Week 39 · 2026

NVIDIA clarifies the parameters of the Nemotron 3 Diarization speaker recognition model

NVIDIA has published additional technical details on the open-weight model Nemotron 3 Diarization: according to the company, it is 41% more accurate than its predecessor and leads the Diarization-Bench leaderboard with an error rate of 14.72% (DER).

New: 41% better error rate than Streaming Sortformer (specific percentage); The model can be combined with Parakeet to create transcripts with speaker tags; Audio buffer adjustable to four levels (30.4–0.32 seconds); Error rate increases with more participants, strong noise, and hares of sound; The Voice Arena benchmark is strict – it counts overlapping speech and misalignments

Spotify launches Taste Profile feature in the US for adjusting recommendations via natural language

Starting Wednesday, Spotify is launching the Taste Profile feature in the US for Premium subscribers (18+), allowing users to adjust the recommendation algorithm for music, podcasts, and audiobooks using natural language. The feature also affects Discover Weekly and Spotify Wrapped and is currently in beta.

AT&T is deploying AI in network operations; an internal source speaks of a target to reduce the workforce to 85 000 by 2030

AT&T is deploying AI in network management, customer service and code development, and starting next year it will replace hardware at thousands of nodes with cloud software from DriveNets. An internal source cites a target of reducing the workforce from 131 000 to 85 000 employees by 2030; the company called this figure inaccurate.

OpenAI hires three former leaders from Patreon for a new Creator Product division, will showcase the first tools at DevDay on 29 September

OpenAI is establishing a Creator Product division led by Sam Yam, co-founder of Patreon, and aims to present the first preview of new tools for content creators at DevDay on 29 September 2026. The former heads of product and engineering also left Patreon alongside Yam.

New: OpenAI DevDay is scheduled for 29 September 2026; Patreon reduced its workforce by 20% and is restructuring its product; Sam Yam is CTO and co-founder of Patreon; Drew Rowny and Shannon Ma stepped down from Patreon in recent weeks; OpenAI is entering the creator monetization business

Czech startup PangeAI won Nápad roku and received an investment of 300 thousand Kč

Startup PangeAI, which is building a geospatial model of Earth for AI agents, won Nápad roku among 189 entrants and received 300 thousand Kč from JSK Investments.

Australia and the US diverge in their approaches to regulating AI and social media

Australian Prime Minister Anthony Albanese is advocating regulation of AI and social media at the UN (the Digital Duty of Care law), while US President Donald Trump rejects any “global control” of AI. The US embassy in Canberra described the Australian proposal as an attack on freedom of speech.

Informal gathering of engineers working in agentic engineering in San Francisco

An informal "Birds of a Feather" gathering for people working on agentic engineering will take place in San Francisco on 14 October 2026 – sharing experimental and unfinished projects through open discussion, with no presentation required.

GitHub added OpenTelemetry support to Copilot for agent monitoring

The GitHub Copilot app now supports OpenTelemetry configuration through enterprise-managed-settings.json — administrators can export agent activity data to their own monitoring tools, with prompt content excluded by default.

GitHub Copilot for JetBrains 1.18.0 adds AI-assisted tool approvals and planning mode for the Codex agent

GitHub has released version 1.18.0 of the GitHub Copilot plugin for JetBrains IDEs with automatic approval of low-risk agent actions, the ability to rewrite an earlier message, organizational skills/instructions, and planning mode in the Codex agent.

Liquid AI and Qualcomm optimized the on-device context layer Liquid Context for Snapdragon processors

Liquid AI and Qualcomm Technologies announced at Snapdragon Summit 2026 that the on-device context layer Liquid Context has been optimized for Hexagon NPU in Snapdragon processors — it is intended to enable device manufacturers to build personalized AI agents with local context processing.

GitHub Copilot CLI has introduced whole codebase indexing for faster analysis of C++ projects

GitHub Copilot CLI has introduced whole codebase indexing for C++ projects – a persistent symbol index across the entire codebase that speeds up searches for definitions, references, and symbols. The feature is enabled by default; in large repositories, initial indexing requires more time and memory.

Snorkel AI tripled its valuation to 3.5 billion dollars after a Series E round of 350 million dollars

Snorkel AI raised 350 million dollars in Series E at a valuation of 3.5 billion dollars (previously 1.3 billion), led by Insight Partners and S32. Annualized revenue run-rate of 375 million dollars, with 18x growth over a year driven by demand for AI training data.

OpenAI has improved prompt caching in the GPT-6 model

According to its own announcement, OpenAI has improved prompt caching in the GPT-6 model - a higher cache hit rate, new diagnostics, explicit breakpoints in prompts, and controls to reduce latency and costs.

Qualcomm introduced Snapdragon 8 Elite Gen 6 and Extreme Gen 6 processors for running AI models locally on phones

Qualcomm announced the Snapdragon 8 Elite Gen 6 and Extreme Gen 6 mobile processors focused on local AI. According to the company, the Extreme version can run a 30-billion-parameter MoE model directly on the phone, while the new sensing hub runs models with up to 200 million parameters for voice agents.

Russia has deployed the fully autonomous V2U drone with AI target selection, while Ukraine is developing similar systems

According to the source, Russia has deployed the V2U drone with a chip made by Nvidia, which independently selects and strikes targets after launch without an operator. Ukraine is developing similar AI-assisted drones that can also operate without a connection to an operator, but they are not yet in mass production.

OpenAI makes GPT-6 Sol and GPT-6 Luna available on the Amazon Bedrock platform

GPT-6 Sol and GPT-6 Luna from OpenAI are now generally available on the Amazon Bedrock platform at lower prices than GPT-5.6; according to OpenAI, GPT-6 Sol makes about half as many factual errors as GPT-5.6 Sol.

A creator on TikTok described typical signs of AI-written scripts for social media

On TikTok, creator @therealcornpop described recognizable signs of AI-written scripts for social media – hackneyed phrases, choppy punctuation, and above all, the lack of an opinion of the author’s own. Simon Willison quoted the commentary on his blog.

Claude Opus 5.5 available on Amazon Bedrock with lower costs per task

Anthropic, together with AWS, has made Claude Opus 5.5 available on Amazon Bedrock – according to the company, it is more efficient and cheaper per task than Claude Opus 5, with adaptive thinking and stricter safety classifiers.

Amazon Bedrock AgentCore adds evaluators for agents with skills

Amazon Web Services added three evaluators (Skill Selection Accuracy, Skill Instruction Following, SkillInvoked) to Amazon Bedrock AgentCore Evaluations and the Strands Evals SDK that measure whether an agent selected the right skill and followed its instructions.

The Claude Opus 5.5 model is available in GitHub Copilot

The Claude Opus 5.5 model from Anthropic is now available in GitHub Copilot for Pro+, Max, Business and Enterprise plans. According to Anthropic, it handles tasks comparably to Opus 5, but with fewer steps and tokens.