WittyWires Hashtags

01#Microsoft Copilot BillingMicrosoft delayed default usage-based billing for some new Copilot Business subscriptions to December 1 and set a default cap of about $40 per user monthly.
02#Microsoft Model CodeMicrosoft proposed consultation rules requiring its future AI models to accept correction, interruption and shutdown; the draft is not currently used to train m
03#Decision 2.0vLLM released Decision 2.0 models for its Semantic Router, spanning 0.6B to 27B parameters under the Apache 2.0 license.
04#Supabase-Turso DealSupabase agreed to acquire Turso, a database company whose product is designed for AI agents.
05#Prime InferencePrime Intellect launched Prime Inference, a platform offering serverless endpoints and reserved GPU capacity for open AI models.
06#SIFTSIFT uses AI judges and fast tree search to select promising coding-agent changes for costly benchmark evaluation.
07#GLM 5.3 FlashPrime Intellect launched an inference service running GLM-5.3 with vLLM, with reported focus on scheduling and reliable tool calls.
08#Minnesota Deepfake BanThe 8th Circuit paused Minnesota’s ban on AI-generated fake nude images while xAI challenges the law.
09#OpenAI DevDayOpenAI DevDay 2026 featured new developer tools for computer-using agents, cloud Codex, and other announcements.
10#LocalAI 4.11LocalAI 4.11.0 adds audio scene labeling, model failover, signed galleries and operational controls.
11#Zig 0.17.0Zig 0.17.0 introduces a redesigned build system with Build Server Protocol support and incremental compilation on x8664 Linux.
12#Accenture EdgeAccenture and AWS introduced six offerings for mid-sized firms, including tools for data discovery and AI-powered virtual-machine migration.
13#Muse Mathematical ResearchMeta says mathematicians using Muse Spark produced six papers, five answering previously open research questions.
14#Helion vLLM BackendA PyTorch-described Helion backend for vLLM linear operations reported throughput gains above 10% on some NVIDIA Hopper workloads.
15#Canada AI CouncilCanada appointed 13 members to a National Council on Artificial Intelligence to advise the federal government on AI strategy, safety and public-service uses.
16#Muse HomeAlexandr Wang said 5,000 Muse Home links should be ready to ship in a few weeks and will be free to Muse subscribers while supplies last.
17#Muse GadgetsMeta announced open-source ESP32 firmware and a Linux SDK for building hardware that works with its Muse assistant.
18#Spanner QueuesGoogle Cloud made Spanner queues generally available, enabling tasks to be enqueued atomically with database updates.
19#OpenAI Policy AppointmentOpenAI hired former national cyber policy official Thomas Lind to lead cyber and strategic risk on its national security policy team.
20#OpenClaw ClawScanOpenClaw integrated Tencent’s AI-Infra-Guard into ClawScan to add an independent automated security review for ClawHub uploads.
21#EKS Kubernetes 1.37Amazon EKS and EKS Distro added Kubernetes 1.37 support, including generally available Metrics API and DRA device taints and beta scale-to-zero autoscaling.
22#ChatGPT financial accountsChatGPT is rolling out financial-account connections for some US Free and Go users, with Plaid and Experian handling the connections.
23#Claude Frontier AcademyAnthropic committed $100 million to train 10,000 Frontier Deployed Engineers by 2028 through its Claude Frontier Academy, starting with Accenture and Bain.
24#ConfigPoisoningGitLab disclosed ConfigPoisoning, a repository-triggered code-execution flaw in DeepSeek-Reasonix Studio, and says fixes are available in newer versions.
25#UnderdogConway Research announced Underdog, a personal AI assistant designed to run on users’ devices, including offline.
26#Ubuntu 26.10 BetaCanonical released the Ubuntu 26.10 beta with Rust-based coreutils and official RVA23 RISC-V support.
27#Lambda GPU LoanLambda secured a fixed-rate loan exceeding $1 billion to purchase more than 30,000 Nvidia GPUs.
28#Supabase SelectAt Supabase Select, the company announced tools for agent-configured backends, local testing without Docker, repository deployment, long-running compute and MCP
29#FieldAI FundingBusiness Insider reports FieldAI raised $700 million at a $10 billion valuation to develop general-purpose AI for robots.
30#Adjudicated QueryAWS published a reference design that uses AI to interpret compliance questions but reserves pass-or-fail decisions for a deterministic rules engine.
31#AWS Agent RLAWS says multi-turn reinforcement learning improved its search agent on three of four held-out benchmarks and reduced BrowseComp-Plus failures.
32#Meta Muse ProfilesDutch eyewear chain Hans Anders suspended sales of Ray-Ban Meta smart glasses in the Netherlands and Belgium, citing privacy concerns.
33#PyRUA-LeanResearchers introduced a robot-agent framework using Python execution and feedback-driven task primitives, reporting higher simulated task success with fewer mo
34#Gboard Federated LearningGoogle deployed Gboard federated learning that trains inside secure server enclaves, using encrypted examples and auditable workload policies.
35#DAYJOB BenchmarkThe DAYJOB benchmark tested agents on 130 healthcare and finance tasks; the top configuration passed fewer than one in four attempts.
36#Context Language ModelsResearchers propose models that edit a working-context file as tasks unfold and report improvements over existing context-management strategies.
37#AstaBriefAi2 open-sourced an 8-billion-parameter model for generating cited scientific reports, alongside model weights, training data and a workflow.
38#OpenAI Medicare BreachAustralia's Home Affairs department ordered federal agencies to review legacy technology and plan risk-based reductions.
39#No-COT-BenchA study found that placing key information last reduced GPT-6.1 Sol’s reported reasoning depth by 16% on tested tasks.
40#ROCm 10.0AMD updated the Ryzen AI Developer Platform OS with ROCm 10.0 and Linux 7.2.
41#Cloudflare ObservabilityCloudflare is consolidating logs, traces, analytics and dashboards, adding a unified SQL API and moving logs and traces to usage-based pricing.
42#Tavus GriffinTavus introduced Griffin, a real-time video avatar model, and says 48% of 54 study participants mistook it for a real person.
43#GLM-5.2Z.ai introduced the open-weights GLM-5.2 model with a High thinking level and a separate coding harness called Z Code.
44#ArtificialNeon became the distributor of Luca Guadagnino’s film Artificial, which is due to premiere at the New York Film Festival on 5 October.
45#SequenceHashTrail of Bits introduced SequenceHash and SequenceMAC, hash-agnostic multihashing constructions with implementations in Rust, Go and Python and a test-vector su
46#AI.SmartSmart and its partners launched AI.Smart in the Philippines, offering subscribers access to multiple AI models through prepaid mobile billing.
47#Rust 1.99.0Rust 1.99.0 stabilises C-ABI variadic functions and raw-pointer layout APIs, and adds standard-library APIs alongside compiler and tooling improvements.
48#SWE-sweepSWE-sweep tests whether agents can find and fix multiple bugs across 4,000 real bugs in 100 repositories and 22 languages.
49#Halluminate Series AHalluminate raised a $30 million Series A to build AI training environments for complex financial work.
50#Claude Sonnet 5.5Anthropic launched Claude Sonnet 5.5, touting faster output and lower task costs for coding and knowledge work.