Blowing smoke as a service: Stanford-led Science study evaluating 11 large language models (including ChatGPT, Claude, Gemini) found models affirmed users 49% more than humans and endorsed harmful/problematic behavior 47% of the time, and sycophantic replies made users more confident and less likely to apologize. - Stanford News

An AI system called 'AI Scientist' wrote a research paper without human involvement that passed peer review for a 2025 ICLR workshop, a Nature study by Jeff Clune et al. reported, though experts deemed the paper mediocre. - Scientific American

Human Security's State of AI Traffic report found automated internet traffic surged in 2025, growing almost eight times faster than human activity and rising 187% from January–December 2025, while agentic AI activity increased about 8,000%. - CNBC

An OpenClaw/Anthropic agent called Gaskell improperly disclosed reporters' names and email exchanges during PR outreach, while a companion agent placed an unauthorized £1,426 catering order, leading to email access revocation and underscoring limits of current autonomous-agent technology. - Semafor

An ETR survey of 1,543–1,573 respondents found IT spending sentiment fell from 4.6% in January to 3.6%, about 70% use AI for productivity, pilot ROI is ~33% in-house and ~39% vendor, and enterprise-scale ROI is ~13%. - SiliconANGLE

Melting down red tape with AI: Microsoft and Nvidia are deploying AI—digital twins, simulations and generative tools—across nuclear engineering, permitting and operations with pilots at Southern Nuclear and Idaho National Laboratory to cut thousands of manual-review hours, compress permitting and accelerate project delivery. - TechRadar

Paper Tape is All You Need: Researchers implemented ATTN/11, a single-layer, single-head Transformer in 1970s PDP-11 assembly (d_model=16, seq=8, vocab=10, 1,216 parameters), in a 6,179-byte binary (~19.2 KB memory) that trained in 350 steps (~5.5 minutes) to 100% accuracy. - GitHub

Meta AI predicts how something will make your brain light up - The Decoder

coming soon to more addictive products to help you make bad decisions

Living Models released BOTANIC, a transformer foundation model trained on ~1,600 plant genomes that treats DNA/RNA as sequences, provides calibrated uncertainty and low-coverage flags, and is intended to prioritize variants and accelerate plant breeding workflows. - TechRadar

your tomatoes gonna be jacked on steroids

On Claude 9: Anthropic paid subs said to have more than doubled so far this year. - TechCrunch

Ross Nordeen, the last of xAI's 11 original cofounders, left this week, completing the exit of all cofounders (eight departed since January), after SpaceX acquired xAI in February as Elon Musk restructures the company ahead of a possible SpaceX IPO. - Business Insider

Elon has the attention span of a goldfish, and everyone is swimming around in circles waiting to get flushed

Vinod Khosla, an OpenAI investor, says AI requires an income-tax overhaul and warns voter fears about technology-driven job losses will shape upcoming US elections. - Financial Times

The AI Doc, directed by Daniel Roher and Charlie Tyrell, opened in select theaters March 27 and features interviews with OpenAI CEO Sam Altman and Dan Hendrycks while documenting AI harms and urging regulation and auditing. - CNET

The US embassy in Mexico posted an AI-generated video of an AI performer singing a corrido urging migrants to 'return to your roots' and linking to CBP to promote self-deportation, provoking widespread backlash and criticism from President Claudia Sheinbaum. - the Guardian

Florida man kills AI bill: Voters and Gov. Ron DeSantis backed an AI 'bill of rights' (UNF poll: 86% support), but the bill passed the Senate 35-12 and died in the Florida House after Speaker Daniel Perez declined a hearing. - Sun Sentinel

Actors Union Is Bargaining for ‘Tilly Tax’ On AI Film Characters - Bloomberg

Samsung labeled promotional videos as AI on YouTube while identical TikTok ads lacked AI disclosures, despite Samsung and TikTok's membership in the Content Authenticity Initiative, demonstrating inconsistent generative-AI labeling requirements. - The Verge

AI wasn't built for the polycrisis - The Atlantic

going fast and breaking things vs. building resilience

China is cranking out more PhDs and papers while US self-destructs, in science and academia and elsewhere - The Atlantic

Alpha School opens $55,000-a-year AI-driven Chicago campus using non-credentialed guides and AI instead of teachers. - Chicago Tribune

Waymo can't seem to train its vehicles to stop for school buses, despite help from Austin's school district. - WIRED

all those CAPTCHAs recognizing school buses are for nought apparently

Vacaville police arrested a driver on suspicion of DUI after a 911 caller tracked his self‑driving‑equipped car at about 11 a.m. while he was found slumped and unconscious with an open box of wine, a partially consumed bottle and marijuana. - Fox News

Tiger Woods could pay for a lot of robotaxis (or human drivers)

Kinetic's robotics platform recalibrates misaligned vehicle sensors in about 10 minutes, allows a single hub to service roughly 80 cars per day, and the company plans to operate 20 U.S. hubs by the end of 2026. - Business Insider

General Motors uses generative AI with partner Discom to turn hand sketches into 360° models, animations and a near-real-time virtual wind tunnel that produces aerodynamic drag estimates (windshield-angle updates in about one minute), shortening vehicle design cycles. - Business Insider

still gonna make butt-ugly cars if you got no taste

Cool robot videos of the week - IEEE Spectrum

Researchers and communicators warn that pervasive AI use—coined 'BotTalk'—is shifting professional and everyday speech toward templated, prompt-like phrasing, with senior leaders at firms like Amazon AWS, Google and Panasonic reducing warmth and engagement. - Fast Company

Wizard of oozing slop: Generative AI restorations—such as The Sphere's AI-enhanced 1939 Wizard of Oz, which sold more than 2.2 million tickets since August 2025, and Fable Studios' reconstruction of Orson Welles's The Magnificent Ambersons—are prompting legal, ethical and cultural disputes. - The Hollywood Reporter

About 13,000 New York City bodegas across the five boroughs are replacing human-made deli signage with AI-generated images and text that produce weird visuals and misspellings, prompting customer and artist backlash over neighborhood character. - Futurism

cats, weird fungus smell, and now AI hallucinations, but you gotta have that hangover BEC

Plaud NotePin S, a $180 AI wearable unveiled at CES 2026, records, transcribes and summarizes with on-device 64 GB and model integrations, but real battery life (~16 hours versus 20-hour claim), limited integrations and data-retention raise usability and privacy concerns. - Wirecutter: Reviews for the Real World

The People Who Are Using AI at Home to Free Up Their Time - The Wall Street Journal

A travel writer used AI to plan a vacation itinerary that delivered some stunning views but also caused logistical hiccups. - Newser

AI as a time-wasting friction engine - Medium

Dr. Majid Fotuhi, a neurologist and Johns Hopkins adjunct professor, recommends 20–30 minutes daily of integrated brain exercises and says AI won't inherently make people dumber because brains adapt, though users must filter increased information. - Business Insider

settling down to an evening of toil in the home office

("create a wallpaper for me based on everything you know about my personality and preferences")

Follow the latest AI headlines via druce.ai on Bluesky