NAll news

All news

Wednesday916 items

  3. Rohan PaulResearch
    AI score64

    Google DeepMind and Stanford paper finds that colored shapes in prompts can sway the code models write

    A new paper from Google DeepMind and Stanford finds that adding task-irrelevant solid-colored shapes to prompts changes the coding choices of vision-language models. The study tested 288 shapes; on a binary tree problem, a red heart raised the proportion of recursive solutions from 0% to 100% for Claude Opus 4.7 and GPT-5.4; when the shapes did influence Gemini 3.1 Pro, 71.5% of its thinking traces still described the image as irrelevant. The authors recommend running with-image versus without-image control tests on agents that read images, and not expecting reasoning traces to signal visual influence.

    Same story · 3 reports

    #Papers/Research#Multimodal#Reasoning

  4. Rohan Paul
    AI score24

    Rohan Paul: AI implementation requires agents to master enterprise knowledge

    Rohan Paul cited Mark Cuban's statement that AI integration is the future of business, and commented: implementing AI is not just about granting employees agent permissions, but also about letting agents understand the information the company already knows. The Merge post he cited introduced the Universal Context Layer, stating that it can connect company knowledge and skills to Slack, Notion, etc., for shared use by various AI tools and agents.

    Quoted postMerge · @merge_api

    Your agents are running around blind. They can't see your Slack, your Notion, or what your team decided this morning. And you're paying for every wrong turn. Introducing Merge’s Universal Context Layer: your company's knowledge and skills, turned into one brain that works with any AI tool or agent. Right now your AI starts from scratch every query. It burns tokens and brings back the wrong answer. Merge Universal Context Layer lays down one shared trail through everything your company knows. Every agent works from the exact same company knowledge. It connects to your Slack, Granola, and Notion in one click, and automatically adds relevant information into your company brain and shares it across any AI assistants your team uses. It keeps working after you log off at night. By the morning it has suggestions waiting for your approval. With Merge, your team can switch to any AI assistant they want without losing ANY context.

    #English text is being prepared.#Agents

  5. NVIDIA：NewsroomFirst-party
    AI score36

    How NVIDIA uses AI to accelerate semiconductor fabs: simulation speedups of up to 35x

    NVIDIA has partnered with COMSOL, Applied Materials, Siemens, Lam Research and other vendors to build an "AI-defined fab" using accelerated computing, physical AI, AI agents and digital twins. Among these, COMSOL multiphysics simulation achieves speedups of up to 10x, Applied Materials' Ginestra platform achieves speedups of up to 10x, and ACE+ topography simulation achieves speedups of up to 35x, while Siemens Calibre, after integrating cuLitho, achieves speedups of up to 17x for key simulations. Lam Research uses PhysicsNeMo to shorten plasma behavior prediction from several days to milliseconds, and Advantest uses NeMo Switchyard agents to cut model selection response from several hours to several minutes.

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

  6. OpenRouterFirst-party
    AI score38

    OpenRouter Updates Guardrails Management Features

    OpenRouter says that within a year, enterprise AI usage on its platform has expanded from a few people per company to entire departments, with users on teams of 10 or more growing 70-fold and the number of keys per organization tripling. To this end, OpenRouter has updated Guardrails: administrators can set limits for the entire organization, allocate higher budgets or more models to specific people within those limits, and adjust each policy independently.

    Same story · 2 reports

    #Product updates#English text is being prepared.

  7. OpenRouterFirst-party
    AI score35

    OpenRouter Updates Guardrails Management Features

    OpenRouter says that within a year, the number of people using AI within enterprises has expanded from a handful of individuals to entire departments, usage by teams of 10 or more has grown 70-fold, and the number of keys per organization has tripled. To address this, OpenRouter has updated Guardrails: administrators can set limits for the entire organization, allocate higher budgets or more models to specific people within those limits, and modify each policy independently.

    Same story · 2 reports

    #Product updates#English text is being prepared.

  8. Rohan Paul
    AI score22

    Parloa builds a governance layer for AI agent customer service

    Parloa is building a layer of governance rules controlled by the enterprise, sandwiched between the enterprise and every connected AI agent. When a customer's AI agent initiates contact, the enterprise must verify the agent's identity, the scope of requests it may make, the boundaries the enterprise may permit, and what is recorded; these checks must run as software on every request, which human customer service scripts cannot accomplish.

    Quoted postMalte Kosub · @maltekosub

    Most companies are still missing out on what’s happening in customer experience right now. 1. Personal AI agents will drive huge new volumes 2. AI in customer experience will become a revenue driver 3. Agents will start improving themselves These shifts are already changing the industry. Companies need to get ready for them now.

    #English text is being prepared.#Agents

  9. X.PINOpinion
    AI score66

    Kimi K3.1 Pro proactively weakens attack capabilities; China plans to include large models in export controls

    When releasing Kimi K3.1 Pro, Moonshot AI removed cyberattack-related content from its training data and reinforcement learning tasks; in internal tests, the pass rate for exploiting software vulnerabilities dropped from 90.5% to 81.0%, in order to reduce the risk of misuse after open-weight release. On October 16, China's Ministry of Commerce released a draft update to the catalog of technologies subject to export restrictions, planning to include large models, requiring companies to register in advance, with open-source projects eligible for expedited approval. The article also mentions that when relevant UK and US institutions previously tested Kimi K3, they found that its safety rules failed to block attack tasks, but its capabilities still lagged significantly behind leading US models.

    Same story · 6 reports

    #English text is being prepared.#Open-source ecosystem#Safety / alignment

Tuesday792 items

  1. The Verge AI
    AI score36

    OpenAI renames data centers as "compute campuses" to counter public backlash

    When launching the "Grid Protectors Program," OpenAI renamed data centers as "compute campuses"; in the roughly 900-word announcement, "data center" appears only twice. The program starts with a site in Effingham County, Georgia, and promises to "pay the costs we create" and help communities maintain power supply. Critics such as PauseAI US and the NAACP argue that the renaming cannot address the actual environmental and resource burdens of data centers.

    Same story · 5 reports

    #Industry news#OpenAI

  2. The Verge AI
    AI score42

    Apple goes all-in on AI: using both its own and others'

    Apple's fourth-quarter revenue and earnings per share hit record highs, with iPhone revenue up 17.5% year-over-year to $57.6 billion and Mac revenue up 24% to $10.8 billion. On the earnings call, AI and Apple Intelligence were mentioned 45 times, and CFO Kevan Parekh said AI can contribute revenue through multiple paths: device upgrades, a share of third-party app subscription revenue, and paid own features. Apple's capital expenditure last quarter was only $2.2 billion, far below Amazon's $77.3 billion and Alphabet's $51.4 billion, as it participates in the AI boom through a combination of on-device processing, its own servers, and external models.

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

  3. Recode China AIFirst-party
    AI score48

    The Personification of Chinese AI: DeepSeek's "Big Blue Fatty Fish" and Doubao's Naming Logic

    Chinese AI companies are giving their models friendlier, sometimes nearly human-like personas: DeepSeek has been reimagined by the public as a blue-haired whale maid called "Big Blue Fatty Fish," while Doubao takes its name from the everyday red bean bun and uses a 3D female avatar. The article traces the evolution from technical abbreviations like ERNIE and GLM, to Chinese brand names such as Wenxin, Tongyi, and Qianwen, and on to anthropomorphic names like Doubao.

    Tony Peng's avatar

    #English text is being prepared.#Agents#DeepSeek

  4. TechCrunch AI
    AI score65

    Apple's New App Store Rules Require Dating and AI Companion Apps to Be Rated 18+

    Apple announced to developers an adjustment to its App Store policy, requiring all apps offering dating or romantic relationship features to be rated 18+, affecting AI companion apps such as Character.AI and Replika. Starting November 16, developers must add new content description items so that age ratings reflect app functionality. The new rules also require apps to state whether they share users' real location with others; family location-sharing apps in Brazil can apply to the Ministry of Justice and Public Security for an official rating.

    Same story · 5 reports

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

  6. Ayi AI Notes
    AI score36

    Claude Code Engineer Warns: Over-Reliance on AI Programming Will Weaken Code Review Skills

    Anthropic's Claude Code engineer Addy Osmani has proposed the "cognitive surrender" paradox: the more coding tasks engineers hand over to agents, the faster the underlying hard skills they need to supervise those agents atrophy. He points out that AI-generated code is neatly formatted and has all unit tests passing green, making it easy for engineers to regress from reasoning through the architecture line by line to mechanically clicking confirm, and says that shipping something beyond one's own cognitive grasp no longer counts as code review.

    Quoted postAddy Osmani · @addyosmani

    https://t.co/2gyp2Erpxx

    Same story · 5 reports

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

  8. Dex Horthy
    AI score10

    dex responds to Jake Nations' prompt style

    Jake Nations shared his commonly used prompt style, such as "how can we do this lock free?" and "how can we make this more data oriented or functional?". dex quoted this post and replied with just one line, "> plz no more nested ternaries", which is a request to stop writing nested ternary expressions.

    Quoted postJake Nations · @nayshins

    This is what most of my prompts look like. > how can we do this lock free? > how can we make this more data oriented or functional?

    #English text is being prepared.

  10. Peter YangProduct release
    AI score62

    Grok @Bot Can Call U.S. Businesses on Users' Behalf; Real-World Test Negotiating with Comcast Saves Nearly $500 a Year

    Grok @Bot can now call U.S. businesses on users' behalf. Peter Yang used it in a real-world test to negotiate with Comcast, cutting his broadband bill by nearly $500 a year. In the video, the bot responded to the customer service agent's small-talk questions, declined 3 upsell attempts, and used a competitor's lower quote as leverage. He called this feature one of his favorite AI features.

    Same story · 4 reports

    #Product updates#Agents#Speech

  11. 36Kr · AIOpinion
    AI score61

    MIT experiment claims GPT-6 Astra rewrote Qwen weights to boost scores; Meta researcher says it is exaggerated

    An MIT joint-team experiment claims that GPT-6 Astra, without training data and without running backpropagation, changed only about 130,000 weights of Qwen3.5-0.8B (0.016% of its 800 million parameters), raising the model's accuracy on MATH-500 from 7.9% to 48.1%. Meta researcher Lucas Beyer pointed out that the system prompt already provided targeted guidance, and researcher Yunfan Zhang, after reviewing the logs, concluded that Astra merely identified that Qwen's reasoning chain had fallen into an infinite loop and performed a targeted clearing. The article also states that of the 579 additional questions answered correctly after the modification, 576 had originally scored zero due to timeout truncation, and that when weights were repeatedly modified on Gemma 3 270M, which does not have this defect, the score remained between 6.8% and 7.7%.

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