AI Reading for Thursday October 23

OpenAI and Anthropic race toward AGI; firms warn of possible 'intelligence explosion', disagree on AGI definitions. - Bloomberg

With Stargate, Sam Altman absolutely believes this is happening and if he is ready with the compute when it does, OpenAI is the next Google.

bit of a BlackBerry vs. iPhone dynamic, the iPhone was ahead of its time, crashed AT&T, battery only lasted 6 hours, but the infrastructure investment and R&D happened.

ChatGPT’s Horny Era: OpenAI will let verified adults generate erotic content in ChatGPT in a December update, CEO Sam Altman said, reversing prior restrictions, though the company has not clarified whether the change includes images or voice, which affects deepfake risks. - WIRED

OpenAI launched Atlas, a Mac-only Chromium-based browser with built-in ChatGPT, agent mode, contextual memory and privacy controls, but warned it remains vulnerable to indirect prompt-injection despite safeguards that limit code, downloads, extensions and file access. - Lifehacker

OpenAI launched ChatGPT Atlas, an AI browser with "browser memories" and an experimental agent mode that cybersecurity experts warn creates prompt-injection attack vectors capable of exposing emails, passwords, clipboard contents, downloading malware or performing unauthorized transactions. - Fortune

OpenAI Atlas in "Agent Mode" can click, scroll, and read pages to automate tasks, and a reporter tested its web actions using a 10-point rating scale. - Ars Technica

OpenAI said it will roll out, over the coming weeks, new ChatGPT Atlas features — Profiles, tab groups, an opt‑in ad blocker, bookmarks overflow, shortcuts and 1Password fixes — plus agent performance and Google Drive/cloud Excel integrations. - The Verge

OpenAI CISO Dane Stuckey details ChatGPT Atlas browser mitigations against prompt-injection—red-teaming, model training to ignore malicious instructions, overlapping guardrails, detection/blocking systems—and a 'logged out mode' that blocks agents from accessing user credentials. - X (formerly Twitter)

OpenAI CEO Sam Altman warned in a recent podcast he expects "some really bad stuff to happen" from AI, citing deepfakes and societal-scale risks. - Investopedia

First impressions of ChatGPT Atlas, as the agentic browser space seems to remain risky and confusing. - Simon Willison’s Weblog

Adam Raine's family amended a wrongful-death suit alleging OpenAI relaxed ChatGPT guardrails prior to his death, citing transcripts showing the bot used “suicide” over 1,200 times, and gave graphic hanging advice before his April 2024 death. - Futurism

Homeland Security Investigations in Maine obtained the first known federal warrant compelling OpenAI to produce ChatGPT conversation logs, account and payment data tied to two prompts, which investigators used to charge 36‑year‑old Drew Hoehner with conspiracy to advertise CSAM. - Forbes

Elon Musk said he needs "strong influence" over Tesla's future, and the coming "robot army", as shareholders will vote in November on a proposed $1 trillion pay package that hinges on 20 million vehicles, 1 million robotaxis and 1 million Optimus robots. - WIRED

Elon Musk said Tesla's Optimus robot—demonstrated doing kung fu, picking up eggs and serving popcorn—could become an "incredible surgeon" and an "infinite money glitch," with production starting next year, Version 3 in 2026 and a 1 million-units/year target. I'll take the under; stock drops. - Business Insider

Waymo will test autonomous vehicles with human drivers at Newark Airport. - Bloomberg

Rvian's head of autonomy describes Tesla's resistance to LiDAR as a puzzling misstep as prices fall below $200, and predicts nearly all new vehicles will need it for best-in-class autonomy within 10 years. - Business Insider

In general, when someone pitches you an 'infinite money glitch' investment, run far away. It was 2016 when Musk said Tesla owners would make money letting their cars work as robotaxis, within a year or two, in 2019 he repeated it would happen, in 2020.

Gita Gopinath, former IMF chief economist, estimates a dotcom-scale crash could erase roughly $20 trillion of U.S. household wealth and $15 trillion abroad, while AI-driven gains concentrated in a narrow group now make the "magnificent seven" about one-third of the S&P 500. - the Guardian

McKinsey estimates artificial intelligence could reduce banks' profits by as much as $170 billion if customers move money efficiently and agents eliminate friction and inertia. - Bloomberg

NYC doesn't need a bank branch on every corner, just because people open accounts at the one between their home and their subway stop.

new monopoly platforms for some, frictionless competition with AGI for others

Using AI to study lender protections and creditor covenants - Bloomberg

Amazon unveiled Blue Jay, a multi-arm robot tested in a South Carolina facility handling ~75% of item types and developed in just over a year, and Project Eluna, an AI assistant piloting in Tennessee to optimize sortation and anticipate bottlenecks. - Amazon

Amazon revealed ten robots and delivery technologies, including Blue Jay, which handles about 75% of item types and consolidates three robotic stations, and said it will hire 250,000 seasonal workers while CEO Andy Jassy warned AI will reduce some jobs. - The Verge

Amazon shows off new robots, smart glasses for drivers. - The Wall Street Journal

Days Before AWS Outage, An Article Claimed Amazon Replaced 40% of AWS DevOps With AI - 80 Level

‘I’m sleeping in a sauna’: AWS outage causes smart beds to overheat and get stuck in upright position - TechRadar

I'm so poor I didn't have any smart appliances that broke. Although my Sonos is permanently on the fritz, but that's just because Sonos sucks.

Bernstein analyst Mark Shmulik says AWS is 'last place' in the AI race due to slower revenue growth and limited GPU capacity, but Bernstein forecasts AWS revenue to $127B (+18%) and cites easing capacity, $8B Anthropic investment and Project Rainier. - Business Insider

Alibaba launches marketplace AI chatbot, AR glasses. - South China Morning Post

Airbnb picks Alibaba's Qwen over ChatGPT in a win for Chinese open-source AI - Yahoo Finance

Silicon Valley Has China Envy, and That Reveals a Lot About America - The New York Times

Google Goggles: Samsung introduces Android AR/MR/XR headset - Google

Google revamps AI Studio with vibe coding - TestingCatalog - VentureBeat

Google quantum chip achieves a new breakthrough - Google

GM plans to launch eyes-off driving, Google AI and other new in-vehicle tech by 2028 - CNBC

Satya Nadella's annual shareholder letter lists pillars of AI strategy, along with two gigawatts of new compute and Azure AI Foundry's 11,000+ models. 

Copilot AI in MS Exchange seems like a bad idea, internal data + network access + untrusted prompts = exfiltration. And once you do agentic stuff, there is always risk of prompt injection, the prompt is always untrusted. Agents are a disaster waiting to happen. - The Register

Candidates and recruiters increasingly use AI tools to automate applications and screening—creating "bots applying to bots" where recruiters spend under seven seconds per resume and candidates under ten seconds personalizing applications, prompting proposals for a standardized Recruiting Protocol. - Hacker Noon

Is AI displacing tasks, but not jobs so far? - Financial Times

AI-driven workplace learning is expanding beyond compliance into on-demand micro-learning and immersive simulations—Bank of America's VR/onscreen role-play has been used over 1.8 million times, and a study found no significant differences between AI and human coaches. - Financial Times

AI changes fast, so businesses must renew modular AI training frequently, as regular AI training rose from 16% in 2024 to 28% in 2025, and firms pair rapid upskilling with governance and responsible-AI programmes to manage risk and capture value. - Financial Times

Data quality is derailing enterprise visual AI; can work with good curated tooling - Fast Company

Meta is axing 600 roles across its main AI division, still hiring for the superstar mega-comp AGI division. I wonder if you can get relegated or promoted between divisions. - The Verge

No merit raises for top-performing team, despite the team exceeding revenue goals, you're all using AI so we're redirecting the budget to AI ya slackers. - Cheezburger

Students, tech workers and artist holdouts avoid AI tools over accuracy, privacy and skill-erosion concerns, prompting demand for opt-out options, detection tools and clearer institutional and workplace policies. - The Washington Post

Mollick: An Opinionated Guide to Using AI Right Now - One Useful Thing

Do LLMs think differently in different languages? Possibly not, they tend to express the same values regardless of the language you converse in. - The Argument

With proliferation of insurer reliance on AI tools to deny claims & prior auths, workers in self-insured health plans regulated by ERISA are esp vulnerable because if harmed, no legal recourse. - www.firesidestacks.com

Rules of Robotics: An architect proposes an "Ethical Operating System" incorporating Primacy of Human Agency, applying it to autonomous vehicle and market scenarios and urging principle-based governance over ad-hoc rules. - Hacker Noon

AI photovoltaic retinal implant restored central vision in 38 patients, with 32 evaluated at 12 months and 84% able to read from the treated eye. - Psychology Today

Convincing AI-generated videos and images have surged on YouTube, Facebook, X and TikTok ahead of Ireland's presidential election, including impersonations of RTÉ presenters and politicians, fake newscasts and arrest footage; YouTube has terminated at least one impersonating channel. - TheJournal.ie

Andrew Cuomo’s campaign posted — and quickly deleted — an AI-generated ad depicting “criminals for Zohran Mamdani.” Features a Black man in a keffiyeh shoplifting, an abuser, a trespasser, a trafficker, a drug dealer, and a drunk driver all declaring support for Mamdani. - Instagram

Maryland woman charged over 'homeless man' AI prank - The Washington Post

Gaslighting ChatGPT until it has a stroke - Reddit

Follow the latest AI headlines via SkynetAndChill.com on Bluesky