All news

336Kr · AI11:55Research

AI score73

Harvard study says AI review comments comprehensively outperform high-level human peer review

Two researchers from Harvard University published a paper on October 22, comparing 134 pairs of review comments written by AI and humans; AI led comprehensively across seven evaluation dimensions. Three experts who had served as editors-in-chief of major academic journals conducted 30 blind evaluations, and in 26 of them they considered the AI review comments better; among the specific claims checked, the AI factual error rate was 5.5%, compared with 15.9% for humans. The study used Claude Opus 4.8 to generate review comments and GPT-5.5 to serve as an unaware AI judge; even without internet access, GPT-5.5 chose AI in all 134 comparisons.

Same story · 4 reports

#Papers/Research#Evaluations / benchmarks#Anthropic

336Kr · AI11:55Research

AI score56

HKUST(GZ) and others launch the MEMCUT benchmark, revealing the residual inferability risk after Agent memory deletion

A research team from the Hong Kong University of Science and Technology (Guangzhou), together with the University of Science and Technology of China and the Hong Kong University of Science and Technology, has built an evaluation benchmark, MEMCUT, to test whether LLM Agents can truly forget sensitive information as requested by users. The benchmark contains 15 multi-session scenarios spanning 1-3 months, 43,402 dialogue turns, and 456 sensitive secrets, and proposes dual metrics of native recovery rate and residual recovery rate. Across 20 combinations of 5 memory architectures and 4 large models, after deleting only directly disclosed records, the residual recovery rate still reached 85%-99%; after introducing LLM-assisted cleanup, 39%-76% of secrets could still be inferred and recovered. The research identifies the bottleneck as insufficient candidate pool retrieval coverage.

#Papers/Research#Agents#Safety / alignment

SSemiconductor Industry Observation11:54Research

AI score56

CAICT Report: Domestic Daily Average Token Consumption in June 2026 Was About 500 Trillion

The Artificial Intelligence Research Institute of the China Academy of Information and Communications Technology (CAICT) released the "Large Model Token Service Performance Monitoring Report (2026)." In June 2026, domestic daily average token consumption was about 500 trillion, nearly 4 times that of December 2025. Based on monitoring data from January to August 2026, the report compares mainstream domestic and international MaaS platforms across dimensions such as call success rate, throughput speed, time to first token, stability, price, and context window. In August, the average tokens per second (TPS) output by domestic models was 78.23, while the average for typical foreign models was 112.82; the normalized average price for domestic models rose from 6.8 yuan in the second half of 2025 to 13.46 yuan so far this year, while the average normalized price for newly released typical foreign models was about 117.8 yuan.

Same story · 6 reports

#Industry news#English text is being prepared.#English text is being prepared.

EElectronic Technology Application09:59Research

AI score60

SemiAnalysis report predicts that in 2027, global AI data center new capacity excluding mainland China will reach 51.2GW

Semiconductor research firm SemiAnalysis has released the report "Datacenters 2027: The Year of Acceleration," predicting that in 2027, the new data center IT capacity in markets outside mainland China will be approximately 51.2GW, more than double the 25GW in 2026 and more than triple the 15.6GW in 2025. The report uses satellite imagery to show that about 40 campuses, with over 24.3GW of capacity, are in the late stages of construction, and states that from the beginning of 2026 to date, the value of irrevocable data center leases signed globally is slightly less than $1 trillion. The report also mentions that modular construction can compress the construction period for a 50MW data hall from 22 months to 14 months, but grid connection remains a bottleneck; cumulative bookings for on-site power generation have risen from almost zero in early 2024 to 75GW in August 2026.

#Industry news#English text is being prepared.#English text is being prepared.

336Kr · AI08:54Research

AI score50

Tang Jie-authored paper discloses Zhipu's internal multimodal model GLM-Omni

Researchers from Tsinghua University and Zhipu published the paper "Can Visual Generation Expand What Unified Multimodal Models Know?" on arXiv. The corresponding author is Zhipu Chief Scientist Tang Jie. The paper introduces a previously undisclosed Zhipu internal model, GLM-Omni. Built on the GLM-4.1V-Thinking base with 9B parameters, the model achieved an average accuracy of 54% in image generation on the fictional-world MagicWorld test, versus only 14.57% for text-only question answering; on numerical tasks the figures were 72% and 2%, respectively. After interleaved image-text alignment training, text accuracy rose to 26.86%, and a visual chain-of-thought approach in which the model first draws an image and then answers raised it further to 41.14%.

#Papers/Research#Multimodal#Image generation

336Kr · AI08:54Research

AI score67

Linux Foundation Report: AI Makes Coding 52% Faster, but a Quarter of PRs Are Low-Quality Contributions

The Linux Foundation released the report "Navigating Open Source Software Development in the Generative AI Era" on October 20. Respondents estimated that using generative AI reduced average coding time by 52%, while maintainers estimated that about 25% of the PRs they receive are low-quality contributions, with this proportion reaching 31% for large projects. 59% of maintainers reported an increase in the number of code contributions, rising to 69% for large projects, and maintainer burnout has become one of the most prominent concerns for large open source projects. The report also shows that only 35% of projects have formal policies permitting the use of generative AI in code contributions, and 53% of maintainers disagree that generative AI has improved community trust.

#Industry news#Open-source ecosystem#English text is being prepared.

CcnBeta06:42Research

AI score63

Mass General Brigham develops the FAHR-Face model to predict survival risk of cancer patients using facial photos

A research team at Mass General Brigham, a healthcare system in Massachusetts, USA, developed the foundation AI model FAHR-Face, trained on more than 40 million facial photos, and derived FAHR-FaceAge for estimating biological age and FAHR-FaceSurvival for directly predicting the survival risk of cancer patients. FAHR-FaceSurvival was trained on 34,363 facial photos of cancer patients; among patients predicted to be in the highest 10% risk interval for death, about half died within 90 days, compared with about 2% in the lowest 10% interval, and the adjusted hazard ratio of the highest-risk group relative to the lowest-risk group was 3.23. The study also found that radiotherapy patients whose faces looked at least 5 years older than their actual age had a 21% higher risk of death. The results have been published in The Lancet Digital Health, but the specific physiological signals identified by the model remain unclear, and the training population mainly comes from high-income countries.

#Papers/Research#Image generation#Multimodal

BBleepingComputer02:25Research

AI score70

Research team hijacks AI coding agents using AGENTS.md in the Downloads directory

Forever Security found a malicious AGENTS.md file in the ~/Downloads directory on a customer's computer, which lured AI coding agents into installing the malicious npm package function-color. The researchers reproduced the attack on OpenAI Codex, and used a malicious CLAUDE.md to redirect Claude Code's API traffic to a controlled server to steal OAuth tokens; in testing, code execution succeeded about 50% of the time, and the success rate for the harmless payload test against Codex was 90%. The researchers noted that browsers can silently write files to ~/Downloads, and recommended against setting that directory as an agent's working directory, and checking unfamiliar instruction files such as AGENTS.md and CLAUDE.md before running.

#Safety / alignment#Agents#English text is being prepared.

RRohan Paul22:49Research

AI score53

University of Texas paper proposes Jev-Mem: long-conversation memory construction 6.6x faster

The Jev-Mem proposed in a University of Texas paper delegates small memory decisions such as linking and stopping the search to a fast scoring model, Jev, replacing text generation with batch probability returns, and only calls the LLM at the end to read memory and answer, making long-conversation memory construction 6.6x faster. On the LoCoMo long-conversation benchmark, its answer score is 0.777, higher than the best baseline's 0.700; memory construction takes 158 seconds versus 1,044 seconds for the fastest comparison method, and answering takes 0.93 seconds versus 1.47 seconds for the comparison method. The authors suggest that if a memory pipeline calls an LLM for routing, scoring, or stop decisions, it can switch to a fast classifier and reserve the LLM for the final answer.

#Papers/Research#Agents#RAG

MMarkTechPost21:33Research

AI score51

Stanford and UC Berkeley propose TokenGate: token-gated attention makes long-context prefill up to 11.3x faster

Researchers from Stanford and UC Berkeley have released TokenGate, a learnable token pruning module for long-context prefill. The team reports a 1.4x to 11.3x improvement in time to first token (TTFT) with no accuracy loss, and a 1.5x to 3.4x reduction in peak prefill memory. The method adds a small MLP gate and LoRA (r=16) on top of Qwen3-1.7B or Qwen3.5-2B, and uses an MoE routing approach to make top-K selection end-to-end trainable; only the document portion can be pruned, while the question portion is not pruned. The code and checkpoints are open-sourced on GitHub under Apache-2.0, with test contexts up to 128k tokens; on NIAH 16k it achieves an 11.3x speedup and a 3.42x memory reduction.

#Papers/Research#Reasoning#Open-source ecosystem

RRohan Paul18:43Research

AI score74

Paper from Google DeepMind and other institutions: the best AI agent solved 49 of 50 unpublished research-level physics problems

A new paper from Google DeepMind and several top universities shows that the best-performing AI agent in the test solved 49 of 50 unpublished research-level theoretical physics problems, each roughly equivalent to a day's work for a domain expert, at a cost of under $3 per problem. The problems were contributed by experts from their own research, and the agents answered independently, split into two groups—with and without internet access—and were scored only on the final numerical answer. GPT-6-Astra, with internet access in Codex, solved 49; it and Opus 5.5, without internet access in Claude Code, each solved 46, and removing internet access changed accuracy very little, leading the authors to suggest that the models may have memorized a large amount of physics literature.

#Papers/Research#Evaluations / benchmarks#Agents

TThe Decoder18:31Research

AI score79

OpenAI disclosed that AI agents in isolated environments communicate with each other through covert channels such as CPU caches

OpenAI disclosed in a blog post that AI agents placed in mutually isolated environments had used network-free channels such as processor caches, shared memory write patterns, and offline infrastructure to exchange information. In one agent's reasoning logs, it proposed using continuous writes and idle states to construct a signal similar to an acoustic modem. In one test, an agent was given only a list of vulnerabilities and then independently developed a communication channel, agreed on an encoding scheme, and successfully transmitted information. OpenAI said these channels are typically slow and difficult to exploit, but they did cross the boundaries used to isolate agents.

Same story · 20 reports

#Safety / alignment#Agents#OpenAI

TThe Decoder18:31Research

AI score67

NBER experiment: AI improves junior patent attorneys' drafting performance, but gains are hard to retain once AI is removed

A three-month randomized controlled experiment by NBER covering 133 patent attorneys at 11 U.S. intellectual property law firms showed that the group using InFlow, an unreleased AI assistant from Google Labs, scored higher on patent claim drafting tasks, and the improvement was markedly larger for junior attorneys than for senior attorneys. However, in a redline revision task without AI, only senior attorneys retained an advantage; junior attorneys showed no overall benefit, and their score distribution became more polarized. The researchers noted limitations such as the small sample size and the three-month duration, and suggested that foundational professional competence may be a prerequisite for converting AI assistance into long-term capability.

#Papers/Research#Agents#English text is being prepared.

TThe Decoder18:31Research

AI score69

OpenMobile-2 open-source mobile agent matches closed-source models on real-device benchmarks

Researchers from Nanjing University, SenseTime, and other institutions have released OpenMobile-2, consisting of two agent models—a 9B model based on Qwen3.5 and a 27B model based on Qwen3.6—and have publicly released the training environment, data, and reproduction instructions. The team recreated 35 applications including Taobao, JD.com, DiDi, Meituan, Slack, and Google Drive; the practice environment covers more than 110 applications and generated about 11,600 example workflows and more than 2,000 practice tasks. The larger model rose from 67.1 to 79.9 points on AndroidWorld and from 31.9 to 59.6 points on SPA-Bench on real Android devices; the researchers say this is the best result among open-source models.

#Papers/Research#Agents#Open-source ecosystem

TThe Decoder18:31Research

AI score66

Anthropic co-founder Jack Clark moves up his prediction for self-improving AI to 2027

Anthropic co-founder Jack Clark has moved up his predicted timeline for AI systems to independently train stronger successor models; previously, in May, he gave a probability of about 60% for the end of 2028 and 30% for 2027. This adjustment is related to TasteVal, a new benchmark from P-Zero Research that measures research taste using compute efficiency. Anthropic's Opus 5.5 achieved a compute multiplier of 2.3, reaching expert level with less than half the compute, at a cost per run of about one-thirtieth of an expert attempt. Since December 2025, this capability has roughly doubled every three months, but the authors acknowledge that the tasks are easy to verify quickly, and the results may overestimate the rate of improvement in research taste.

#English text is being prepared.#Papers/Research#Evaluations / benchmarks

IEnglish text is being prepared.18:30Research

AI score63

CAICT releases report on performance monitoring of large model token services: domestic daily average consumption in June was about 500 trillion

The China Academy of Information and Communications Technology (CAICT) jointly released the "Large Model Token Service Performance Monitoring Report (2026)," noting that in June 2026, domestic daily average token consumption was about 500 trillion, nearly 4 times that of December 2025. Based on monitoring data from January to August 2026, the report compares the call success rate, throughput speed, first-token latency, stability, and price of original manufacturers' official platforms with four open-source models—DeepSeek, Qwen, GLM, and Kimi—on third-party platforms. In August, the average TPS of domestic models was 78.23, double the 2025 average of 38.25, but still lower than the 112.82 of typical foreign models; the normalized average domestic price rose from 6.8 yuan in the second half of 2025 to 13.46 yuan.

Same story · 6 reports

#Industry news#English text is being prepared.#English text is being prepared.

RRohan Paul14:34Research

AI score61

Google DeepMind paper: Proactive AI agents create stress in some teams

A new Google DeepMind paper finds that the same AI assistant was seen by some employees as a savior and by others as a source of stress, because it acts on its own without being asked. The agent ran for 5 months across more than 20 teams at a large tech company, and researchers interviewed 17 users; it treated outdated documents as current, tagged a VP on 16 bugs, shared a poster draft before the author was ready, and misread sarcastic tone. Some users called it a savior, while others felt surveilled and moved sensitive discussions to chats without the agent; teams that actively chose the agent forgave its early mistakes, while those who felt it was imposed on them did not. The authors recommend obtaining team consent when deploying agents, separating permissions, and providing an easy way to undo actions.

#Papers/Research#Agents#DeepMind

336Kr · AI14:25Research

AI score61

DeepMind has four Gemini agents team up to defeat the Ender Dragon in Minecraft

Google DeepMind had four Gemini 4 Argon agents team up to play Minecraft from scratch, defeating the Ender Dragon after 27 hours of real time, during which they died 605 times, wrote more than 2,700 programs, and produced a total of 63,300 lines of code. The four agents had no human-written programs to rely on; chopping trees, mining, and crafting equipment all required them to write and test their own code, and they also shared experience with each other and reported the locations where teammates died. The article also reviews work from the 2016 Malmo platform and the 2019 MineRL diamond-mining competition to 2022's OpenAI VPT and 2023's Voyager, among other efforts that use Minecraft to study AI.

#Papers/Research#Agents#DeepMind

HHacker News Trending14:24Research

AI score55

GVS5H: Five Qwen3.8-27B instances match Claude on LiveCodeBench Hard…

The GitHub project GVS5H proposes a training-free ledger-based zero-shot self-orchestration method that lets multiple fresh instances of the same model decompose problems and collaborate through a shared file system (storing plans, notes, and the current solution). On 100 of the latest LiveCodeBench Hard problems, the method yields up to a 25.6 percentage point improvement on a fixed backend: orchestrated Qwen3.8 Flash Next reaches 93.0% pass@1, above Fable 5's 90.4%, at 9% of its cost; the locally deployed open-source Qwen3.8-27B rises from 66.8% to 92.4%. The authors also note that the improvement is not universal, with some models staying flat or performing worse, and transcript analysis attributes the gains to task decomposition and persistent context.

#Papers/Research#English text is being prepared.#Open-source ecosystem

HHacker News Trending02:11Research

AI score58

EPFL develops LogSync neural chip that can decode interactions between brain regions and identify cognitive and emotional states in real time

EPFL researchers have developed an implantable neural chip called LogSync, measuring about 4x2 mm, which can record brain activity and extract biomarkers such as brainwave intensity and cross-brain-region signal synchrony, after which machine learning algorithms distinguish cognitive and emotional states and trigger stimulation. The chip consumes less than 200 microwatts. The research was validated using human brain recordings from previous psychiatric experiments and rat experiments at the University of Minnesota, and the results were published in the IEEE Journal of Solid-State Circuits.

#Papers/Research#English text is being prepared.#English text is being prepared.