Google released the world model Genie 4, Apple launched a touchscreen laptop, and Microsoft's first-quarter results beat expectations. Nvidia announced it will subscribe to $3 billion of Meta's newly issued shares, and Meta will expand deployment of Nvidia's Vera Rubin computing platform, Vera CPU, and Spectrum-X networking equipment; this is the first time Nvidia has extended this circular transaction model beyond Meta to other "Magnificent Seven" stocks.
Unitree Technology released its 2026 third-quarter financial report on October 28. Third-quarter revenue rose 69.93% year on year to 665 million yuan, while total profit fell 27.78% year on year to 112 million yuan. From 2023 to 2025, Unitree's revenue was 159 million yuan, 393 million yuan, and 1.699 billion yuan, respectively, with a three-year compound growth rate of 226.78%. Net profit shifted from a loss of 11.1451 million yuan in 2023 to 278 million yuan in 2025. The report noted that robots are mainly used in niche areas such as commercial exhibitions and scientific research, and have not yet been deployed on a large scale in production scenarios.
Unitree Technology released its 2026 third-quarter financial report on October 28. Third-quarter revenue rose 69.93% year on year to 665 million yuan, while total profit fell 27.78% year on year to 112 million yuan. From 2023 to 2025, Unitree's revenue was 159 million yuan, 393 million yuan, and 1.699 billion yuan, respectively, with a three-year compound growth rate of 226.78%. Net profit shifted from a loss of 11.1451 million yuan in 2023 to 278 million yuan in 2025. The report noted that robo
Anthropic has filed a prospectus with the SEC, planning to list on Nasdaq under the ticker "ANT", with first-half revenue of US$16.3 billion and operating profit in the second quarter. Founded five years ago, the company has raised nearly RMB 140 billion in the private market; its IPO valuation is expected at US$1.5 trillion to US$2 trillion, with projected compute investment of US$800 billion. OpenAI announced in September that it would not go public this year, leaving Anthropic to push ahead with this IPO alone.
Anthropic has filed a prospectus with the SEC, planning to list on Nasdaq under the ticker "ANT", with first-half revenue of US$16.3 billion and operating profit in the second quarter. Founded five years ago, the company has raised nearly RMB 140 billion in the private market; its IPO valuation is expected at US$1.5 trillion to US$2 trillion, with projected compute investment of US$800 billion. OpenAI announced in September that it would not go public this year, leaving Anthropic to push ahead w
Google has released the large language model Gemini 4 Carbon and the world model Genie 4; the former surpasses its own Gemini 4 Argon and OpenAI's GPT-6 Astra on comprehensive capability benchmarks. Gemini 4 Carbon focuses on vulnerability detection and patching and is currently available only to select cybersecurity experts; Genie 4 adds audio capabilities, allowing developers to add background sound and dialogue to AI worlds, and a waitlist is now open.
Cofense discovered a phishing campaign that impersonates a private Anthropic beta invitation, using branded GoTo Webinar emails to lure enterprise employees into clicking, ultimately stealing Google and Microsoft login credentials. The attack uses URL shorteners and two Cloudflare CAPTCHA pages to block security scans; the fake Google page collects only usernames, while the fake Microsoft page collects both usernames and passwords. Researchers recommend enabling phishing-resistant MFA such as passkeys, and watching for subtle spelling flaws such as the sender "Austin from nthropic".
Cofense discovered a phishing campaign that impersonates a private Anthropic beta invitation, using branded GoTo Webinar emails to lure enterprise employees into clicking, ultimately stealing Google and Microsoft login credentials. The attack uses URL shorteners and two Cloudflare CAPTCHA pages to block security scans; the fake Google page collects only usernames, while the fake Microsoft page collects both usernames and passwords. Researchers recommend enabling phishing-resistant MFA such as pa
Replit has launched Agent Habitat, allowing agents to access enterprise knowledge and systems within boundaries the company can control, building security governance around four principles: least privilege, fault isolation, auditable production changes, and traceable delegated work. Its credential broker retrieves credentials from the vault and injects them only after each request is authorized, so credentials never enter the agent's context; enterprise connector management supports SAML SSO, SCIM, and role-based permissions. Replit said the kernel vulnerability disclosed last week did not affect its existing isolation environment, and it has concentrated on patching tools and deploying firewall protection.
Replit has launched Agent Habitat, allowing agents to access enterprise knowledge and systems within boundaries the company can control, building security governance around four principles: least privilege, fault isolation, auditable production changes, and traceable delegated work. Its credential broker retrieves credentials from the vault and injects them only after each request is authorized, so credentials never enter the agent's context; enterprise connector management supports SAML SSO, SC
a16z's survey of 100 CISOs at companies with more than 1,000 employees shows that 87% expect cybersecurity spending to increase, driven mainly by AI-related threats and the expanded attack surface from enterprises' own AI deployments. All seven of the most-watched security initiatives for 2027 are AI-related; SOC automation received the most first-priority votes, but only 18% have put AI SOC tools into production. 45% plan to make new purchases or upgrades within 12 months for internal AI agent security, and 42% plan to procure offensive security capabilities.
a16z's survey of 100 CISOs at companies with more than 1,000 employees shows that 87% expect cybersecurity spending to increase, driven mainly by AI-related threats and the expanded attack surface from enterprises' own AI deployments. All seven of the most-watched security initiatives for 2027 are AI-related; SOC automation received the most first-priority votes, but only 18% have put AI SOC tools into production. 45% plan to make new purchases or upgrades within 12 months for internal AI agent
At the OpenAI DevDay Exchange 2026 in Berlin, OpenAI said that after talks with regulators, its AI assistant Dots is expected to open more broadly in Europe "within a few weeks"; individual Pro accounts in the European Economic Area, Switzerland and the UK still cannot use it. Weekly active Codex users in Germany have grown twentyfold since January 2026, and nearly one in five desktop users use skills to handle documents, presentations and spreadsheets. OpenAI runs its own agent infrastructure through the Agents API and, with 1.2 billion weekly active ChatGPT users as its entry point, is pushing developers to build their businesses on its platform.
At the OpenAI DevDay Exchange 2026 in Berlin, OpenAI said that after talks with regulators, its AI assistant Dots is expected to open more broadly in Europe "within a few weeks"; individual Pro accounts in the European Economic Area, Switzerland and the UK still cannot use it. Weekly active Codex users in Germany have grown twentyfold since January 2026, and nearly one in five desktop users use skills to handle documents, presentations and spreadsheets. OpenAI runs its own agent infrastructure t
Jessica Sellam, head of Rothschild & Co's private markets group, said at Calcalist's Tech1 conference that AI's impact depends on each company's degree of agility, not on the industry it is in. She believes the sectors in Israel most likely to attract financing are defense and cybersecurity, and noted that there is currently more capital than opportunities, and that investors will become more selective in IPOs, strategic acquisitions and private equity deals.
After WIRED reporter Jason Parham used the AI life-expectancy prediction app Death Clock, he was told he would die on August 10, 2060, at the age of 74; after he added his family history of diabetes, the predicted age dropped to 73. The app launched in 2024 and claims to be trained on 1,200 longevity studies, life-insurance actuarial data, and life-expectancy tables; a $69 subscription adds a basic blood test, and nine other tests cost $780 in total. Mike Ludkovski, a professor at UC Santa Barbara, points out that the triggers of individual death are inherently random and can never truly be predicted.
After WIRED reporter Jason Parham used the AI life-expectancy prediction app Death Clock, he was told he would die on August 10, 2060, at the age of 74; after he added his family history of diabetes, the predicted age dropped to 73. The app launched in 2024 and claims to be trained on 1,200 longevity studies, life-insurance actuarial data, and life-expectancy tables; a $69 subscription adds a basic blood test, and nine other tests cost $780 in total. Mike Ludkovski, a professor at UC Santa Barba
Latent Space spoke with Supabase founders Paul Copplestone and Ant Wilson about why agents choose Supabase as their preferred application backend. The article mentions that Supabase brings roughly 5 million databases online each month, and introduces the Docker-free local runtime alpha, a declarative approach that stores the desired database schema in Git, as well as Scale-side moves such as the private alpha of Supabase Compute, Multigres, and the acquisition of Turso.
AnySpeechRecognizer supports engine switching on top of the Apple SpeechAnalyzer API, with options including NVIDIA Parakeet TDT 0.6B v3 (supporting 25 European languages), Moondream post-trained Parakeet Ultra, the English version Parakeet Unified, and Alibaba Qwen3-ASR (0.6B/1.7B, covering 30 languages and 22 Chinese dialects). These models have been converted to Core ML and published on the Hugging Face Hub; Parakeet takes about 20–40 seconds to transcribe 1 hour of audio, while Qwen3-ASR is slower and requires several GB of memory. The package also adds a speaker diarization module, SpeakerDiarizer, which the Apple framework lacks; it is based on NVIDIA Streaming Sortformer and can track up to 4 speakers.
AnySpeechRecognizer supports engine switching on top of the Apple SpeechAnalyzer API, with options including NVIDIA Parakeet TDT 0.6B v3 (supporting 25 European languages), Moondream post-trained Parakeet Ultra, the English version Parakeet Unified, and Alibaba Qwen3-ASR (0.6B/1.7B, covering 30 languages and 22 Chinese dialects). These models have been converted to Core ML and published on the Hugging Face Hub; Parakeet takes about 20–40 seconds to transcribe 1 hour of audio, while Qwen3-ASR is
Zhongji Innolight, Eoptolink, and T&S Communications saw third-quarter revenue and net profit surge on strong demand for 800G and 1.6T high-speed optical modules, yet their share prices have fallen for several consecutive months from mid-year highs. At the close on October 29, Zhongji Innolight fell 8.53% to 582.05 yuan, Eoptolink dropped 11.06% to 228.40 yuan, and T&S Communications plunged 19.77% to 161.60 yuan, with all three companies' share prices down more than 56% from their peaks.
A New York Times investigative report claims that Nvidia CEO Jensen Huang may be a target of Chinese intelligence activities, and that the US has identified Alice Huang, an executive at the Singapore data center company Megaspeed who has close ties to him, as a suspected Chinese government intelligence operative. Alice Huang is a Chinese citizen; Megaspeed purchases chips from Nvidia and sells computing power to Chinese AI labs including Alibaba. While conducting electronic surveillance of the company over suspected improper resale of advanced Nvidia chips to China, the US found that she was also suspected of passing information to the Chinese government. Nvidia responded that Huang's contact with Alice Huang was not unusual and that customers such as Megaspeed would not receive sensitive trade secrets.
A New York Times investigative report claims that Nvidia CEO Jensen Huang may be a target of Chinese intelligence activities, and that the US has identified Alice Huang, an executive at the Singapore data center company Megaspeed who has close ties to him, as a suspected Chinese government intelligence operative. Alice Huang is a Chinese citizen; Megaspeed purchases chips from Nvidia and sells computing power to Chinese AI labs including Alibaba. While conducting electronic surveillance of the c
Elorian and Mercor jointly released the DecoyCount benchmark, which tests whether models truly count objects or rely on visual shortcuts. Frontier models all scored below 50% accuracy, with the highest being GPT-6 Astra at 43.17%. Elorian says its visual reasoning model Glimpse scored 50.48% in a single run on 1,660 questions, making it the only model to exceed 50%, but this result comes from internal testing and did not use an agent or tool framework, whereas the six models in the paper each took the average of three runs. Elorian CEO Andrew Dai said the benchmark examines both vision and reasoning, and can distinguish whether a model failed to see the objects or miscounted them.
Elorian and Mercor jointly released the DecoyCount benchmark, which tests whether models truly count objects or rely on visual shortcuts. Frontier models all scored below 50% accuracy, with the highest being GPT-6 Astra at 43.17%. Elorian says its visual reasoning model Glimpse scored 50.48% in a single run on 1,660 questions, making it the only model to exceed 50%, but this result comes from internal testing and did not use an agent or tool framework, whereas the six models in the paper each to
Apple has equipped the 14-inch MacBook Pro and iMac with the M6 chip; the former features a 12-core CPU, 12-core GPU and dual 16-core Neural Engine, delivers up to 24 hours of battery life, and offers 30% better graphics performance than the previous generation; the iMac's graphics performance is twice that of the M4 version. The Creator Studio suite will be updated in November: Final Cut Pro adds the MotionVFX effects library and Pixelmator Pro editing features, Logic Pro enhances Stem Splitter and introduces an intelligent assistant powered by a combination of Apple's in-house models and OpenAI, Pixelmator Pro adds natural-language Generate Fill, and Keynote, Pages and Numbers add, in beta form, an assistant that understands document context and MCP support. Both Macs open for pre-order on October 28 and arrive in stores on November 6, with the iMac starting at $1,499.
Apple has equipped the 14-inch MacBook Pro and iMac with the M6 chip; the former features a 12-core CPU, 12-core GPU and dual 16-core Neural Engine, delivers up to 24 hours of battery life, and offers 30% better graphics performance than the previous generation; the iMac's graphics performance is twice that of the M4 version. The Creator Studio suite will be updated in November: Final Cut Pro adds the MotionVFX effects library and Pixelmator Pro editing features, Logic Pro enhances Stem Splitter
At Oracle AI World 2026, Oracle released Oracle App Composer. After employees describe their requirements in a chat interface, OpenAI's Codex generates business application code, which is then deployed to Oracle Cloud after security checks and shared internally via URL. The information department can centrally manage these applications, assign owners, and take offline applications that have long been unused. The product is planned for launch in spring 2027.
Anthropic hiring manager Felix Reiseberg announced that the company will no longer use traditional coding interviews to evaluate candidates, citing the fact that Claude is already stronger than humans at both React challenges and assembly problems, and that truly impactful builders do not win through such tests. He advises junior engineers to pick an idea that can drive business impact, build a working prototype over the weekend, focus on just one good idea, and leverage their fresh perspective on users. On the same day, Google DeepMind released the Genie 4 world model, and Claude Opus 5.5 (58 points) topped the Artificial Analysis Intelligence Index v5.
Anthropic hiring manager Felix Reiseberg announced that the company will no longer use traditional coding interviews to evaluate candidates, citing the fact that Claude is already stronger than humans at both React challenges and assembly problems, and that truly impactful builders do not win through such tests. He advises junior engineers to pick an idea that can drive business impact, build a working prototype over the weekend, focus on just one good idea, and leverage their fresh perspective
NVIDIA has released the open-weight model Nemotron 3.5 Transcribe on Hugging Face and ModelScope, integrating transcription, speaker diarization, translation, and audio question answering into a single checkpoint. It supports transcription in 40 languages, translation between 31 languages and English, and speaker labeling in 13 languages, can process recordings up to 1 hour long, and can label up to 8 speakers. The model ranks first among open-weight models on the English leaderboard of the Hugging Face Open ASR Leaderboard with an average WER of 5.33%, and achieves 2.8% AA-WER on Artificial Analysis with approximately 4B activated parameters. The simultaneously released Granary v2 covers 48 languages and 463,062 hours of speech data, and the accompanying NeMo Speech fine-tuning recipe reduced WER on held-out speakers from 22.44% to 12.17% in a medical case.
NVIDIA has released the open-weight model Nemotron 3.5 Transcribe on Hugging Face and ModelScope, integrating transcription, speaker diarization, translation, and audio question answering into a single checkpoint. It supports transcription in 40 languages, translation between 31 languages and English, and speaker labeling in 13 languages, can process recordings up to 1 hour long, and can label up to 8 speakers. The model ranks first among open-weight models on the English leaderboard of the Hugg
When AI vision systems move from the lab to homes, hospitals, and streets, they face three major challenges: generalization, human behavior, and resources. Most models are trained only on well-lit, high-quality images, and their human pose estimation performance drops noticeably in low-light or nighttime scenes. Using an unsupervised domain adaptation method, the authors' team, in a 2026 study on low-light human pose estimation, substantially improved model performance on two low-light benchmarks without requiring manual joint annotations for low-light images. The number of action combinations in human-object interaction is enormous and difficult to collect and annotate one by one, so the research instead uses large vision-language models to recognize new interactions unseen during training.
When AI vision systems move from the lab to homes, hospitals, and streets, they face three major challenges: generalization, human behavior, and resources. Most models are trained only on well-lit, high-quality images, and their human pose estimation performance drops noticeably in low-light or nighttime scenes. Using an unsupervised domain adaptation method, the authors' team, in a 2026 study on low-light human pose estimation, substantially improved model performance on two low-light benchmark
