AI Reading for Saturday November 15

Meta Chief AI Scientist Yann LeCun, reportedly leaving after being sidelined, says LLMs are a dead end for AGI and backs multi-modal world models instead - The Wall Street Journal

Meta plans up to $72 billion in AI infrastructure spending this year, CMO Alex Schultz said the AI-driven pivot to 'unconnected' content and ad tools has generated billions in revenue and is 'aggressive, but not crazy.' - Business Insider

Yann did not share Zuck’s LLM enthusiasm.

Meta will make 'AI-driven impact' a core employee performance expectation from 2026, reward exceptional AI-driven impact in 2025, and roll out an 'AI Performance Assistant' for reviews beginning Dec. 8, allowing Metamate and Google's Gemini for review content. - Business Insider

Chat-PIP?

Oracle of Omaha orders the Alphabet soup with his Cherry Coke? Berkshire Hathaway discloses a $4.3B stake in Alphabet in Q3, a surprising move given Warren Buffett's reluctance toward high-growth, tech-driven stocks. - CNBC

5 tool player, Google has all the pieces, foundation model, strongest enterprise cloud AI offering, access to proprietary big data, now a deal with Apple.

CEO Sundar Pichai said Gemini 3 will launch by the end of the year, promising coding and multimedia upgrades, and the Gemini app has about 650 million monthly users versus ChatGPT's roughly 800 million weekly users. - Business Insider

Google-Backed Robot Maker Apptronik in Talks to Raise Funding at $5 Billion Valuation - The Information

On Oct. 27, a Waymo driverless taxi in San Francisco struck and killed a Mission District cat, which Waymo acknowledged occurred when the animal "darted under our vehicle as it was pulling away," prompting local outrage and regulatory scrutiny. - The New York Times

A multimodal model in Google AI Studio achieved near-expert handwritten-text recognition—strict CER 1.7% and WER 6.5% (0.56% CER, 1.22% WER excluding punctuation) on a 50-document benchmark and showed implicit symbolic reasoning by converting '145' to '14 lb 5 oz'. - The Potential Surface

is 145→ 14 lb 5 oz actually correct in context tho? does it pass the doctor chicken scratch test?

Models commoditize, platforms monetize? Microsoft CEO Satya Nadella warned model creators face a "winner's curse" because foundational models are "one copy away from being commoditized," saying lasting value will lie in platform scaffolding, integration and distribution rather than a single dominant AGI model. - Windows Central

Perplexing business model…300 founders and investors voted Perplexity most likely to fail, despite reported valuations between $14 billion and $50 billion. - Business Insider

pioneered AI search but everyone has caught up, they pivoted to agents, it’s hard for web sites to block Google but a no-brainer to block Perplexity.

JPMorgan projects an AI-driven boom will generate a record $1.8 trillion in bond sales in 2026. - Bloomberg

Retirees connect to server farms: Insurance companies funding retirement benefits with data-center debt. - The Wall Street Journal

hype locomotive pulls a train of debt

Backdoor Blackwells: Shanghai-based INF Tech gained legal access to 2,304 Nvidia GB200 'Blackwell' GPUs in October 2025 by contracting compute from Indonesian telco Indosat Ooredoo Hutchison, which deployed 32 Aivres-built server racks in Jakarta, raising export-control loophole concerns. - Tom's Hardware

Data centers are being upgraded to support AI workloads and increased infrastructure and compute demands. - The Wall Street Journal

Achievement unlocked: No em-dashes! OpenAI's GPT-5.1 adds stronger instruction-following and per-user personalization controls that let users suppress em dashes, CEO Sam Altman said on Twitter, though the change applies per account and does not fix the model globally. - Gizmodo

Understanding neural networks through sparse circuits: We trained models to think in simpler, more traceable steps—so we can better understand how they work. - OpenAI Help Center

Eli Lilly CEO David Ricks said he runs "at least one or two AIs" during every meeting and prefers xAI's Grok and Anthropic's Claude over OpenAI's ChatGPT for science questions, while warning about hallucinations and urging robotic experiments. - Business Insider

Grok-blocked: Elon Musk now says xAI is delaying its Grok 5 launch until next year - MarketWatch

Ternus around with our AI pls? Apple intensifies succession planning for CEO Tim Cook, Ternus seen most likely successor - Financial Times

Old Age Will Be Different in the Robotic Age, as robots help elders - The Wall Street Journal

Watch a Unitree robot doing some house chores - Reddit

now help me find my glasses and grandma will knit you a sweater

Bill Gates’s daughter adopts abusive business practices, say it ain’t so? Security researchers found Phia's Chrome extension copied full HTML of every webpage users opened, including bank statements and private emails, compressed and sent snapshots to Phia's servers; Phia removed the feature after alerts but did not notify users. - Fortune

Tracker… never heard of her! At least she’s not in the Epstein files, or dipping a pen in the company ink.

The American Psychological Association issued an advisory warning for use of general-purpose generative AI chatbots and wellness apps for unmet mental-health needs, outlining harms and mandating developer safeguards, clinician disclosure, policy reforms, and rigorous independent research and audits. - American Psychological Association

One day my smartwatch will know me better than I know myself. How much can AI fitness tech help us avoid injury? - Lifehacker

Nature published a video series featuring six AI leaders—Mustafa Suleyman, Pushmeet Kohli, Timnit Gebru, Jared Kaplan, Anima Anandkumar and Amandeep Gill—discussing AI adoption, risks, trillions in investment, and UN-led governance. - Nature

Robot robbers…AI is reinventing crime and cops aren't ready - Axios

Isolde Hegemann's Oct 2025 USApp/LSE study finds AI-powered fact-checking reduces Republicans' engagement with online political misinformation more than traditional fact-checking methods. - Impact of Social Sciences Blog

Alibaba Group cut Qwen3-Max API prices by about 50%, reflecting release of new Ernie, Kimi, MiniMax models, and intensified competition in China’s foundational-model market. - South China Morning Post

Databricks unveiled ai_parse_document, integrated into Agent Bricks, which parses PDFs into Delta tables preserving complex layouts, claims 3–5× lower cost, and Erich Elsen said 80% of enterprise knowledge remains in PDFs; early deployments include Rockwell, TE Connectivity and Emerson. - VentureBeat

AI Rules for thee, not for me: Nitro-commissioned Oct 2025 survey of 1,003 US/UK/Canada workers found 68% of executives admitted violating their own AI policies in the past three months, about 50% of employees used unapproved AI tools, and 33% processed confidential data with AI. - The Register

Upwork study says agents do better with human in the loop - VentureBeat

Small-Business Owners vibe-code and asking AI to automate stuff a big company might have a big team and ERP for. - The Wall Street Journal

Business Insider interviewed 16 people across India, Singapore, Silicon Valley and Sydney identifying four routes into AI—graduating, transitioning, pivoting, founding startups—and reported strong cross‑industry demand driving high pay, including a $400,000 Meta offer and a 30% base-pay increase. - Business Insider

OmniFocus is adding on-device, offline generative AI features that preserve privacy and enable task summarization, inbox processing, natural-language task creation, and customizable behaviors, according to The Omni Group. - WIRED

Two months of prompting v two years of training: a systematic approach to classifying car problems lets prompts beat a highly trained XGBoost model at lower cost, and without needing a lot of labeled training data. - Levs

Coca-Cola ran an AI-generated Christmas ad for a second consecutive year that, despite continuity glitches, scored maximum short- and long-term commercial effectiveness in scale testing, while John Lewis's human-crafted spot drew similar consumer reactions. - The Curious Brain

Can AI unlock a new generation of human creativity? - Fast Company

AI-powered design accelerates the development of mRNA therapeutics - Nature

Linear algebra at the speed of light: researchers demonstrate efficient fully parallel optical matrix–matrix multiplication (POMMM) via a single coherent light propagation, matching GPU results with mean absolute error <0.15 and NRMSE <0.1, IIUC would drop TFLOPS/W a couple of orders of magnitude. - Nature

I know what you are thinking: Japanese scientist turns MRIs into text using ‘mind-captioning’ technology. - CNN

How do ‘AI detection’ tools actually work? And are they effective? - The Conversation

Andrej Karpathy wrote on X that autonomous vehicles will "terraform" outdoor urban spaces, reducing parked cars and lots, increasing public space and safety and reducing noise pollution, though he said the transformation will not happen overnight. - X (formerly Twitter)

More traffic though. Maybe jitneys in super high density NYC. I could see cities using a ton of micromobility, less using streets for car storage and garages and parking lots everywhere, needs some redesign.

U up? The app Text With Jesus uses artificial intelligence and chatbots to offer spiritual guidance to users who are looking to connect with a higher power - NBC10 Philadelphia

WWJD? just ask. Premium users can talk to Satan (not even kidding).

A tech-world faction called "cheerful apocalyptics" advocates allowing AI to surpass and replace humans, citing figures like Larry Page and thinkers such as Hans Moravec, and their influence could shift investment, corporate roadmaps and AI-safety regulation. - Cranach

Follow the latest AI headlines via SkynetAndChill.com on Bluesky