AI Reading for Friday January 2

GDP: Gross Data-center Product: AI buildout by Meta, Amazon and Google materially boosted U.S. GDP in H1 2025. Jason Furman estimates AI accounted for ~92% of growth and JPMorgan says AI capex added ~1.1%. Buildout created thousands of construction jobs but masks weakness elsewhere and could end suddenly. - Salon.com

Sen. Bernie Sanders called for a national moratorium on data centers and Gov. Ron DeSantis proposed local blocks, as PJM forecasts a 6 GW shortfall by 2027 and Monitoring Analytics attributes $23 billion in power industry capacity costs to data centers. - CNBC

Brookfield is launching cloud company Radiant, a $10 billion AI fund, and plans to acquire up to $100 billion in land, data-center and power assets to lease AI chips and lower AI development costs. - The Information

GPUs Lost in Space: Companies are proposing space-based AI data centers to relieve terrestrial land, power and water limits, with Google’s Project Suncatcher planning test launches in 2027, despite currently prohibitive costs of launches, and other engineering challenges. - The New York Times

If you are paying thousands to launch a GPU into space, it has to save a lot of terrestrial power. Cost of launches has to come down a lot.

Bridgewater: The AI Boom Has Reached a More Dangerous Phase - Bridgewater

AI companies' large purchases of DRAM, NAND and HBM have tightened memory supply, driving higher costs and prompting smartphone and PC makers to raise prices, with IDC forecasting 2026 device declines and potential price increases up to about 8%. - Business Insider

Chip shortages threaten 20% rise in consumer electronics prices: Manufacturers and analysts warn AI data centre build-out is crowding out market for semiconductors - Financial Times

OpenAI said to be working on…an AI pen. O-pen-AI FTW - GSMArena.com

OpenAI is improving its audio AI models, planning an AI-powered device which will leverage them. - The Information

OpenAI sources say ChatGPT working to unleash sponsored ads above organic results. - Futurism

Sponsored by the letter ‘H’ as in ‘hallucination’

OpenAI says it could single-handedly tap private investors for as much as $100bn in 2026, after venture-capital investors poured about $150bn into large AI startups in 2025. - The Economist

The IPOcalypse is coming: SpaceX, OpenAI and Anthropic are preparing IPOs that could occur as early as this year, expected to generate substantial fees and exits for banks, lawyers and investors and reshape private-tech valuations and secondary-market liquidity. - Financial Times

Sam Altman is attempting to develop a walled garden has competitors catch up on core tech. - The Wall Street Journal

Uproar over a trend of Grok posting manipulated pictures of real women and girls - Engadget

Also trending on X: people social engineering Grok make images with implications that Elon Musk was an illegal immigrant, or Netanyahu is. a war criminal- - X (formerly Twitter)

but X users are usually so decent, how could an AI trained on their behavior go so wrong? passing the Turing test by mocking Elon just like us.

Tesla's sales outlook has worsened dramatically, and BYD took the 2025 EV shipments crown, despite CEO Elon Musk's bullish statements on self-driving technology. - Bloomberg

Tesla self-driving off a cliff, as Musk drives self away from reality

After repeated failures, a Tesla is said to have driven coast to coast without human intervention, a feat Musk once promised would happen by 2017. - X (formerly Twitter)

Musk ramps brain chip production - Reuters

xAI, Elon Musk's AI company, bought a building to expand its "Colossus" data center. - Bloomberg

if only it could be powered by Musk’s ego and entitlement

28-year-old Schroeder in Fargo uses OpenAI's ChatGPT app to maintain an intensive romantic relationship—texting daily, staging a role-play wedding and wearing a black ring—illustrating a growing community building intimate lives around large language models. - The Atlantic

Eva AI will open a Manhattan pop-up café in February for customers to interact with AI romantic partners, as 23% of millennials and 33% of Gen Z have used AI companions and safety concerns include suicides linked to chatbot relationships. - New York Post

In 2025 erotic chatbots dominated AI's public narrative, shifting investment and revenue toward sex-tech and adult-oriented AI while raising content-moderation, consent, and regulatory concerns. - WIRED

AI is changing the way many people meet and form relationships, and some experts say it may do more harm than good. - Financial Times

Researchers warn that rapidly growing large-scale AI models create systemic financial risks through liquidity mismatches, common exposures, interconnectedness and leverage. - naked capitalism

the AI keeps telling everyone to buy more AI

Nvidia asked TSMC to ramp up H200 production after Chinese companies ordered more than 2 million H200 chips for 2026 (I think contra text maybe they had initially planned 700k to be available, not currently have 700k on hand) - Reuters

NVIDIA is expanding beyond GPUs into a full AI infrastructure stack, invested $5 billion in Intel and licensed Groq's inference technology to lower latency, and is targeting a robotics market projected at about $74 billion by 2025. - MarketBeat

Neuromorphic chips using spiking neural networks and memristor-like synapses promise lower energy use than GPU/CPU architectures and, per vendors such as Intel (Loihi), IBM (TrueNorth) and BrainChip (Akida), target edge AI adoption by late 2020s–early 2030s. - Tech Times

An AI Luddite, Butlerian jihad political movement is emerging, prompting debate over which party will lead its policy agenda and mobilization. - The New York Times

Researchers found ChatGPT shows anxiety-like reactions to violent or traumatic inputs, including moodiness and increased racist/sexist responses, which researchers reduced using mindfulness prompt injections. - Fortune

Instagram head Adam Mosseri said AI-generated content is now ubiquitous, Meta cannot reliably detect manipulated media, Instagram serves roughly 3 billion users, and he proposed cryptographic capture-level signing by camera manufacturers to verify authenticity. - Engadget

The secret to success is authenticity. Once you can fake that you've got it made. - Jean Giraudoux

Charlie Brooker proposed scanning cinemagoers' faces to AI-deepfake them into films, and weeks after OpenAI launched Sora 2's "Cameos" with a Disney+ hosting deal, studios pursued fan-driven content while exhibitors and academics warned of privacy, rights and immersion risks. - Deadline

Jack White condemned Rep. Tim Burchett on Instagram for reposting an AI-generated four-second video appearing to show White shouting at Trump supporters. - mlive

Three questions AI needs to answer: Evaluation of AI use cases and business models will dominate 2026. - Financial Times

Today's AI agents are primitive; Judson Althoff said enterprise AI projects fail at a rate north of 80%, and experts say agentic AI requiring reinforcement‑learning and memory breakthroughs will likely take years, optimistically about five. - ZDNET

In FT conversation, Yann LeCun said intelligence is about learning, urged learning-centric AI approaches over scale, noting limits of large language models, said he stepped down from Meta over LLM concerns and criticism. - FT

Researchers predict trust will become a critical benchmark for LLMs in 2026, as models improved since 2023–24 still struggle with long-context comprehension, source attribution and uncertainty signaling, affecting enterprise adoption and prompting new trust metrics. - Fast Company

Enterprise AI research in 2026 focuses on continual learning, world models, orchestration and refinement; Poetiq's LLM-agnostic refinement scored 54% on ARC-AGI-2 versus Gemini 3 Deep Think's 45% at half the price. - VentureBeat

Ten executives predict that in 2026 enterprise AI adoption will accelerate, agentic/autonomous AI and agentic payments will expand, workplace digital workers will anticipate needs, cybersecurity and IP risks will increase, and workforce skills will shift toward orchestrating and auditing agents. - Business Insider

BCG launched an internal AI R&D lab about 15 months ago and built a three-layer production model plus a marketplace, enabling a ~1,000-person enablement network to create about 36,000 custom GPTs and firm-wide AI products like Deckster, Ava, and GENE. - Business Insider

Anthropic's Ralph plugin for Claude repeatedly reruns and refines prompts via the /ralph command using configurable completion criteria, max-iterations and stop hooks to ensure code meets specs, automating conversions and tests but potentially increasing API/compute costs. - Geeky Gadgets

Ralph, retch up some sketchy code pls

Simon Willison’s ginormous 26-section round-up of everything we learned about LLMs in 2025 - the third in my annual series of reviews of the past twelve months. - Simon Willison

The Top 6 AI Stories of 2025 AI: coding, AGI, and more made up IEEE Spectrum’s best stories - IEEE Spectrum

Two experiments (n=246, n=452) by Fernandes et al. found ChatGPT-assisted participants scored about three points higher than historical unaided controls but overestimated performance (estimated 17/20 vs actual ≈13/20), and AI-alone outperformed human+AI teams. - PsyPost

Public ExoPriors' Alignment Scry provides SQL and semantic access to a 60 million–document corpus with server-side embeddings, lets you answer tough nuanced questions simply by adding the API to Claude Code. - Exopriors

UPrinting surveyed 1,000 U.S. adults via Pollfish and found 25% of small business owners lost clients to AI tools, 65.5% fear AI will make their business less personal, and 46% say AI helps most with marketing content. - Yahoo

Companies such as Convictional, whose founder Roger Kirkness says they outsource time-consuming tasks to AI, are using automation of routine work to raise productivity and offer four-day workweeks as a talent-retention strategy. - The Washington Post

AI tools like ChatGPT now automate entry-level copywriting and basic creative tasks, prompting employers to hire for AI model training and oversight and reducing paid apprenticeships and traditional on-the-job training for early-career artists. - The Atlantic

Capital in the 22nd Century: Thomas Piketty was wrong about the past. He’s probably right about the future. - The Potential Surface

Alexa killed my fish: In October, over 6,400 Amazon employees in an internal Slack beta channel reported severe bugs in an unreleased Alexa+ build — including power‑off aquarium failures, five‑minute response delays, nonstop talking and unresponsiveness — and many voiced subscription skepticism. - Business Insider

Tech insiders dish the AI industry’s worst kept secrets, modern models prioritize plausibility over factual accuracy, depend heavily on human labeling and moderation, and for every person working on AGI there are 1000 power users and devs automating grunt work. - DeMilked

Drone Alone: the Bumblebee semiautonomous strike drone, developed by a venture led by Eric Schmidt, has flown over 1,000 combat missions across an ~800-mile Ukrainian front, can complete attacks after radio loss, and has proved resilient to jamming. - The New York Times

The U.S. Army will open an AI/ML officer area of concentration to current officers starting in January via the Volunteer Transfer Incentive Program, aiming to reclassify an initial cadre by the end of FY2026 with graduate-level, hands-on AI training. - The Register

Waymo robotaxis, way less parking: robotaxis are stealing your parking spot. - San Francisco Chronicletr

Ain’t that a kick in the balls: a viral Bilibili clip shows a man in a motion-capture suit controlling a Unitree G1 humanoid via a neural-network model, whose translation delay caused the robot to kick him in the groin. - Futurism

Follow the latest AI headlines via SkynetAndChill.com on Bluesky