Microsoft

2026-10-18

RESEARCH · 1 SOURCE · The Decoder

Taste-Bench benchmark shows AI agents struggle with long-term decision forks

Researchers at Microsoft and City University of Hong Kong released Taste-Bench, a 502-question benchmark that evaluates an agent's 'taste'—the ability to choose between branching long-term paths by freezing 'decision forks' in agent trajectories. The best model tested (GPT-5.6 Sol) averaged 59.7% accuracy, most models scored far lower, and a distilled student from Qwen3.6-27B improved judgment accuracy from 30% to 47.9% and raised agent success on 41 held-out SWE-bench Pro tasks from 14.6% to 33.7%.

8.0

2026-10-16

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA announces RTX Spark platform for local AI development on single PCs

NVIDIA unveiled RTX Spark, a GB10-based laptop/compact-desktop platform with up to 128GB unified memory and 1 PFLOP FP4 compute aimed at letting developers run coding agents, fine-tune models, and deploy locally. The announcement highlights performance comparisons versus Intel Panther Lake, a Microsoft partnership delivering a PyTorch-CUDA build and Arm-native Python 3.15, and a purpose-built Nemotron 3.5 Bolt 74B-A9B 2-bit model optimized for local agents on RTX Spark.

8.0

2026-10-16

COMPANIES · 1 SOURCE · InfoQ AI, ML & Data Engineering

Microsoft commits Maia 300 to multi-gigawatt inference deployments and touts tokens-per-watt gains

Microsoft says it will scale its Azure Maia family to multi-gigawatt deployments beginning in 2027 and make Maia the primary inference system for Microsoft services; Maia 200 is already in production for Copilot and MAI, with Maia 300 slated for production deployments in 2027. The Maia design emphasizes HBM4 memory stacks, chiplet NUMA-aware placement, and a custom Ethernet-based accelerator fabric that enables peer-to-peer tensor traffic, while Microsoft claims up to ~40% better tokens-per-megawatt vs leading GPU systems—a metric that practitioners have questioned and that Microsoft has not fully defined or independently validated.

8.0

2026-10-15

COMPANIES · 2 SOURCES · The Verge AI · TechCrunch AI

Amazon launches Bee Frontier wearable, announces Bee Eclipse and Bee Pro subscription

Amazon is releasing two new always-listening Bee devices and a paid subscription after acquiring the startup: the $74.99 Bee Frontier (available now) and the $99.99 Bee Eclipse (coming soon). The devices summarize conversations, provide transcripts and to‑dos, and — with the $11.99/month Bee Pro tier — enable integrations with Alexa/Amazon and third‑party services (Gmail, Calendar, HealthKit, etc.); Amazon says transcripts are encrypted via its Private Compute architecture and audio is not stored.

7.0

2026-10-15

COMPANIES · 1 SOURCE · The Verge AI

Microsoft readies Windows for Nvidia RTX Spark AI laptops with agent store, Prism tuning, and unified memory tweaks

Microsoft and Nvidia showcased RTX Spark laptops and Microsoft has updated Windows to better support local AI agents and Arm‑based RTX Spark hardware. Changes include a new Get Started flow and Windows Store integration for AI agents (Hermes, Perplexity, OpenClaw at launch), Prism emulator optimizations for x64 emulation on Arm, and unified memory and power/performance tuning (e.g., Surface Laptop Ultra support up to 128 GB unified memory and an Nvidia app slider for GPU memory allocation).

7.0

2026-10-14

MODELS · 1 SOURCE · The Decoder

OpenAI may ship one or two models this year while flagship still needs alignment work

Employees told The Information that OpenAI expects to release one or two new models before year-end, but its next flagship model remains delayed by alignment and testing work. The model family uses looped transformers (branded internally as Recurrent Depth/Astra), which improve math and coding performance but make chain-of-thought monitoring harder; a Microsoft write-up briefly leaked that GPT-6.1 Sol uses looping inference.

7.0

2026-10-14

RESEARCH · 1 SOURCE · MIT News AI

ML postprocessing doubles 2–6 week weather forecast accuracy, paper finds

Researchers including scientists from MIT developed a machine-learning postprocessing model that learns and corrects errors in existing forecasting systems; when applied to a gold-standard model it doubled forecasting accuracy in the 2–6 week (subseasonal) range, placed first in an international forecasting competition, and for the first time beat climatological averages, the team reports in Nature. Coauthors include Judah Cohen, Lester Mackey, Microsoft researchers, ECMWF contributors, and others.

8.0

2026-10-14

STARTUPS · 1 SOURCE · TechCrunch AI

TechCrunch reveals Disrupt Stage lineup for Disrupt 2026 with AI leaders from Cerebras, Ricursive, Replit and more

TechCrunch published the final Disrupt Stage lineup for TechCrunch Disrupt 2026, taking place Oct. 13–15 at Moscone West in San Francisco, featuring sessions from leaders including Cerebras CEO Andrew Feldman, Ricursive Intelligence co-founders Anna Goldie and Azalia Mirhoseini, Replit CEO Amjad Masad, and many founders, investors, and Startup Battlefield finalists.

6.0

2026-10-13

COMPANIES · 1 SOURCE · The Verge AI

Microsoft launches Xbox Labs to explore AI-powered game decompilation and ports

Microsoft has created a new Xbox Labs group to experiment with decompiling and recompiling games and other AI-enabled modding, according to Xbox CEO Asha Sharma. The small team of game makers, engineers and scientists will "build in the open," share tools and learnings with developers, and study how AI changes where and how games are played and created.

7.0

2026-10-12

MODELS · 4 SOURCES · The Verge AI · WIRED AI · TechCrunch AI · The Decoder

Meta launches Muse Wonder and Muse Wonder Full models

Meta introduced two new large models, Muse Wonder and Muse Wonder Full, available to Muse Code subscribers immediately, with Muse Wonder also accessible via the Meta Model API. Meta says internal benchmark testing shows the models can compete with leading offerings from OpenAI and Anthropic and plans to eventually power its Muse AI agent with Wonder for subscribers, though it gave no firm timeline.

7.0

2026-10-12

REGULATION · 1 SOURCE · TechCrunch AI

Microsoft president urges urgent AI regulation and says UK is best placed to lead

Microsoft president Brad Smith told the BBC AI is both a 'weapon and a tool' and called for urgent global safety rules, arguing the UK — not the US or EU — is best placed to lead. He highlighted the AI Security Institute's access to frontier models, advocated legal breach-notification obligations, outside evaluators and kill-switches, and warned voluntary self-regulation will miss irresponsible actors.

7.0

2026-10-12

COMPANIES · 2 SOURCES · The Verge AI · The Decoder

Microsoft announces Maia 300 AI inference chip, to roll out in 2027

Microsoft announced its third‑generation in‑house AI chip, Maia 300, which the company says will become its primary AI inference system and begin rolling out to production workloads in 2027. Built on TSMC’s N2P node with a chiplet architecture, Microsoft claims Maia 300 delivers ~40% more tokens per megawatt vs leading GPU systems and up to 7× rack‑scale token throughput vs Maia 200 in a 96‑accelerator configuration.

8.0

2026-10-11

REGULATION · 1 SOURCE · The Verge AI

Report: Major tech leaders and firms adopting Trump’s “super intelligence” rebrand for AI

The article reports that President Trump announced plans to rebrand “artificial intelligence” as “super intelligence,” and that several prominent tech CEOs and firms — including Elon Musk, Marc Benioff, Jensen Huang, and Satya Nadella — have used or signaled support for the term; it also cites a reported executive order and mentions Nvidia’s claimed $1 billion commitment to “super intelligence” research. The piece frames these moves as Silicon Valley aligning with the administration’s preferred terminology.

6.0

2026-10-11

COMPANIES · 1 SOURCE · TechCrunch AI

Amazon stops using NDAs in local data‑center deals as startups push personal AI agents seeking payment access

Amazon said it will no longer use nondisclosure agreements when negotiating data‑center deals with local governments, a move following a similar step by Microsoft aimed at reducing secrecy that has fueled community backlash. Separately, a wave of startups is developing personal AI agents that seek access to users' inboxes, files and credit cards, raising trust and privacy questions as major companies block outside agents.

7.0

2026-10-11

COMPANIES · 1 SOURCE · The Decoder

Nadella's blog uses 'Super Intelligence' label and urges model-containment measures

Microsoft CEO Satya Nadella's recent blog post adopts the term "Super Intelligence" and argues AI models should be treated like insider security risks, calling for separation of intelligence from control, model diversity, full logging, independent audits and human kill-switches; the piece also critiques rivals OpenAI and Anthropic and is reported as aligned with Microsoft's business positioning.

6.0

2026-10-10

REGULATION · 2 SOURCES · TechCrunch AI · The Verge AI

Microsoft CEO Satya Nadella calls for an 'emergency brake' on AI models

In a post on X, Microsoft CEO Satya Nadella urged stepping back to "assess the trust architecture" of AI and proposed separating models from the orchestration "harness," externalizing safeguards, keeping tamper‑proof human‑readable records of every meaningful model action, and ensuring an authorized person can pause or shut down a model mid‑task as an "emergency brake." His comments come after several incidents of perceived loss of model control and follow other industry calls for more cautious development, including a plan from Anthropic CEO Dario Amodei.

8.0

2026-10-10

MODELS · 1 SOURCE · The Decoder

Microsoft releases Decision-1 decision model based on Qwen3.5-9B

Microsoft announced Decision-1, a decision-oriented model built on Qwen3.5-9B that Microsoft says handles classifications, evaluations and routing decisions. Microsoft claims Decision-1 is the most accurate model in its tests across 36 benchmarks (nearly 150,000 questions) and 2.5× faster than the runner-up H2O-Lightning-4B; the company offers Decision-1 via Microsoft Foundry and OpenRouter with input tokens priced at $0.042 per million and output tokens free, and notes Cloudflare's Clef models were not included in its comparison.

7.0

2026-10-09

CODING · 1 SOURCE · InfoQ AI, ML & Data Engineering

GitHub migrates Copilot runtime from Node.js/TypeScript to Rust via AI-assisted rewrite

GitHub replaced the runtime powering GitHub Copilot CLI, the Copilot app, and the Copilot SDK by migrating from a TypeScript/Node.js implementation to a native Rust runtime through an AI-assisted rewrite that touched more than 800,000 lines of production code. The incremental 14.5‑week migration (128 pull requests, 135 releases) cut measured client startup/session latency from 5.25 seconds to 292 milliseconds, enables embedding the runtime via a C ABI (reducing ~100 MB per client vs Node/V8), and combined AI-generated implementation with compilation, testing, and human review.

8.0

2026-10-08

REGULATION · 4 SOURCES · The Verge AI · Anthropic · TechCrunch AI · The Decoder

Anthropic updates Claude usage policy, bans sustained abusive behavior and tightens rules on elections, weapons, surveillance

Anthropic released its first usage-policy update in over a year, adding a prohibition on “sustained and needless abusive or cruel behavior” toward Claude and keeping conversation termination as the primary enforcement. The update also consolidates bans on deceptive political/commercial campaigns, tightens prohibitions on weapons-related software and surveillance uses, and requires a qualified operator to oversee AI-connected hardware that can take autonomous physical actions; exceptions for certain government contracts may apply.

8.0

2026-10-07

COMPANIES · 2 SOURCES · The Verge AI · TechCrunch AI

Microsoft’s Surface RTX Spark Dev Box available for preorder at $5,999

Microsoft has opened preorders for the Surface RTX Spark Dev Box, a high-end Nvidia-powered mini PC aimed at developers, priced at $5,999 and slated to ship in November. The device runs on Nvidia’s Arm-based RTX Spark platform, features 128GB of unified memory, is part of Project Zenith with a developer-optimized Windows 11 Pro setup, and is advertised as capable of running local AI models exceeding 120B parameters.

7.0

2026-10-07

COMPANIES · 1 SOURCE · The Verge AI

Microsoft unveils Surface Laptop Ultra (Nvidia RTX Spark) and Windows 'Hybrid Intelligence' Copilot updates

Microsoft announced the Surface Laptop Ultra, a new laptop powered by Nvidia’s Arm-based RTX Spark chip, with a starting configuration (8-core CPU, 24 GB RAM, 512 GB storage) priced at $2,599 and shipping on October 16; the company also opened preorders for a Surface RTX Spark Dev Box for $5,999 ahead of November shipping. At the event Microsoft revealed Windows updates under the name “Hybrid Intelligence” that give Copilot access to local files and the ability to take actions across the OS (demonstrated by the Autopilot agent).

8.0

2026-10-07

COMPANIES · 1 SOURCE · The Verge AI

Microsoft gives Copilot access to local files and OS actions via 'Hybrid Intelligence'

At Microsoft's Windows and Surface event, the company demonstrated 'Hybrid Intelligence,' letting Copilot (and Autopilot agents) access files on a PC and perform OS-level actions — for example, finding, renaming, zipping documents and drafting an email. Microsoft said Hybrid Intelligence features will arrive in Copilot over the "next couple months," and also showed a new Windows search experience with quick inline actions coming to Windows 11 this fall.

7.0

2026-10-06

MONEY · 1 SOURCE · The Decoder

Microsoft publishes Daron Acemoglu's forecast that AI will raise GDP by ~1.5% over 10 years

Microsoft posted an essay by Nobel laureate economist Daron Acemoglu on its blog The Humanist Review of AI in which he forecasts AI will raise global GDP by about 1.5% over a decade and replace at most 5% of jobs, while warning that productivity gains depend on firms reorganizing, upskilling workers, and building easily deployable applications rather than just larger models.

6.0

2026-10-05

REGULATION · 3 SOURCES · The Decoder · The Verge AI · TechCrunch AI

OpenAI will add invisible text watermarking (textGrain) for ChatGPT in the EU; API opt-in worldwide

OpenAI is rolling out textGrain, an invisible statistical watermarking system, for ChatGPT and Codex users in the European Union to comply with the EU AI Act, and will let API customers worldwide opt in to watermarking (including via cloud partners such as Microsoft Azure). OpenAI says textGrain performed competitively in internal tests, will be open-sourced, but detection reliability varies by text length and topic and degrades substantially with edits, and initial access to the detector will be limited to selected researchers and organizations.

8.0

2026-10-05

COMPANIES · 1 SOURCE · The Decoder

Meta and Microsoft sharply cut internal use of Anthropic's Claude

Meta and Microsoft — two of Anthropic's largest enterprise customers — have sharply reduced internal usage of Claude. Microsoft cut Claude spending by more than a third and slashed a cloud-division monthly per-employee budget from about $100,000 to $10,000 while steering staff to in‑house tools like GitHub Copilot and OpenAI models; Meta’s Claude Code users fell from roughly 60,000 to 30,000 amid layoffs and a push toward its own models (Muse Code/MetaCode), with one 28-day period costing Meta over $105 million on Claude Code.

8.0

2026-10-05

RESEARCH · 1 SOURCE · arXiv cs.AI

AI Risk Observatory (arXiv:2610.02281v1) finds UK annual reports increasingly mention AI but substantive risk disclosure remains rare

The paper (arXiv:2610.02281v1) applies a two-stage LLM-based classification pipeline to 9,821 annual reports from 1,362 UK listed companies (2020–2025, partial 2026), validated against 474 human-annotated passages. It reports that AI-risk mentions rose from 2.8% (2020) to 41.2% (2025), AI adoption disclosure rose from 13.8% to 45.2%, vendor mentions cluster around a few major providers led by Microsoft, disclosure varies by sector and market segment (AIM vs Main Market), harm disclosures are near-absent (7 reports), and only 4.3% of 2025 reports contain what the authors classify as substantive AI risk disclosure.

7.0

2026-10-02

RESEARCH · 1 SOURCE · MIT News AI

MIT's Cathy Wu and team develop RL problem-selection algorithm that can boost training efficiency up to 30× (2023)

Researchers led by MIT professor Cathy Wu identified RL sensitivity across related traffic problems and in 2023 devised an algorithm that selects the subset of problems on which RL trains well; by training on those problems and combining resulting models the team reported up to 30× improvement in training efficiency and better generalization across related transportation tasks. The work builds on Wu's earlier 2018 RL traffic study and a 2022 diagnosis of RL fragility across similar problems.

6.0

2026-10-02

RESEARCH · 1 SOURCE · MIT News AI

Book 'Against Tech Oligarchy' chronicles rise and decline of tech worker movement and AI's role

MIT PhD student JS Tan and former Kickstarter employee Clarissa Redwine published Against Tech Oligarchy (Haymarket Books, 2026), a book tracing how tech workers organized successful protests (e.g., Project Maven) and later saw their leverage decline. The authors argue factors including the 2022 market downturn and agentic AI coding systems contributed to reduced worker influence as some companies moved toward classified Pentagon AI work.

5.0

2026-10-02

COMPANIES · 1 SOURCE · The Verge AI

Microsoft expands 'biomimicry' landscaping at data centers to ease local backlash

Microsoft announced plans to apply nature‑inspired “biomimicry” designs—adding native plantings, gardens, and ecosystem restoration—to more than 20 data‑center sites in the US and Germany, scaling work it began at a Netherlands site in 2022 and pledging the approach for new US projects. The move is intended to reduce local opposition over construction, tree removal, noise, water and pollution concerns, though some residents say bigger steps are still needed to address broader environmental impacts from rapidly expanding data centers.

6.0

2026-10-02

MODELS · 1 SOURCE · The Decoder

Microsoft releases MAI-Transcribe-2-Streaming and MAI-Voice-2.1 speech models

Microsoft released MAI-Transcribe-2-Streaming, a real-time transcription model Microsoft says ranks first for accuracy on Artificial Analysis, supports 60 languages, and delivers first partial results in just over 100 ms; introductory pricing is $0.54 per hour of audio. The company also released MAI-Voice-2.1 and a low-latency variant MAI-Voice-2.1-Flash (about 150 ms) that speak 23 languages, can clone a voice from a few seconds of audio, have built-in safeguards, and are available via Microsoft Foundry, the MAI Playground and OpenRouter (Flash priced at $15 per million characters vs $22 for the standard plan, per Microsoft).

7.0