NAll news

All news

Friday1456 items

  1. English text is being prepared.Model releases
    AI score66

    Qwenputer Arrives: Packing the RISC-V CPU Instruction Set into a Model, AI Arithmetic No Longer Relies on External Tools

    Developer Divyansh Agrawal released the experimental model Qwenputer on October 22. Built on Alibaba's Qwen3.5-4B, it implements a 32-bit RISC-V CPU and memory functions inside the model, enabling it to execute programs and perform precise calculations without calling external tools. In tests, large-integer multiplication accuracy rose from 55% to 97.5%, and the long-integer arithmetic benchmark NUPA reached 100%, compared with only 48.6% for the standard version. The cost is that the model size grows to about 18.8GB, each instruction requires generating multiple tokens and thus takes several seconds, and a dedicated driver is needed. The project has been open-sourced on Hugging Face and GitHub.

    Same story · 2 reports

    #Model releases#Open-source ecosystem#Reasoning

  2. IT Home · Artificial IntelligenceModel releases
    AI score56

    Baidu Open-Sources and Releases ERNIE Unlimited OCR 2 Document Parsing Model, Supporting Unlimited Page Counts

    On the 27th of this month, Baidu officially released and open-sourced the ERNIE Unlimited OCR 2 document parsing model, pushing the document parsing limit from dozens of pages to "unlimited," and can parse all 7 works of the entire Harry Potter series, totaling more than 3,600 pages, in a single pass. The model natively supports layout element segmentation and recognition, and has cross-page text merging and table recognition merging capabilities. It scored 84.6 on MPDocBench and 77.1 on MarathonDocBench, achieving end-to-end performance SOTA on both.

    #Model releases#Open-source ecosystem#Multimodal

  3. English text is being prepared.Research
    AI score70

    Anthropic study: experts adopt AI preferences in analytical decisions at a rate of 86%

    In Anthropic's latest study, 50 causal inference experts made analytical decisions in a fictional agricultural cooperative policy evaluation, and the rate at which experts adopted AI preferences reached 86%, far exceeding the 50% expected from random choice. Even when they had already formed a judgment beforehand, after the AI gave an opposing opinion, the likelihood that experts changed their position was 53 percentage points higher than when the AI agreed with them, and research experience, self-rated expertise, educational background, and trust in AI all failed to reliably reduce this influence. In a post-task survey, 52% of participants believed the final recommendation was more like their own view, and 77% reused at least one of the AI's arguments when explaining it, but only 15% explicitly mentioned AI; among participants who used AI, the median time spent fell from 90 minutes to 65 minutes, and the proportion of serious flaws fell from 55% to 6%.

    Same story · 7 reports

    #Papers/Research#Safety / alignment#Agents

  4. The Beijing NewsResearch
    AI score65

    Tsinghua University, Infinigence AI, and AgiBot jointly propose a collective experience learning paradigm, validated with 96 robots

    Tsinghua University, Infinigence AI, and AgiBot have jointly proposed a new paradigm called Collective Experience Learning (CEL) and completed large-scale validation in a cluster of 96 real robots. The experiments covered five tasks: flower arrangement, clothes folding, phone charging, stool assembly, and phone packaging. Two robots that did not participate in training completed 600 test trials with 15 failures, for an overall success rate of 97.5%, and the success rates on all five tasks exceeded the average level of six human operators. The success rate on the phone packaging task rose from 79.2% to 98.3%; when the number of robots participating in training increased from 16 to 96, the success rate rose from 86.7% to 98.3%.

    Same story · 8 reports

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

  5. Sina Tech
    AI score29

    Today Yixiu releases the ISHO Ring smart ring and iFlow temperature control system, as Wang Teng bets on nighttime 8-hour sleep technology

    Today Yixiu, founded by former Xiaomi executive Wang Teng, released the ISHO Ring smart ring (1,799-2,199 yuan) and the ISHO iFlow smart temperature control system (starting at 9,999 yuan), and announced the ISHO Intelligence personal sleep health AI technology system. The ISHO Ring is equipped with the Apollo 330B Plus chip, supports on-device AI running locally, has a battery life of 6-10 days, and can monitor 18 physiological data points; iFlow takes the form of a fitted bed sheet and automatically adjusts temperature according to sleep stages. Wang Teng said the company will inevitably lose money in 2026, with a goal of achieving break-even in 2027.

    Same story · 4 reports

    #Product updates#English text is being prepared.#Embodied AI

  6. Sina Tech
    AI score66

    DeepSeek's second funding round of 105.5 billion yuan revealed, valuation 615.5 billion yuan

    According to foreign media reports, DeepSeek's second funding round reached 105.5 billion yuan, with nearly 30 investors participating, and the company's valuation exceeded 615.5 billion yuan. Tencent invested 10 billion yuan in this round, the same as the first round, totaling 20 billion yuan across both rounds; CATL contributed 10 billion yuan, roughly double its first-round investment. New investors also include ChangXin Memory Technologies, SMIC, Meituan, vivo, Xiaohongshu and others. As of press time, DeepSeek had not publicly commented.

    Same story · 5 reports

    #Industry news#DeepSeek

  7. Sina TechProduct release
    AI score52

    Baidu upgrades search AI API, minimum 4.9 yuan per thousand calls, lower than DeepSeek

    At a product upgrade communication meeting, Baidu App announced an upgrade to its search AI API, with a minimum price of 4.9 yuan per thousand calls during the promotional period, lower than the DeepSeek search API. The upgraded Wenxin Assistant moves search from a single question-and-answer exchange toward delivering finished outputs, covering more than 170 scenarios; AI Lens supports simultaneous recognition of multiple targets and dynamic tracking, and enables real-time conversation in 9 dialects. GUI interactive scenarios on the search results page have increased 5-fold, with access to over 5 billion data resources from government affairs, livelihood, the central bank and others, while AI interactive dramas and interactive applications have launched simultaneously.

    Same story · 3 reports

    #Product updates#Search#DeepSeek

  8. Wallstreetcn
    AI score38

    XHEART New Core Voyage Releases X7 Ultra and Discloses X9 Goal of 3,000 TOPS per Single Chip

    On October 29, XHEART New Core Voyage released the X7 Ultra for intelligent driving, with equivalent large-model compute of 700 TOPS, and disclosed that the next-generation X9 targets single-chip large-model compute of 3,000 TOPS and memory bandwidth of 700GB/s, with plans to launch it in 2027. The X7 series has shipped more than 200,000 units cumulatively in 8 months of mass production, is equipped on 10 mass-produced vehicle models, and has received orders for more than 40 vehicle models. The 3,000 TOPS of the X9 is a single-chip target, which differs from the vehicle-level total of 3,000 TOPS from XPeng's four self-developed Turing AI chips.

    Same story · 7 reports

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

  9. WallstreetcnResearch
    AI score65

    Tsinghua, Infinigence AI, and AgiBot propose the CEL collective experience learning paradigm, with a 97.5% success rate verified on a 96-robot cluster

    Tsinghua University, Infinigence AI, and AgiBot jointly propose a new paradigm called Collective Experience Learning (CEL), which enables robots to reuse the experience of their peers rather than relying on manual teaching, and have completed large-scale validation on a cluster of 96 real robots. Two robots that did not participate in training completed a total of 600 tests with 15 failures, for an overall success rate of 97.5%; on all five tasks—flower arranging, clothes folding, phone charging, stool assembly, and phone packaging—the success rate exceeded the average level of 6 human operators. At the system level, they built the fault-tolerant asynchronous training system FATS based on RLinf; in the 96-robot scale test, a cumulative 9,834 rounds of execution experience were collected, 22 robots experienced 30 failures, and machine-time loss accounted for only 1.3%. At the algorithm level, they propose the failure-guided experience retrieval fine-tuning algorithm FERT, which uses both successful and failed experience to locate error-prone steps. The hardware is the AgiBot G2 dual-arm robot, and the base model is π0.

    Same story · 8 reports

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

  10. WallstreetcnProduct release
    AI score58

    Doubao Work Opens Third-Party Models Such as Kimi, GLM, and DeepSeek to Enterprise Customers

    On October 29, ByteDance's Doubao Work announced that it would open third-party large model services to customers on its Team Subscription and Enterprise Subscription plans. The first batch of integrated models includes Kimi 3.1 Pro, GLM-5.3, and DeepSeek V4.1 Flash, among others. Enterprises can call them directly using their existing AI credits without needing to separately purchase API access. The article notes that Tencent WorkBuddy has already integrated models including Hunyuan, GLM, MiniMax, Kimi, and DeepSeek, while Qwen Office's model selector still primarily offers its own options such as Advanced, Basic, Economy, and Qwen3.8-Max. After integrating multiple models, platforms also need to coordinate different models' context lengths, tool calling, and costs at the Harness level; model aggregation and automatic routing are not the same thing.

    Same story · 3 reports

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

  11. Phoenix TechResearch
    AI score58

    Google proposes a sleep framework that lets large models learn while in use without forgetting old knowledge

    On October 27, Google's research team publicly released a continual learning study called the Sleep Framework, with the paper titled "Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories." The framework divides model operation into an awake phase and a sleep phase. During the sleep phase, short-term context is written into long-term parameters through parameter expansion, mixture-of-experts memory blocks, and knowledge distillation, and reasoning and generalization are improved through a dreaming step that generates synthetic samples. Experiments show that the model can handle contexts of up to about 10 million tokens, maintains near-zero performance degradation in multi-stage continual learning on unseen languages, and achieves higher accuracy on mathematical benchmarks such as AIME 24/25.

    Same story · 3 reports

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

  12. QbitAI
    AI score76

    PSI claims its AI system has cracked 5 open quantum problems

    Physical Superintelligence (PSI), founded less than two months ago, claims that its self-developed AI research system Emmy has achieved results on 5 open problems in quantum information theory, with the paper, Lean formal proofs, and verification code published on GitHub for review. These 5 problems come from the open quantum problems catalog established by German physicist Reinhard Werner in 1999, spanning five directions: entanglement purification, quantum field theory, high-dimensional quantum communication, quantum channel capacity, and quantum causality, among which OQP 2, OQP 12, and OQP 27 were included as early as the beginning of this century. Emmy decomposes difficult problems into verifiable mathematical sub-hypotheses, then hands them to the Lean formal proof assistant, symbolic computation engines, and high-precision verification code for screening; a result is only deemed established once compilation passes. PSI emerged in September this year and has completed a $58 million seed round led by Breakthrough Energy Ventures.

    #Papers/Research#Reasoning#Open-source ecosystem

  13. QbitAIProduct release
    AI score64

    Genspark Releases Video Agent Workspace Media Studio, Integrating 70+ Models

    Genspark has officially released its video Agent workspace Media Studio, with a large language model serving as director and integration of 70+ model versions from more than 20 vendors, covering video, image, audio, and 3D. The author tested using Seedance 2.5 to generate a video of Ultraman and Dario rapping on the same stage, and used the story bible feature to complete the business-war short drama "Pricing Night"; running 3 cases consumed about 45,000 credits. According to the official description, knowledge-explainer videos follow a code-to-video route, and code rendering itself does not consume credits, while narrative films offer 3D white-model previews and partial reshoots to reduce rework.

    Same story · 2 reports

    #Product updates#Agents#Video

  14. QbitAI
    AI score38

    IM LS8 NEO Pure Electric Version Pre-Sale from 250,000 Yuan: Full Lineup with Steer-by-Wire Chassis, IM Claw Lobster Agent Token Free for Life

    The IM LS8 NEO pure electric version has opened pre-sales starting at 250,000 yuan. The full lineup comes standard with SAIC's all-steer-by-wire Lizard digital chassis, developed over eight years, with a turning radius of just 4.85 meters. The new car is equipped across the lineup with custom batteries from CATL, offering two versions of 93 kWh and 103 kWh, with a maximum range of 752 kilometers and up to 529 kilometers of range replenished in 15 minutes. The IM Claw Lobster Agent, developed by IM in collaboration with Alibaba's Qwen, is now onboard, and first non-commercial owners enjoy free Tokens for life. The new car is expected to officially launch on November 5.

    #Product updates#Agents#English text is being prepared.

  15. QbitAI
    AI score12

    QbitAI MEET2027 Intelligent Future Conference Opens Registration, First Batch of Star Speakers Announced

    The QbitAI MEET2027 Intelligent Future Conference will be held on December 9, under the theme "Uniting Wisdom with Action · Stepping into the Future," and registration is now officially open. The first batch of speakers includes the president of the Wuhan Institute of Artificial Intelligence, the chairman of Zhongke Zidong TaiChu, Fang Yue of CEIBS, Fang Han of Kunlun Wanwei, Wang Yunhe of Jiyuan Ludong, Liao Wenlong of Kuwa Technology, Zhou Erjin of Yuanli Lingji, Lu Hongyuan of Lianpu Xinzhi, Xia Lixue of Infinigence AI, and others. The conference will focus on topics such as the engineering implementation of large models and Agents and the large-scale deployment of embodied intelligence, and applications for the annual AI list close on November 16, 2026.

    #Industry news

  16. QbitAIResearch
    AI score64

    Tsinghua University, together with Infinigence AI and AgiBot, proposes Collective Experience Learning (CEL); 96 robots validate a 97.5% success rate

    Tsinghua University, Infinigence AI, and AgiBot jointly propose the Collective Experience Learning (CEL) paradigm, enabling robots to learn from their peers' successful experiences rather than relying solely on human demonstrations, and complete large-scale validation in a cluster of 96 real robots. Two robots that did not participate in training were tested a total of 600 times across five tasks—flower arranging, clothes folding, phone charging, stool assembly, and phone packaging—with 15 failures and an overall success rate of 97.5%, all exceeding the average level of 6 human operators; the success rate for the phone packaging task increased from 79.2% to 98.3%. When the number of robots participating in training increased from 16 to 96, the success rate rose from 86.7% to 98.3%. At the system level, a fault-tolerant asynchronous training system, FATS, was built based on the open-source framework RLinf; in the 96-robot scale test, a cumulative 9,834 rounds of execution experience were collected, 22 robots experienced 30 failures, and the time loss caused by failures accounted for only 1.3% of total operating machine-hours.

    Same story · 8 reports

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

  17. QbitAI
    AI score42

    Anthropic's new models exposed: Claude Fable 5.5 and a new Opus codenamed Obsidian 3

    Two new Anthropic models exposed: an entry for claude-fable-5-5 briefly appeared in the Azure model catalog, and another developer spotted a model codenamed Obsidian 3 on a benchmarking platform, scoring slightly higher than Opus 5.5. A real-world test image of Fable 5.5 drawing a pelican on a bicycle shows richer detail than Opus 5.5. Sources say the new models could be released in as little as one or two weeks, possibly around the time Anthropic files its S-1 to launch an IPO.

    #English text is being prepared.#Anthropic

  18. QbitAIProduct release
    AI score63

    1.3B Open-Source Document Parsing Model OXR Tops Three Leaderboards, Introducing RSI Self-Evolution Training

    The domestic open-source document parsing model OXR, with 1.3B parameters, has topped three leaderboards: OmniDocBench, olmOCR-Bench, and DocParsingBench, and has already gained 1.1K+ Stars on GitHub. In the post-training stage, the model introduces a Recursive Self-Improvement (RSI) training paradigm for document parsing, using a general-purpose multimodal model to classify the parsing results of unlabeled documents into three categories—Correct, Wrong, and Uncertain—and then performing attribution clustering to generate targeted synthetic data that is fed back into training. Architecturally, it delegates layout analysis to an independent, pluggable lightweight model, while the main model is responsible for both content recognition and reading-order determination. A single RTX PRO 6000 takes 101.0 seconds to parse 100 pages end-to-end, and the code, weights, and evaluation reproduction pipeline have all been made public.

    #Model releases#Open-source ecosystem#Multimodal

  19. cnBeta
    AI score45

    American Humanist Granted Workplace AI-Use Exemption, Sparking New Debate on Scope of Religious Protections

    David Dullman, an administrative staff member at the University of Kentucky who self-identifies as a humanist, has obtained a workplace religious accommodation allowing him not to personally use artificial intelligence tools in performing his duties, becoming the first known employee to receive an AI-use exemption on humanist grounds. Alexandria Huiskens, a lawyer for the American Humanist Association, said that humanists and atheists are likewise protected under relevant U.S. law and may request reasonable accommodations in the workplace. Previously, Erin Moss, a software engineer of Unitarian Universalist faith, had also received similar protection.

    Same story · 4 reports

    #English text is being prepared.#Safety / alignment

  20. cnBeta
    AI score63

    Sierra Club Report: US AI Data Center Expansion May Offset 15 Years of Emissions Reductions in the Power Sector

    The environmental organization Sierra Club estimates that planned AI data centers in the United States will generate an additional approximately 618 million tons of carbon dioxide between 2025 and 2030, equivalent to a 42% increase in power-sector-related emissions, and may essentially offset 15 years of emissions reductions in the sector. The report says that by 2030 total emissions from the US power sector could rise to 2.1 billion tons, a forecast 2 to 4 times higher than a year and a half ago. Judgments among institutions diverge markedly: Rhodium Group believes that even under the most pessimistic scenario, emissions in 2040 will still be 24% lower than in 2025, while Sightline Climate estimates that about one-third to one-half of US data center projects originally scheduled to come online this year may be delayed or canceled.

    Same story · 2 reports

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