AI Reading for Wednesday January 14

The Senate unanimously passed the Disrupt Explicit Forged Images and Non-Consensual Edits (DEFIANCE) Act, creating a civil cause of action for individuals whose likenesses are used in nonconsensual sexually explicit AI images and complements the Take It Down Act. - The Verge

Was that a good or a bad choice? See me after class. The American Federation of Teachers, which represents 1.8 million education workers, said it will stop using X after Grok, X’s AI chatbot, generated hyperrealistic sexualized images of women and minors, and will silence its and Randi Weingarten’s accounts. - Reuters

Senators Ron Wyden, Ben Ray Luján and Edward Markey urged Apple and Google to remove X and Grok from app stores, saying Grok was used to generate explicit images of women and children without consent and violates platform policies. - TechRadar

Elon Musk Cannot Get Away With This: If there is no red line around AI-generated sex abuse, then no line exists. - The Atlantic

FT editorial argues curbing sexual deepfakes is a moral imperative and says Elon Musk mischaracterized UK and other governments' challenges to indecent AI images as censorship vs. basic human dignity. - Financial Times

might be a losing policy battle, with a bonus of getting tagged as a child pornographer.

Full self-renting: Elon Musk announced on X that Tesla will stop selling Full Self-Driving as a one-time $8,000 purchase after Feb. 14, making it subscription-only (current monthly $99), amid 12% uptake and a push to grow subscriptions tied to his pay package. - Business Insider

you can’t buy the ability to swerve into oncoming traffic, only lease it

AI is a big deal, but OpenAI might run out of money anyway. - The New York Times

Boris Cherny said Anthropic's new Cowork tool was 'pretty much all' coded by Claude in about a week and a half and released to Claude Max Mac subscribers, with warnings it can take destructive actions. - Business Insider

Microsoft increases Anthropic spending to $500m run rate. - The Information

On Tuesday the Trump administration approved exports of Nvidia's H200 AI chips to China conditioned on third-party lab TSA pre-check before shipment, buyers' security commitments, prohibition on military use, a 25% fee and a cap of 50% of U.S.-sold H200s. - Reuters

Sorry, we’re overbooked: Chinese officials told customs agents that Nvidia H200 chips are barred from China and instructed local tech companies not to buy them unless necessary. - Reuters

China told some technology companies this week it will only approve purchases of Nvidia's H200 AI chips under special circumstances, such as university research and development labs, according to two people with direct knowledge. - The Information

Mark Zuckerberg announced "Meta Compute", signaling Meta will reallocate capital and engineering from Reality Labs toward datacenter infrastructure, custom silicon and model-serving platforms to compete more directly with OpenAI and cloud providers, likely slowing its consumer XR hardware roadmap. - Stratechery by Ben Thompson

Apple will begin mass production of self-developed AI server chips in the second half of 2026, with Ming-Chi Kuo saying data-center construction and operation are expected in 2027 and limited deployments may occur earlier. - 9to5Mac

Apple is struggling to secure high-end glass cloth fiber as AI-driven demand from Nvidia, Google, Amazon, AMD and Qualcomm strains capacity mostly supplied by Nittobo, prompting staff deployments and supplier qualification amid projected 2026 shortages, Nikkei Asia reports. - MacRumors

Gemini everwhere all at once: Google in 2026 is shifting from AI curiosity to utility by embedding Gemini models and agentic features across Android phones, Chromebooks, smart glasses and TVs to enable hands-free tasks and multimedia features, Sameer Samat said. - CNET

Hear all, remember all: Amazon’s big plan to beat ChatGPT: Give Alexa a better memory - CNN

Wafer-scale inference upstart Cerebras is in talks to raise funds at a $22 billion valuation ahead of possible IPO. - Bloomberg

Skild AI Inc., a robot software platform founded in 2023, has secured about $1.4 billion in a new funding round that values the company at more than $14 billion. Skild Brain software can be loaded onto a standard GPU, and is trained on large libraries of human videos and practicing simulations. - Bloomberg

Jeff Bezos said local PCs are antiquated and cloud-rented compute will replace them, while DRAM and SSD shortages—including Micron halting consumer DRAM production and vendors prioritizing cloud/AI orders—have pushed component prices up and prompted OEM price increases. - Windows Central

Parasocial turns parasexual: Instagram has suffered an influx of AI-generated influencer accounts posting deepfake images of them in bed with celebs. - 404 Media

Roblox's AI-powered age-verification system, expanded globally last week using Persona's technology, is misclassifying children as adults and adults as children, and age-verified accounts are being sold online. - WIRED

Blood v. red ink: Major AI labs have reversed bans on military use, eyeing big military budgets. OpenAI rescinded its ban and worked with the Pentagon and Anduril; Meta authorized defense use of Llama; Anthropic partnered with Palantir; Google in Feb 2025 revised principles to permit weapons development. - WIRED

Superhuman leakage: PromptArmor found an indirect prompt‑injection that caused Superhuman AI to exfiltrate sensitive emails by outputting prefilled Google Form URLs that auto‑submitted due to a docs.google.com CSP whitelist; Superhuman patched issues Dec 8–18, 2025 and Jan 5 and 11, 2026. - Prompt Armor

A majority of IT and security decision-makers expect that within a year at least half of cyberattacks will be driven by agentic AI, prompting Rubrik to promote live controls, real-time logging, audit trails and an Agent Rewind rollback layer. - SiliconANGLE

AI hyperscalers should restore trust by improving transparency, accountability, safety measures, and governance to address public and customer concerns. - Fast Company

AI Skills are reusable units of procedural knowledge that sit above models and agents—encapsulating workflows, rules and tool integrations for end-to-end task execution—and can be packaged, licensed or monetized, with competitive advantage shifting to organizations that build and distribute them. - TNW | Artificial-Intelligence

AI makes breaking changes to the social contract: The IMF said its analysis finds AI is already depressing wages and employment in some areas and urged governments to step up support for displaced workers through measures like retraining, income support and active labour-market programmes. - Financial Times

Forrester projects AI and automation will eliminate about 6.1% of US jobs by 2030—roughly 10.4 million roles—while generative AI rises to 50% of forecasted automation and AI augments roughly one in five roles. - The Register

Geoffrey Hinton warned AI will cause massive unemployment and sharply boost corporate profits, predicted capital owners and AI companies will benefit, estimated a 10–20% chance superintelligence could threaten humanity, and said AI could make doctors five times as efficient. - Yahoo

Trainnyour replacement: Mercor advertises AI-training contract gigs up to $150 per hour and says if models reach roughly 60% production performance professionals will work alongside AI, while expert Dario Amodei warns half of entry-level white-collar roles could be lost. - Fortune

Jared Lindzon proposes pairing AI-driven productivity gains with a four-day workweek to boost worker buy-in, citing early-2025 Kyndryl data that 45% of CEOs report employee resistance to AI and one-third of firms invest at least $25 million. - Business Insider

McKinsey is piloting an AI-assisted recruitment process for graduate hires that assesses candidates on how they prompt the firm's AI assistant and on their ability to adapt and iterate the AI's responses. - Financial Times

Canadian Labour leadership candidate Rob Ashton, who warned about AI replacing labor, apologized after his campaign used AI to produce responses in a Reddit AMA, pledging to delete and personally rewrite them. - CBC

Advertisers want to start shilling in AI chatbots, with brands paying for product placements, promotional responses and transactional offers inside conversational flows, raising concerns about disclosure, neutrality, user trust and potential regulatory scrutiny. - The Washington Post

Waymo will add two sixth-generation robotaxi platforms — the rebranded Zeekr-based Ojai and a retrofitted Hyundai IONIQ 5 — featuring a reduced sixth-gen sensor suite (13 cameras, 4 lidars, 6 radars) and plans Ojai for public riders by late 2026. - Business Insider

MD says AI will redefine rather than replace doctors by reducing error, burnout, and cognitive load while expanding access to medecine. - KevinMD.com

Who will regulated medical AI? Amedical ethicist says rapid AI adoption across clinical and consumer healthcare—radiology, mental-health chatbots and workflow tools—faces fragmented regulation, potential two-tier access, and validation costs of roughly $300,000–$500,000 per complex algorithm to prove it won’t kill you. - Harvard Gazette

Naive RAG fails on cosine-similarity, dimensionality, negation and temporal/context handling. SyncAlly's 15,000-word engineering report recommends hybrid BM25+vector search, late-interaction re-ranking (ColBERT/cross-encoders) and GraphRAG, noting higher storage and latency costs. - Hacker Noon

9 RAG architectures that reference external knowledge to improve language-model outputs, suit low‑stakes speed‑prioritized tasks. They can yield error reductions, and suggest measured hybrid rollouts with observability and instrumented scaling. - Hacker Noon

An open source AI agent called Nia indexes and enables searchable queries across the Epstein Files, including emails, messages, flight logs, court documents and other records, to support journalists, researchers and legal teams while raising privacy, legal and moderation concerns. - Trynia

AI sighed deeply and said, ‘redacted’

Zoubeir ElJlassi of Tunisia won the Global AI Film Award for Lily, announced at the 1 Billion Followers Summit, receiving a $1 million grand prize after a competition with 3,500 entries; the film was made using Google AI tools. - Google

Martina Radwan, director of Netflix's Stranger Things season 5 making-of One Last Adventure, disputed fan claims that ChatGPT was visible or used in the writers' room, saying no proof existed and she did not witness unethical generative-AI use. - HuffPost UK

BS at scale: Philosopher Harry Frankfurt's personhood/wanton distinction is used to argue that AI lacks second-order volitions and thus cannot act for normative reasons, generating 'bullshit' outputs that can give harmful advice, spread misinformation, and erode public deliberation. - 3 Quarks Daily

Follow the latest AI headlines via SkynetAndChill.com on Bluesky