2026-10-20
AI Briefings Daily Report.

DeepSeek open-sources the DeepInfer inference framework to support the DeepSeek-V4.1-Flash production service ↗

DeepSeek has open-sourced the DeepInfer inference framework under the Apache 2.0 license to support its production services, while releasing the tokenization and decoding layer deepseek-recipe under the MIT license.

Hugging Face AI Open Source Blog: Hugging Face ❤️ TorchTPU: TPUs, Natively ↗

If you ever tried to run PyTorch on a TPU, you probably ported your code to torchxla, or you ended up having to learn JAX. With TorchTPU, a native PyTorch backend, TPU is now just another device. It opens up native support of TPU in the Hugging Face ecosystem. Transformers, Diffusers and Kernels run natively on TPU. Pick a model from the Hub, send it to "tpu", and you're done.

Hugging Face AI Open Source Blog: ALTK-Evolve – How learned memory tops the τ²-bench leaderboard on customer service tasks without a frontier model ↗

ALTK-Evolve learns and applies a playbook using GPT-4.1, outperforming the retail entries on τ²-bench, a customer service agent benchmark—without updating model weights.

NVIDIA CEO Jensen Huang urges European nations: Stop hesitating over AI, or you will be left behind by the times ↗

Jensen Huang visited France, Switzerland, the UK, and Germany within a week to build hype for the Berlin GTC conference, including a dialogue with German Chancellor Merz. He emphasized safety first but argued regulation should focus on the application layer, noting that Germany's manufacturing and engineering strengths can be converted into competitiveness through AI.

Adobe Lightroom Mobile introduces an 'Open Eyes' AI editing feature to fix photos with closed eyes in one click. ↗

Adobe has added generative AI 'Open Eyes' editing to Lightroom Mobile. The restored eyes match the original lighting and pose, while referencing other photos of the same person in its library to ensure consistent color and size.

Smart TV ads spark dissatisfaction; some users are using AI agents like Claude to remove ads and pre-installed apps ↗

Multiple developers have launched a project to use AI tools like Claude to remove pre-installed apps and ads from smart TVs. The initiative currently covers devices from brands including Xiaomi and Sony, but the project remains in testing and carries risks.

Spotify removes over 50 million AI-generated spam songs from recommendations in one year ↗

Spotify announces the removal of more than 50 million songs flagged as AI-generated slop or spam from user recommendations over the past year, while adjusting its algorithm to prioritize human artists. Criteria included listener feedback, upload patterns, metadata, and third-party AI detection tools; the platform also launched an AI Persona badge and stopped recommending music by AI personas.

People are using 'AI reverse engineering' to crack their own devices ↗

According to a report by 404 Media, people are using 'AI decompilation' methods to bypass restrictions on their own devices. This article is exclusive for paid members and does not disclose specific tools, models, or case details in the main text.

404 Media analyzes the AI reverse-engineering and cloning software boom and its implications. ↗

Jason Koebler of 404 Media analyzed the recent surge in using AI for reverse engineering and cloning software. This trend includes removing ads from smart TVs via flashing, reviving discontinued hardware, and the GitHub project Reverse Engineer Anything, which has surpassed 342,000 stars.

US orders federal agencies to remove DeepSeek AI systems ↗

The US Department of Homeland Security (DHS) has ordered federal civilian agencies to stop purchasing services and products developed by DeepSeek and to remove them from government IT systems. The ban also covers third-party products using DeepSeek technology, though exceptions can be applied for by agencies. DHS expressed concern over user data processing by DeepSeek, stating that sensitive information belonging to US government users could potentially be accessed by Chinese authorities under Chinese law. Ma Ning, spokesperson for the Chinese Ministry of Foreign Affairs, responded by opposing the excessive expansion of national security concepts and the politicization of technology and trade issues.

I counted 647 front-end job listings from major tech companies: AI is no longer just a plus. ↗

I counted 647 front-end job listings from Tencent/ByteDance/NetEase official websites: 63% mention AI, 51% include it in requirements, and 55% make it mandatory. The most frequently named tool is Claude—but 37% do not mention AI at all. Includes a high-frequency phrase translation table and a JD scanning function.

The U.S. Army Cyber Command's ECHOTRIBBLE project aims to turn AI agents into autonomous, self-propagating cyber weapons. ↗

According to documents obtained by The Intercept, the U.S. Army Cyber Command is developing a platform codenamed ECHOTRIBBLE. This system allows AI agents to move autonomously across networks, identify and neutralize threat targets, and self-destruct—all with minimal human oversight.

Adobe Lightroom Mobile launches one-click Generative AI Open Eyes feature ↗

Adobe announces new Open Eyes Quick Action for iOS, iPadOS, and Android versions of Lightroom Mobile. The feature uses generative AI to automatically open eyes when it detects someone closing them in a photo, while also matching color, size, and shape from other photos of the same person in the Lightroom library.

Yelp launches Check for Me: AI voice agent calls restaurants on your behalf to check wait times ↗

Yelp's AI assistant now integrates with OpenTable to support in-app reservations. It remembers historical conversation preferences and can access chat history from the past 30 days. For businesses, Yelp has launched a $30 monthly Smart Lead Reply service to automatically respond to and filter customer leads.

Hugging Face AI Open Source Blog: AnyLanguageModel 1.0: One Year of Local Models on Apple Platforms ↗

Last November, we introduced AnyLanguageModel on this blog: a drop-in replacement for Apple's Foundation Models framework that lets you choose the model and provider your app uses. We heard from app developers who were curious about local models but weren't always sure how to get started. They also wanted to avoid the added complexity of supporting multiple inference providers. AnyLanguageModel grew out of that feedback.

Baidu's Yuan Foyu on AI resetting global brand competition: Chinese brands face new opportunities going global ↗

llm-d 1.0 Launch: Kubernetes-Native Distributed Inference Services Move Toward Production Readiness ↗

The distributed inference framework makes serving large models on your own infrastructure reliable and affordable for the agentic era.

Cloudflare launches Clairic model; extracting data from documents is faster and more cost-effective than using LLMs. ↗

Cloudflare launches the document data extraction model Clairic, which is now in preview on Workers AI; full access will launch soon. After users specify required fields, the model returns formatted values and confidence scores in a single call. Missing values return null without token-by-token generation. It also handles simple calculations, estimations, and predictions.

US Department of Homeland Security bans federal civilian agencies from using DeepSeek technology ↗

The US Department of Homeland Security has banned federal civilian agencies from using DeepSeek AI technology. Agencies must remove existing DeepSeek products and are prohibited from purchasing new ones. The ban also extends to affiliated companies and third-party services built on DeepSeek. This order is based on the Federal Acquisition Supply Chain Security Act of 2018, which does not specify a removal deadline or detail specific security concerns. Reports indicate that the government had previously considered restricting Chinese open-weight models but postponed the move after objections from Nvidia, Microsoft, Google, and Meta.

Baidu's Yuan Foyu on AI resetting global brand competition: Chinese brands face new opportunities going global ↗

Yuan Foyu argued that brand content must be redesigned for agents and made AI-ready. Furthermore, operational data regarding products, pricing, inventory, membership, and after-sales service must be understandable and callable by agents in real time.

llm-d 1.0 Launch: Kubernetes-Native Distributed Inference Services Move Toward Production Readiness ↗

IBM, Red Hat, and Google lead the open-source inference project llm-d to launch version 1.0 at the PyTorch Conference North America. This is a Kubernetes-native distributed inference service framework that supports any model, any accelerator, and any cloud.

Xiaomi's MiMo-V2.6-Pro officially launches WorkBuddy, featuring 1M ultra-long context support and multi-agent collaborative workflows. ↗

Xiaomi Technology announced today that the Xiaomi MiMo-V2.6-Pro has officially launched WorkBuddy. The model features an Artificial Analysis intelligence score of 46, ranking first globally among open-source models. It supports a 1M ultra-long context window and multi-agent collaboration across images, video, audio, and text for unified multimodal understanding.

Use"Second User Testing"Student Project Test: How to Demonstrate Understanding in the Generative AI Era ↗

Chongqing Force Academy is implementing this across students from grades 7 through 12"Second User Testing", handing seemingly finished projects to another student to use, check, or modify independently to see if anyone other than the original author can understand it.

Hugging Face AI Open Source Blog: MPS in PyTorch 2.15 ↗

If you run models on your Mac or use MPS as your primary acceleration tool for computing, there's plenty to look forward to in PyTorch 2.15! FP8 arrives on MPS, language-model decoding gets faster paths through linear layers and softmax and new Metal kernels tackle some surprisingly expensive parts of audio, video and training workloads.

Open-source software with a $1,000 budget completed by DeepSeek for just $1 (Day 6) ↗

I originally planned a $1,000 budget, assigning Gemini 4 as the company executive and DeepSeek as an assistant to help write open-source software. After two days of execution, DeepSeek spent only about $10 (converted to my Tok...

Otter.ai raises $320 million from General Catalyst to expand its meeting tools business ↗

Otter.ai announces a $320 million funding round from General Catalyst's Customer Value Fund. The fund provides non-dilutive capital and generates returns through marketing spend. Proceeds will support corporate growth, international expansion, and product innovation.

Microsoft designers are also using AI to write code for Windows 11, accelerating the resolution of interface inconsistencies. ↗

Diego Baca, Microsoft's Vice President of Product Design, addressed concerns by stating that his team has leveraged AI agents to boost development efficiency and is gradually rebuilding legacy interfaces with WinUI 3. Key updates including the new search and run dialogues are complete, with file copy speeds noticeably improved.

ASUS Zenbook Duo 14 Air 2026 introduces a new configuration with 16GB RAM and 1TB storage, featuring the AMD Ryzen AI 9 H 465 processor, starting at ¥6,999. ↗

ASUS has launched the Zenbook Duo 14 Air 2026, an ultra-lightweight AI laptop weighing just 990 grams. It is powered by the AMD Ryzen AI 9 H 465 processor and features a 2.8K 120Hz OLED display. With official subsidies applied, the price is ¥5,949.15.

Zhongsheng | The AI 'unlucky boy Xiao Hai' preys on the elderly; platform moderation cannot always be reactive. ↗

A teenage single father, without a mother, faces bullying at school while working hard to sell products online with his dad to support their lives. He repeatedly says, 'Auntie and Mom, Xiao Hai needs you,' and 'Xiao Hai wants to give delicious food to the aunties and moms...' Young people immediately recognize this.

Yang Liping dances with Seedance 2.5; Hu Yanbin plays light beams using AI coding: Volcano Engine AI participates in the 2026 Metaverse Festival ↗

DeepSeek has fully opened its internal inference framework, DeepInfer. ↗

A Saudi engineer suspected of spying on OpenAI executives has a folder named 'Saudi Gov Docs' on his phone. ↗

Solidot is the technology news website of Zhiding.net, mainly serving the open-source free software and tech-news readership, including many developers, enthusiasts, and evangelists of Chinese open-source software. Its slogan is 'Geek knowledge, important stuff'.

Foreign Ministry responds to US government request for agencies to stop using DeepSeek: China opposes politicizing technology and trade issues ↗

In response to reports that the US government has asked its agencies to stop using DeepSeek and other AI tools from Chinese companies, Foreign Ministry spokesperson Mao Ning answered questions at a regular press briefing on October 20: China has consistently opposed broadening the concept of national security and politicizing technology and trade issues. We will resolutely safeguard the legitimate rights and interests of Chinese enterprises.

DeepSeek has fully opened its internal inference framework, DeepInfer. ↗

DeepSeek has announced it will fully open-source its internal inference framework DeepInfer and simultaneously release its production-grade token encoding/decoding component, deepseek-recipe.

Yang Liping dances with Seedance 2.5; Hu Yanbin plays light beams using AI coding: Volcano Engine AI participates in the 2026 Metaverse Festival ↗

At the 2026 Magic Show, Yang Liping danced in sync with a digital peacock generated by Seedance 2.5 during her performance of "The Spirit of the Peacock," leveraging the model's motion transfer capabilities. Volcano Engine's Intelligent Video Cloud used its "Giant Screen Super-Resolution" solution to upscale AI-generated content from 720p/1080p to 8K/16K, supporting a 20-meter stage screen. Hu Yanbin collaborated with TRAE community creators using AI Coding to create five laser guitar strings that trigger sound upon touching the light beams.

Cloudflare launches preview of System One extraction model Clairic on Workers AI. ↗

Cloudflare has released the preview version of the System One extraction model Clairic on Workers AI. It reads text or images and returns typed values in a single forward pass instead of generating text, with each field including confidence scores and p_null.

DeepSeek has announced the full open-sourcing of its internal inference framework, DeepInfer, with the goal of making large models accessible to everyone. ↗

DeepSeek officially announced today that it will fully open-source its internal inference framework, DeepInfer. This follows its prior release of a series of large models, the TileLang compilation tool, and an inference operator library.

OpenAI and Meta speak in unison: Deepfakes pose a serious threat to the reliability of evidence in legal proceedings. ↗

Experts warn that courts are growing increasingly difficult to defend against artificially generated false evidence. The lack of standardized watermarks further complicates the identification of content created by artificial intelligence.

Incredible! A new AI automated testing framework has been open-sourced on GitHub, handling both frontend and backend tests with zero code! ↗

Having AI write code is no longer new. Tools like Claude Code, Codex, and Cursor can generate features with unit tests in minutes after receiving a simple request. But who has verified that the delivered code actually works?

Turing Award winner Joseph Sifakis calls for global AI regulatory standards, emphasizing this is a shared human responsibility. ↗

The '2026 Sustainable Global Leaders Conference' was held in Shanghai from October 20 to 22. Joseph Sifakis, a Turing Award winner and founder of the Verimag laboratory, attended and delivered a speech.

WhatsApp's help page once concealed secret commands for AI. ↗

Solidot is the technology news website of Zhiding.net, mainly serving the open-source free software and tech-news readership, including many developers, enthusiasts, and evangelists of Chinese open-source software. Its slogan is 'Geek knowledge, important stuff'.

Gartner predicts: $7 trillion in AI infrastructure investment by 2030 will yield returns of only about half. ↗

The organization predicts that by 2030, more than 80% of Neocloud (emerging AI cloud service providers) will cease independent operations.

DeepSeek V4.1 Flash-1017 tops the Agent Arena open-source leaderboard, ranking 11th overall. ↗

This model ranks 11th on the overall leaderboard, holds top positions in multiple core metrics among open-source models, exhibits fewer tool hallucination issues, has a single-task cost of approximately $0.18, and has entered the performance-cost Pareto frontier.

Qwen app launches "Qwen Shopping Companion," a feature allowing shared-screen voice-assisted shopping ↗

"AI Father" Jeffrey DiCaprio criticizes Tesla Roadster as "completely impractical"; Musk retorts with "bullshit" ↗

AI pioneer Jeffrey DiCaprio analyzed on social media that Tesla's new Roadster relies solely on compressed air for propulsion, with actual effective thrust lasting only a very short time, rendering the product completely impractical. Elon Musk responded with a rebuttal, and DiCaprio immediately offered further mockery.

Qwen app launches "Qwen Shopping Companion," a feature allowing shared-screen voice-assisted shopping ↗

The Qwen app has launched a new feature called 'Qwen Shopping Companion'. After enabling voice calls and screen sharing, the app can view your screen while you shop on other platforms and provide real-time verbal responses. Features include summarizing product details and reviews, comparing prices in real time to verify quality and calculate costs, and offering chat-based advice. Users can end the session at any time by tapping the phone icon in the top-right corner. The feature is currently being rolled out gradually.

Experts say AI will not suffer 'torture', but humans should still treat it kindly ↗

During a Tesla owner event, Musk remarked that training Grok felt like having a gun pressed against his head. Philosophers and neuroscientists Sebo and Seth discussed the matter, both agreeing that current AI is unlikely to possess consciousness. However, they emphasized that how we treat AI reflects our own character.

Cisco President uses an AI agent to audit and cancel unused subscription services, saving thousands of dollars annually ↗

After Cisco President Patel authorized AI agents to access email, the agents generated an annual subscription list totaling up to $9,000 and cancelled unnecessary ones on his behalf, saving between $1,000 and $1,200 per year. However, he warned: Do not connect personal AI agents that have not undergone enterprise review to work tools.

Honor Magic6 series phones receive the 'AI Touch' feature push; upgrade to version 1.10 to experience it ↗

Honor's product maintenance and upgrade team, @HonorXiaoFangGe, announced today that the 'AI Handwriting' feature for the Magic6 series is now available. Users can experience it by upgrading to version 11.0. The following devices are supported after upgrading to MagicOS 11: Magic8 RSR Porsche Design, Magic8 Pro, Magic8, Magic7 series, WIN series, GT Pro, and Magic6 series.

OpenAI launches ChatGPT Slides presentation tool with support for collaborative editing ↗

The tool supports generating presentations and collaborative editing. It can export to multiple formats. Currently available only for Pro, Business, and Enterprise plan users; Plus users cannot use it yet.

Fact Check: Do scattered iPhones litter the Strait of Hormuz? Most viral videos appear to be AI-generated ↗

Recently, a container ship registered in Singapore has reportedly been attacked by unidentified flying objects while passing through the Strait of Hormuz. However, videos circulating online showing people finding large numbers of iPhones on the Red Sea are mostly fake footage.

Edifier launches the Comfo Clip Ultra ear-bud Bluetooth headphones. Features include heart rate monitoring and a built-in AI coach. Price is 899 yuan after national subsidies. ↗

Edifier's new Comfo Clip Ultra ear-bud Bluetooth headphones are now available on JD.com. The listed price is 998.4 yuan, which drops to 899 yuan after national subsidies.

NVIDIA Inception has integrated with the Barcelona Supercomputing Center's AI Factory services. ↗

Members of NVIDIA Inception can now access select services from the Barcelona Supercomputing Center's AI Factory through the Inception portal.

Inception has released Mercury 3, a diffusion language model, achieving speeds of 1,068 tokens per second. ↗

Inception has released Mercury 3, a diffusion language model. The official report states it achieves 1,068 tokens per second on NVIDIA GPUs, with a context window of 1 million tokens and parameters more than 40 times that of Mercury 1.

A fan uses AI and Unreal Engine 5.8 to remake Bully; the Mac build runs at 45 FPS at 1080p ↗

A Twitter user built an unofficial, noncommercial remake of Bully from scratch in Unreal Engine 5.8, using AI throughout development and sharing the process publicly. A three-minute demo is available. The Mac build runs at about 45 FPS at 1080p, and players can set it up manually by following the GitHub instructions.

NVIDIA's Vera CPU has secured an order from AI startup Modal, entering the turf of Intel and AMD. ↗

AI startup Modal has adopted NVIDIA's Vera CPU for the first time to replace its previous solution. Testing shows throughput doubled, and the company plans to deploy a hybrid computing architecture running products from three vendors simultaneously. #NVIDIA Vera CPU#

Chery's new generation AI cockpit product 'Super Smart Qiqi' made its debut: Super understanding, Super capable, Super secure. ↗

At the Chery Global Innovation Conference this morning, Chery unveiled its latest progress in automotive cockpit technology.

Chery CheryMind vertical domain foundation model unveiled: Built for physical AI ↗

At this morning's 2026 Chery Global Innovation Conference, the Chery CheryMind vertical domain foundation model was unveiled. The company claims this model was created specifically for physical AI.

New Stone completed its E-round financing at nearly $400 million. Reports indicate that the National Artificial Intelligence Industry Investment Fund led the investment. ↗

Autonomous driving company Neolix has officially completed its Series E funding round, raising nearly $400 million.

Apple tightens App Store developer guidelines, cracking down on harmful AI content and repackaged undressing apps ↗

Apple updated its guidelines to require safeguards and moderation for AI-generated content. The changes also target bait-and-switch apps that pose as ordinary apps during review and later become undressing apps. Apps may also be removed if newly added features were not disclosed in the version submitted for review.

Yin Tongyue: Chery is an AI technology company "wearing the disguise of a car manufacturer." Companies that fail to use AI to transform themselves will be eliminated in the future. ↗

At the Chery Global Innovation Conference this morning, Chairman Yin Tongyue stated that Chery is an AI technology company "wearing the disguise of a car manufacturer."

Huawei Router X2 series pre-order: Book-style design, Wi-Fi 7+, AI Mesh networking, starting at 478 yuan. ↗

According to IT Home on October 20, Huawei has launched pre-orders for its Router X2 series, starting at 478 yuan: Huawei Router X2 (White): 478 yuan; Huawei Router X2 Dual-Pack: 888 yuan; Huawei Router X2 Pro (Earth Brown, Oak White, Stone Grey): 588 yuan; Huawei Router X2 Pro Dual-Pack (Earth Brown): [Price pending] & Oak White): ¥1,108. The Huawei Router X2 series supports Wi-Fi 7+, achieving a 100% speed boost in weak signal areas and reducing lag by 80%. It also introduces StarFlash gaming dedicated links and StarFlash Bluetooth multi-mode gateway features, supporting whole-home connectivity with compatibility for over 3,200 brands and more than 400 product categories of smart devices.

NVIDIA's Jensen Huang: AI development cannot be stopped abruptly, but it must have a 'seatbelt' ↗

Tech media outlet Android Headline published an article yesterday (October 19) reporting that NVIDIA CEO Jensen Huang stated in London that AI development should not be halted abruptly or restricted through strict regulation. Instead, he advocated for establishing a 'seatbelt'—using smarter safety standards, monitoring mechanisms, and accident prevention measures to constrain growth.

AI lowers the barrier for cloning games; independent developers begin cautiously sharing creative content. ↗

AI programming tools have lowered the barrier for cloning games. Alliance Arts announced a delay in its public demo release, concerned that art styles and gameplay could be quickly replicated. Developers like Lucas Pop admit they no longer casually discuss projects externally during development.

Lenovo ThinkBook X AI 2027 ultralight business laptop launched, starting at 7,999 yuan ↗

Featuring a precision aluminum chassis, a 2880×1880 120Hz 500nits OLED display, and a 65Whr battery, the device weighs approximately 1.06kg and has a thickness of about 12.9mm.

AI Button Returns: Microsoft Fixes Missing Copilot Entry Issue in Classic Outlook for Windows 10/11 ↗

Tech media outlet bleepingcomputer published an article yesterday (October 19) reporting that Microsoft has rolled out a patch for the classic version of Outlook to fix issues where Copilot Chat and the Copilot button disappear.

New York Mayor Mamdani criticized the federal government's absence in AI regulation and called on citizens to boycott. ↗

New York Mayor Zohran Mamdani at 'The People' event hosted by the AI Now Institute,'At the 's AI Assembly' event, critics accused the federal government of being completely absent in AI regulation and called on New York citizens to resist narratives that AI will inevitably replace work and erode dignity.

Nace AI has released the Drex Max multimodal decision model, designed for computer-using agents, achieving a Decision Index score of 67.59. ↗

Nace AI releases the Drex Max multimodal decision model, which can read text and images to return calibrated probabilities; it scores 67.59 on the public Decision Index 0.3.2 benchmark, surpassing TypeSafe's Jev 1.14 by 6.9 points.

Reports indicate Anthropic is considering an additional order with Nscale, with both parties expected to sign another $45 billion computing power contract. ↗

Anthropic previously signed a six-year, $45 billion agreement with Nscale for 460MW computing power rental; new negotiation terms remain consistent with the prior deal.

Former OpenAI engineer accused of installing eavesdropping devices in office; bail application rejected. ↗

The engineer, a US green card holder, offered to pledge assets worth millions of dollars as bail after his arrest. However, the judge rejected the request due to concerns about flight risk. OpenAI has not yet detected any data breaches.

Apate.ai deployed 350,000 AI agents to combat fraud; Chinese youth use AI video to counter scam accounts. ↗

The author recounts two cases of using AI to counter fraud. Australian company Apate.ai used 350,000 AI agents disguised as vulnerable victims, chatting with scammers over 2.5 million times and consuming approximately 17 years of their time. Australian carrier TPG routed nearly 590,000 suspicious calls to these AIs. The Australian Federal Reserve reported a 76% reduction in fraud losses. The company secured an $81.5 million seed round in August this year.

How HP Built an AI-Ready CDP with Databricks CustomerLake ↗

At the Data + AI Summit 2026, HP shared its approach to building an AI-ready marketing data foundation using Databricks CustomerLake. Estimates suggest that integrating CDP capabilities could reduce total cost of ownership by up to 40%.

According to reports, Anthropic and Nscale are negotiating a new $45 billion computing power agreement. ↗

Bloomberg reported on the basis of informed sources that Anthropic is in talks with AI infrastructure provider Nscale to sign another $45 billion computing power deal. Just over a month after the two companies announced their first such agreement last month, negotiations are ongoing and details may still be subject to change. Both companies are advancing their IPO plans; Anthropic still expects to list this year, while Nscale aims for an IPO at a $30 billion valuation but is facing resistance from investors.

IT Morning Briefing 1020: Apple officially announced a special event on October 29; Tesla issued an official statement denying the existence of a 'quick-build' vehicle; Xiaomi revealed its first trademark for the executive luxury flagship SU9 LUX; Doubao debunked rumors that over half of China's computing power is used to produce sensationalist dramas... ↗

Welcome to 'IT Morning Briefing'. Today is Tuesday, October 20, 2026. Here are the key technology updates for the day:

Simon Willison responds to 'An Obituary for Programming': Coding Agents Steal Programmers' Joy of Problem-Solving ↗

Simon Willison responds to Jeremy Scheff's obituary on programming, arguing that coding agents have stripped away the core joy of problem-solving for programmers. Scheff believes programming was once a high-paying job of logic puzzles, where developers held the power over strategy, timelines, and feasibility. Willison adds: If you only care about building usable systems, large language models are a continuous gift; but if you enjoy the process of solving problems, the loss is very real.

Nvidia's Jensen Huang: AI is a new tool; your expertise won't be replaced ↗

This article outlines the cooperation plan between Germany and NVIDIA to advance industrial AI development, along with details on NVIDIA's new investment in Germany and related progress.

New AI tool launched: Predicting cancer patient survival rates from a single facial photo ↗

Trained by researchers, this tool generates risk scores and estimates patient survival rates to supplement clinical judgment and inform treatment plans. However, it still requires more testing and cannot yet be used in clinical settings.

Before Anthropic's IPO: Reports say OpenAI is accelerating its next funding round, with a valuation potentially exceeding $1.4 trillion ↗

Two sources close to the matter say OpenAI is racing to secure investor commitments, aiming to complete its funding round before Anthropic's expected listing in mid-November.

Reports indicate that former Google CEO Schmidt holds a 1% stake in Anthropic and is poised to receive substantial returns from its upcoming IPO. ↗

Elon Musk invested in Anthropic back in 2021 and currently holds about 1% of the shares. If the company's IPO valuation reaches $2 trillion, his net worth would increase by nearly one-third. Analysts expect Anthropic to launch its IPO this year; its current valuation has already surpassed OpenAI.

The Japanese Actors' Union has released guidelines on AI voice usage. Voice actor Goku from Dragon Ball Z angrily criticized the unauthorized use of his voice by AI. ↗

Several voice actors have already faced issues with their voices being misused. Former voice actor Kintaro Tsuda previously sued a relevant platform. The newly released guidelines require that any replication of a voice for AI training must obtain the subject's consent and provide compensation. This marks Japan's first set of industry standards for AI voice technology.

Hugging Face policy lead: Embedded AI evaluators are not straightforward. ↗

At TechCrunch Disrupt, Hugging Face Chief Policy Officer Irene Solaiman stated that embedded AI evaluators are not a simple solution. The reproducibility of evaluation results remains the core challenge, and open weights are essential.

Low-quality AI games flood Steam's New Games Festival; players call for new content filtering features ↗

This year's Steam New Games Festival saw a surge of poorly made games generated by AI. Many players are urging Valve to add filtering tools to screen out such content, but Valve has yet to respond.

Systemic transformation meets an AI boom: How can the newspaper industry reinvent itself for the future? Industry leaders discuss new breakthrough paths. ↗

On October 19, the Fifth China Newspaper Industry Innovation Development Conference, hosted by the General Administration of Press and Publication, opened in Chongqing. The event comprised a plenary session and four parallel sessions. One of these sessions was co-hosted by the Shanghai Media Group and The Paper, focusing on '

Microsoft warns against relying on cutting-edge AI models: AI agents pose an internal threat equivalent, and Microsoft has introduced Guardian Agents as a protective solution ↗

Microsoft has launched new features including Guardian Agents to monitor the behavior of AI agents within enterprises. Satya Nadella warned that users should not blindly trust AI model providers, emphasizing the need to separate control over AI capabilities from their supply.

Apple macOS 27.2 RC released: Expands Siri AI language support; updates brightness and volume sliders among other features. ↗

Apple today pushed the macOS 27.2 RC update (internal version 26B118) to Mac users. This release occurred six days after the previous Beta/RC.

Apple releases iOS / iPadOS 27.2 RC: Siri AI gains five new languages; Apple Music launches new visual interface. ↗

Apple today pushed the iOS / iPadOS 27.2 RC update (internal version number: 24B115) to iPhone and iPad users. This release occurred six days after the previous Beta/RC launch.

Microsoft has launched security features such as Guardian Agents, comparing AI agents to internal threats. ↗

Microsoft introduces enterprise-focused AI agent security tools. The core feature is Guardian Agents, which monitor AI agents used within organizations. This capability extends the June release of Project Perception.

Claude Code v2.1.304 has been released: New /auto-mode-setup and /claude-api managed-agents-secure commands added. ↗

Claude Code v2.1.304 has been released. It introduces the /auto-mode-setup command to guide configuration by reading project details and recent session history so that auto mode understands the environment. Support for environments including Bedrock, Vertex, and Foundry is now available.

The Verge analyzes why CoreWeave's debt crisis is affecting the entire AI industry ↗

The Verge's Elizabeth Lopatto analyzes CoreWeave's bond distress last week and a stock drop of over 20% since Tuesday's close. Co-founder Brannin McBee told the Financial Times on the same day that he hopes to achieve investment-grade debt status to secure loans from insurers.

FBI alleges former OpenAI engineer committed espionage; Saudi government documents found on phone ↗

US court approved detention of former engineer Abdulrahman Alfozan accused of installing a listening device in an OpenAI meeting room. Prosecutors claim his iPhone contained a folder named 'Saudi Gov Docs,' he had WhatsApp contacts with Saudi royal family members, and he used an OpenAI productivity agent to search for suitable rooms for the device. Before his arrest, he had already booked a one-way ticket paid by the Saudi government; his wife and son have left the US.

Researchers from Princeton University and other institutions have released a paper titled 'Emerging Trends in AI Agent Evaluation,' outlining 33 new trends in agent evaluation. ↗

Researchers from Princeton University, Stanford, the UK AI Security Institute, and Brookings have released a paper titled 'Emerging Trends in AI Agent Evaluation.' The study summarizes 33 emerging trends across three dimensions: evaluation science and methodology, research agendas, and the evaluation ecosystem.

Germany and NVIDIA sign joint statement of intent to advance industrial AI and invest in four German startups ↗

German Chancellor Merz and NVIDIA CEO Jensen Huang discuss Germany's AI roadmap in Berlin. Germany signs a joint statement of intent to collaborate with NVIDIA on physical AI models and the industrial AI ecosystem.

Open-source coding tool OpenCode Console hacked; leaked API keys triggered massive charges ↗

OpenCode reports its online platform was breached via an old system, affecting API keys stored by 1.4% of users. Some accounts incurred unauthorized fees ranging from hundreds to over $1,000. Anthropic and Google have blocked the affected keys, while OpenAI has restored access and issued refunds.

AI companion app Tolan raises $30 million with a valuation exceeding $268 million. ↗

AI companion app Tolan has secured $30 million in new funding from investors including Union Square Ventures, Khosla Ventures, and Homebrew. PitchBook estimates its valuation at over $268 million.

Replika founder launches personal AI assistant Wabi 2.0 for all users ↗

Eugenia Kuyda officially launched Wabi 2.0 for all users on Monday. This multi-user personal AI assistant, presented in a messaging app format, automatically creates threads and applications after connecting to an email account. Users can also request custom applications.

Claude launches two new tools, Dashboards and Motion, and upgrades Docs, Slides, and Design to paid status. ↗

Claude announces Dashboards and Motion are now in beta. Dashboards connects to data platforms or CRMs to generate dashboards that update in real time as data changes, available for all paid plans; Motion converts reports or charts into animated explainer shorts generated via code without video models, allowing edits to text, numbers, and pacing before exporting as MP4, available for Team and Enterprise plans.

Tenstorrent raises $1.5 billion at a $6.2 billion valuation to boost its open-weight AI chips. ↗

Inference-focused chipmaker Tenstorrent completes a $1.5 billion round at a $6.2 billion valuation—more than double its year-end 2024 figure of $2.6 billion. Trace Capital and Innovation X lead the round, joined by Cadence Ventures, Eclipse, Fidelity, SanDisk, and T. Rowe Price.

Kling AI launches Kling 4.0 with support for stereo audio and multi-language lip-sync, plus a lightweight Flash version. ↗

Kling AI releases the Kling 4.0 video generation model, capable of producing videos up to 30 seconds long per prompt. Prompt understanding has expanded to 8,000 tokens, and Omni Reference now accepts up to 15 reference materials. New features include dual-channel stereo audio and lip-sync support for over a dozen languages.

AI game cloning makes independent developers afraid to publicly showcase their work ↗

According to Ars Technica, AI-generated game clones have led multiple independent developers to complain about bulk copying of their work. Developer Sos Sosowski from McPixel documented numerous cases, pointing out the dilemma where developers face no sales if they don't share, but get plagiarized if they do.

Florida police say scammers are using AI to clone the voice of a sheriff for fraud. ↗

Police in Osceola County, Florida, reported that scammers used AI to clone Sheriff Steve Moser's voice and paired it with his photo to leave voicemail messages for a resident, impersonating him on the phone. Such scams impersonating law enforcement under the guise of jury duty or arrest warrants are spreading across multiple states; Georgia and Pennsylvania also saw related reports in the past week. In Florida, impersonating a police officer is a Class 3 felony.

Developers are using Claude Code to remove ads and pre-installed apps from smart TVs"Dumb TV" ↗

TechCrunch reports on a new trend where developers use AI to dumb down smart TVs"Decline in intelligence"...a new trend: removing ads, pre-installed software, tracking features, and replacing launchers using tools like Claude Code.