMs like Claude Fable 5 in complex coding tasks. The framework includes specific techniques like 'blindspot pass' for unknown unknowns, rapid HTML prototypes for unknown knowns, and a three-stage workflow (pre-implementation, during, post-implementation) with practices such as implementation notes and quizzes.

rss · meng shao(@shao__meng) · Jul 4, 00:53

Background: The 'map is not territory' metaphor highlights the difference between mental representations (maps) and reality (territory). The 'known unknowns' and 'unknown unknowns' categories originate from project management and strategic planning. In this context, the map refers to the prompt and context given to the model, while the territory is the actual codebase and real-world constraints.

Tags: #LLM, #Prompt Engineering, #认知框架, #Claude, #AI协作

EU Probes Android for Favoring Gemini Under DMA
欧盟调查安卓系统因偏袒 Gemini 违反数字市场法 ⭐️ 8.0/10

The European Commission is investigating Google's Android operating system for violating the Digital Markets Act by giving its Gemini AI system-level advantages over third-party AI services. The probe focuses on Gemini being pre-installed and deeply integrated into Android, with special privileges not available to competitors. This case could set a precedent for how AI assistants are regulated on dominant mobile platforms, potentially forcing Google to offer equal access to third-party AI services. It also highlights the DMA's role in curbing self-preferencing in emerging AI markets. The investigation specifically targets Gemini's built-in status and system-level integration on Android, which the Commission believes may constitute a form of self-preferencing prohibited under the DMA. Google has until a yet-unannounced deadline to respond and potentially make changes.

rss · Michael Tsai · Jul 3, 18:23

Background: The Digital Markets Act (DMA) is an EU regulation that became applicable in May 2023, targeting large digital platforms deemed 'gatekeepers' to ensure fair competition. It prohibits gatekeepers from self-preferencing their own products or services over those of third parties. Android, as a core platform service, is subject to these rules, and the Commission previously enforced requirements allowing users to remove pre-installed apps and choose default browsers.

Tags: #Android, #DMA, #Gemini, #Regulation, #AI Competition

a16z: AI Increases Drug Supply, Reduces Per-Drug Value Due to Clinical Bottleneck
a16z 称 AI 导致药物过剩，临床瓶颈降低每款药价值 ⭐️ 8.0/10

Venture capital firm a16z argues that while AI accelerates drug discovery and expands pipelines, the number of FDA approvals per year remains stuck around 50, meaning each new drug becomes less valuable. The analysis, based on a long article by Formation Bio's CEO, identifies clinical trials as the new bottleneck. This contrarian view challenges the prevailing optimism around AI in pharma, highlighting that clinical translation remains the true limiting factor. It could reshape investment strategies and focus efforts on solving clinical bottlenecks rather than just discovery. Over the past decade, drug pipelines have doubled, yet FDA approval numbers have flatlined at about 50 per year. Formation Bio's CEO argues that discovery is no longer scarce; clinical capacity is the binding constraint.

rss · AI Will(@FinanceYF5) · Jul 4, 11:17

Background: AI drug discovery uses machine learning to screen and design candidate molecules faster, but all drugs must still pass rigorous clinical trials for safety and efficacy. Clinical trials are costly, time-consuming, and limited by patient recruitment, regulatory hurdles, and site capacity — making them the primary bottleneck. Despite market projections of rapid AI-driven growth in drug discovery, without resolution of clinical constraints, increased supply may not translate into more approved drugs.

Tags: #AI in Drug Discovery, #Pharmaceutical Industry, #Venture Capital, #Clinical Bottleneck

HOLA: Hippocampal Linear Attention for Long-Range Recall
HOLA：海马线性注意力实现长程召回 ⭐️ 8.0/10

Researchers introduced HOLA (Hippocampal Linear Attention), which augments linear attention with a bounded exact KV cache alongside the compressive delta-rule state, achieving improved long-range recall without sacrificing O(1) inference efficiency. This addresses a key limitation of linear attention and state-space models where early information is overwritten in the compressed state, enabling robust needle recall up to 32k tokens and outperforming full-attention Transformers on perplexity at 340M parameters. At 340M parameters trained on 15B SlimPajama tokens, HOLA reduces Wikitext perplexity from 27.32 to 22.92, below a full-attention Transformer++ at 26.88, and maintains robust RULER needle recall out to 32k tokens, 16x its training length.

rss · elvis(@omarsar0) · Jul 3, 15:38

Background: Linear attention and state-space models compress the entire sequence into a fixed-size recurrent state, enabling O(1) memory per step. However, this compression causes early key-value associations to be overwritten when many tokens compete, degrading long-range recall. HOLA combines this compressive memory with a small exact cache, inspired by the hippocampus, to store critical associations that should not be compressed.

Tags: #linear attention, #state-space models, #long-range recall, #efficient sequence modeling, #HOLA

NASA launches rescue satellite to save aging Swift telescope
NASA 发射救援卫星拯救老化 Swift 望远镜 ⭐️ 8.0/10

On July 3, 2026, NASA launched the LINK spacecraft, built by Katalyst Space Technologies, to rendezvous with and boost the orbit of the 20-year-old Neil Gehrels Swift Observatory, preventing its uncontrolled reentry. If successful, this will be the first private spacecraft to capture a US government satellite that was not designed for on-orbit servicing, demonstrating a new capability for extending satellite life and reducing orbital debris. LINK will use a robotic arm to grapple Swift and then fire thrusters to raise its orbit by about 240 kilometers. Swift could resume observations as early as September 2026 after the maneuver.

telegram · zaihuapd · Jul 3, 15:43

Background: The Swift Observatory, launched in 2004, studies gamma-ray bursts and other cosmic phenomena. Its orbit has been decaying due to solar activity, and without intervention it was expected to burn up in the atmosphere by late 2026. The LINK mission is part of a growing trend of commercial on-orbit servicing.

Tags: #NASA, #space telescope, #satellite rescue, #orbital boost, #private space

Tencent Xuanwu's Atu Yin AI beats Mythos in CyberGym test
腾讯玄武实验室阿图因 AI 在 CyberGym 测试中超越 Mythos ⭐️ 8.0/10

Tencent Xuanwu Lab announced that its Atu Yin AI scored 84.0% on the CyberGym cybersecurity benchmark, surpassing Anthropic's Claude Mythos Preview. The tool is built on the open-source model GLM-5.1 and costs less than 0.1% of Mythos's 'Glass Wing Project' budget. This demonstrates that open-source models can outperform top commercial AI in cybersecurity at a fraction of the cost, potentially democratizing advanced vulnerability detection. It also identifies multiple high-severity bugs missed by Mythos, highlighting practical security impact. Atu Yin AI found multiple critical logic vulnerabilities in curl, gnark, OpenSSL, Python cryptography, and Java bc-java, with scores up to 9.3. In Berkeley BVI's real-world vulnerability list, it ranked 1st in severity and 5th in total count.

telegram · zaihuapd · Jul 3, 16:12

Background: CyberGym is a cybersecurity benchmark created by UC Berkeley, based on 1,507 real-world vulnerabilities from 188 open-source projects. GLM-5.1 is an open-source large language model developed by Zhipu AI, focused on coding capabilities. The test evaluates AI's ability to detect security flaws in real software.

Tags: #AI安全, #网络安全, #开源模型, #大语言模型, #漏洞检测

Huawei Proposes Tao's Law: Time Scaling Replaces Geometric Scaling
华为提出韬定律：时间缩微替代几何缩微 ⭐️ 8.0/10

At the 2026 International Conference on Circuits and Systems in Shanghai, Huawei introduced 'Tao's Law', a new semiconductor principle that replaces geometric scaling with time scaling. Over the past six years, Huawei has designed and manufactured 381 chips based on this law, and plans to launch a new Kirin chip using LogicFolding technology this autumn. Tao's Law offers a potential path forward as Moore's Law approaches physical limits, shifting focus from transistor density to time-based optimization across device, circuit, chip, and system layers. This could help China's semiconductor industry bypass EUV restrictions and achieve 1.4nm-class chip performance by 2031. The 'LogicFolding' architecture physically folds logic circuits into a dual-layer structure, enabling higher transistor density without requiring extreme ultraviolet lithography. Huawei's paper claims a 55% improvement in transistor density compared to traditional scaling, with the first commercial chip expected in the Kirin lineup in fall 2026.

telegram · zaihuapd · Jul 4, 04:56

Background: Moore's Law has historically driven semiconductor progress by shrinking transistor dimensions (geometric scaling), but this is slowing due to physical and economic limits. Tao's Law instead proposes reducing a characteristic time constant across all system layers, optimizing heat removal and signal propagation. Huawei's approach combines this time scaling with LogicFolding to stack circuits vertically, mimicking density gains of advanced nodes without requiring cutting-edge lithography.

Tags: #semiconductor, #Huawei, #Moore's Law, #logic folding, #chip design

China Bans Mobile Phone Homework, Summer Tutoring
教育部禁止手机布置作业及暑假补课 ⭐️ 8.0/10

China's Ministry of Education issued a notice banning the use of mobile phones for assigning or completing homework, and prohibiting parents from checking and correcting homework. The notice also forbids summer tutoring, early school start, delayed vacations, and disguised exams under names like summer camps. This policy directly impacts millions of students, teachers, and parents across China, aiming to reduce screen time and academic pressure. It also addresses concerns over technology overuse in education and emphasizes standardized school schedules and assessment practices. The notice targets summer vacation specifically, banning any regional or inter-school unified exams before and after the break. It also prohibits disguised testing under the names of 'summer camps' or 'study tours'.

telegram · zaihuapd · Jul 4, 13:13

Background: In recent years, Chinese parents have complained about excessive homework and the burden of checking assignments via mobile apps. The Ministry of Education has been implementing 'double reduction' policies to alleviate academic pressure on students. This notice extends those efforts by specifically targeting mobile phone use and summer tutoring.

Tags: #Education Policy, #Mobile Phones, #Homework, #Summer School, #China

📊 Run stats · Total 4m 57s · AI analysis 1m 40s · Tokens 0.31 MCY (input 0.22 / output 0.10 MCY)