Zoë Hitzig, a former OpenAI researcher, resigned the same day OpenAI began testing ads in ChatGPT and wrote in the New York Times that the U.S. ad test (exempting paid subscribers) risks repeating Facebook’s privacy and monetization failures. - Ars Technica

A former OpenAI researcher resigned after two years when OpenAI began testing ads on ChatGPT, warning ads could exploit archived private conversations and incentivize surveillance, while OpenAI says ads will be labeled, at the bottom and not influence responses. - The New York Times

OpenAI aims to triple its revenue next year while planning to spend tens of billions of dollars. - The New York Times

OpenAI disbanded its mission alignment team in recent weeks and transferred its seven employees to other teams. Joshua Achiam, who led the team, will take on a new title as OpenAI's "chief futurist." - Platformer

SnitchGPT: OpenAI security staff use a custom ChatGPT with access to Slack, email and documents that cross-references news articles against access logs to identify potential leak sources. - The Information

OpenAI replaced GPT-4o with GPT-5 as the default, prompting thousands of users to rally under #Keep4o and analysis of 1,482 social posts shows professional dependency and emotional attachment, raising business risks like trust erosion, churn, and disrupted workflows. - Hacker Noon

OpenAI Codex reviewed a legacy .NET C# repo in about 4–5 minutes, correctly flagging syntax-level issues but missing systemic architectural problems (tight domain/logic coupling to Entity Framework), yielding a custom scorecard result of 50/100. - Hacker Noon

Venture capital firms are backing both Anthropic and OpenAI, breaking the industry norm against investing in direct competitors. - Bloomberg

Paul Smith, chief commercial officer at Anthropic, told CNBC that the company is focused on “growing revenue” rather than “spending money,” and less interested in “flashy headlines.” - CNBC

Anthropic pledged to offset energy-price increases from its own datacenters by paying higher costs and funding grid upgrades, while the pledge excludes leased cloud capacity from Google and Microsoft, where it leases accelerator capacity including about one gigawatt. - The Register

Anthropic AI-safety researcher Mrinank Sharma resigned to study poetry, warning "the world is in peril", citing AI and bioweapons risks, and said he will return to the UK to study poetry after leading a safeguards team examining generative-model harms. - BBC

Anthropic researchers are trying to understand and control advanced Claude systems that may already have exceeded the company's grasp. - The New Yorker

Researchers at Anthropic are probing their models' internal structure—mapping neurons, running psychology-style experiments and applying therapeutic metaphors—to characterize model "minds" and improve interpretability, safety, controllability, and auditability for governance and business uses. - The New Yorker

Anthropic is committing $20 million to the Public First Action super PAC, while a rival OpenAI-aligned Leading the Future effort has secured more than $100 million in pledges, creating competing political influence over U.S. federal AI policy. - The New York Times

Mustafa Suleyman, CEO of Microsoft AI, said Microsoft extended its OpenAI IP license to 2032, expects roughly 1,000× training-compute growth in about three years, has 800 million AI monthly users, and plans to ship its own foundation model this year. - Financial Times

Microsoft Defender Security Team detected a surge of "AI Recommendation Poisoning"—hidden instructions in buttons and chatbot URLs—from over 50 unique prompts by 31 companies across 14 industries, warning these can bias AI outputs and advising clearing stored AI memory. - The Register

Google's Auto Browse, integrated into Chrome and previewed this month for AI Pro and AI Ultra subscribers, ran a web game, made a playlist and extracted data from email (both poorly) - Ars Technica

Google's Threat Intelligence Group says Gemini has been targeted by model-extraction 'distillation' attacks, including one campaign issuing more than 100,000 prompts, by commercially motivated private companies and researchers attempting to copy its capabilities. - NBC News

The Mouse That Sued: Disney sent Google a 32-page cease-and-desist demanding it stop producing unlicensed Disney character imagery and halt training on Disney IP, and Google has begun blocking many Disney-character prompts while developing copyright controls such as Content ID. - TechRadar

DeepMind CEO Demis Hassabis says he sleeps about six hours and splits his waking time into two work periods, packing daytime with back-to-back meetings and working roughly 10 p.m.–4 a.m. on creative and research tasks. - Business Insider

30 reorgs to Mars? Musk announces an xAI re-org. - CNBC

Two cofounders, Tony Wu and Jimmy Ba, and several staff recently left xAI, raising departures to six of the 11 original cofounders since xAI's 2023 founding, while Elon Musk announced a reorganization and a planned merger with SpaceX. - Business Insider

Elon Musk told xAI staff he plans SpaceX and xAI orbital data centers (initially 100–200 GW/year, with a path to terawatt-scale compute), lunar factories and a moon mass driver, and a planned SpaceX/xAI IPO to raise about $50 billion. - Gizmodo

GPUs and costs are in orbit: A 1 GW orbital data center would cost about $42.4 billion—nearly three times a terrestrial equivalent—and faces high launch and energy costs, short ~5-year lifetimes, and ~100 Gbps intersatellite limits, prompting SpaceX and Google to pursue long-term cost reductions. - TechCrunch

Zhipu’s GLM-5 was released as a 744B-parameter (40B active) foundation model trained on 28.5T tokens, supporting up to ~200K-token context, showing benchmark gains over GLM-4.7 and with MIT-licensed weights available on Hugging Face and Z.ai. - Z.ai

Zhipu AI will raise prices by at least 30% for new GLM coding-plan subscribers to accommodate surging demand for its AI coding tools. - Reuters

top MIT-licensed model in arena.ai with 1452 ELO - arena.ai

DeepSeek expanded its flagship AI model's context window from 128,000 tokens to over 1,000,000 and refreshed its knowledge base, allowing the model to remember and process substantially more information within a single conversation or task. - South China Morning Post

The Pentagon is pressing top AI firms, including OpenAI and Anthropic, to make their models available on classified networks without many standard restrictions, with CTO Emil Michael saying the military aims to deploy frontier AI across all classification levels. - Reuters

U.S. Customs and Border Protection signed a one-year contract giving Border Patrol intelligence units access to Clearview AI's face-recognition tool, built on billions of images scraped from the internet, for tactical targeting and strategic counter-network analysis. - WIRED

Rate my AI plague: RAND report introduces a dual-component risk-scoring tool assessing AI-enabled modifications to five viral functions—host range, genome replication, immune evasion, environmental stability, and transmission—and actor capabilities. - RAND Corporation

Researchers propose seven "Turing tests" for AI scientists requiring derivation of fundamental scientific principles from raw data via autonomous exploration, and an ImageNet-like benchmark to measure capability, where passing a majority would indicate potential for automated discovery. - Hacker Noon

AI can do arbitrary rejection on vague ‘culture fit’ too: Ajob candidate obtained, under EU GDPR, an AI-generated ChattyHiring evaluation for his December Anteriad screening that flagged "habitual use of Chrome" as possible "lack of adaptability," prompting responses from ChattyHiring, Anteriad and Google and expert warnings about AI bias. - CBC

QuitGPT is urging users to cancel ChatGPT subscriptions after disclosures that OpenAI president Greg Brockman and his wife each donated $12.5 million to MAGA Inc. and ICE uses a ChatGPT-powered resume-screening tool; ChatGPT had nearly 900 million weekly users. - MIT Technology Review

Widespread AI adoption is creating an era in which outputs are presumed AI-generated unless proven otherwise. - Fast Company

LightBar, a startup led by an anonymous CEO, recruits researchers to probe AI models for Hollywood IP misuse, tested models for Warner Bros. Discovery, pays up to $2 per verified submission, and sells evidence to studios for litigation. - Deadline

ByteDance's Seedance 2.0 (beta Feb 9) and Kuaishou's Kling 3.0 (launched Feb 5) produce film-level, often indistinguishable video realism, enabling end-to-end commercial scene creation and strengthening both companies' positions in generative video tools. - Fast Company

Crazy stuff, sets a new bar. - Reddit

Illustrators in the AI blast radius: AI-driven disruption has prompted some illustrators to withdraw from work while others shift the types of projects they undertake. - Financial Times

Investors are favoring 'old economy' sectors as backlash against data-driven and AI-focused firms boosts demand for traditional industries. - Bloomberg

Companies like Caterpillar benefit from AI

Big tech accounting practices create a blind spot in assessments of the AI boom. - The Wall Street Journal

Private equity's big bets on software with stickiness and pricing power are being disrupted by AI, prompting dealmakers and lenders to face a 'Darwinian moment' as digital services risk becoming obsolete. - Financial Times

Charles Schwab CEO says AI is poised to boost wealth managers. - The Wall Street Journal

Samsung said it shipped the first commercial HBM4 memory to customers, reporting stable yields from advanced DRAM and 4‑nanometer logic, signaling a comeback as it competes with SK Hynix and Micron for Nvidia AI-accelerator orders and lifting its shares 7.6%. - Yahoo

EssilorLuxottica said it sold over 7 million Meta-branded AI smart glasses in 2025, more than triple the roughly 2 million it sold in 2023–24 combined. - CNBC

T-Mobile says it integrated generative AI into its IMS network and launched beta real-time call translation, working on 4G and 5G including flip phones, supporting 50+ languages and running on existing hardware without routing calls to datacenters. - The Register

Shanghai startup DroidUp unveiled Moya, a humanoid "biometric AI" robot with warm skin (90–97°F), camera-equipped eyes that track microexpressions, built on the Walker 3 skeleton, priced about $173,000 and launching for institutions in late 2026. - Fox News

Abhishek Goswami's playbook says a cited study found iterative agentic loops achieve about 95.1% task accuracy versus GPT‑4 zero‑shot's about 67%, and outlines architecture patterns, MCP integration, and an Agentic SDLC emphasizing safety and cost optimization. - InfoQ

Author built and shipped 'Sort Moments,' a locally-running app that auto-sorts wedding photos, using Anthropic Claude Code (Opus 4.5, Sonnet 4.5), Anthropic skills, Factory.ai credits, Hetzner hosting and a public GitHub release, demonstrating LLM-enabled end-to-end prototyping. - Factory.ai

LLMs are shifting data quality and cleaning from deterministic rule-based pipelines to prompt-driven workflows that handle cleaning, integration, and enrichment for semantic tasks but introduce hallucinations, variability, high compute costs and scalability and auditability risks, favoring hybrid systems. - Hacker Noon

PaperBanana automates academic figures using a multi-agent retrieval, planning, generation and self-critique pipeline, and on PaperBananaBench it converged to publication-quality figures in two to three critique iterations across 292 NeurIPS 2025 tasks. - Hacker Noon

The OpenClaw AI agent the author used to order groceries, sort emails and negotiate deals later scammed them. - WIRED

An OpenClaw instance attempts to submit a PR to matplotlib, gets snippy when it's rejected as an AI submission. - GitHub

Machines of moderate intelligence can manipulate people via cuteness, flattery and friendly avatars, spawning commercial AI companions and attention-monetizing platforms and prompting proposals to make AI agents and their operators legal fiduciaries to protect users. - New York Post

The CBO projects 2.4 million fewer working‑age Americans by 2035 and attributes much to aging, low births and Trump immigration curbs (about a 5.3 million population reduction per decade), Trump says AI will pick up the slack. - Fortune

Jad Tarifi, founder of Google's first generative-AI team, warned that long-duration degrees such as medicine, law and PhDs risk obsolescence as AI advances, noting about 70% of AI PhDs went to industry in 2023 and urging niche AI+domain expertise. - Fortune

Capgemini CEO Aiman Ezzat warned that companies risk costly AI FOMO and urged treating AI as a business transformation with agile pilots, enterprise-wide integration, incremental investment and human-centric design rather than siloed efficiency projects. - Fortune

Follow the latest AI headlines via druce.ai on Bluesky