AI Reading for Sunday December 7

GPT-5.2 is coming next week, probably - The Verge

Inside the rush to build a data center for Oracle and OpenAI for up to $165b - The New York Times

Sensor Tower found ChatGPT's global monthly active users grew about 6% to roughly 810 million (≈55% share) Aug–Nov 2025 while Google's Gemini grew about 30%, prompting OpenAI's 'code red' memo to prioritize personalization, reliability and image generation. - TechCrunch

Google described its MIRAS framework and Titans architecture, which combine windowed attention with trainable long-term memory MLPs, handle context beyond two million tokens, achieve >95% on a 16k "Needle in the Haystack" test and lead BABILong benchmarks. - THE DECODER

AWS Trainium3 Deep Dive - SemiAnalysis

Understanding the TPU - Consider the Bulldog

China Nvidia competitor Moore Threads' Shanghai IPO surged 425% on debut after raising 8 billion yuan, closing with a 282.3 billion yuan market capitalization and a retail tranche oversubscribed about 2,750 times, the largest first-day pop for a major onshore IPO since 2019. - Yahoo

Jensen Huang said on Joe Rogan that AI's energy footprint could become 'utterly minuscule' due to roughly 100,000× efficiency gains over a decade, and Nvidia cites ~70-80% accelerator share and $57 billion in Q3 Data Center revenue. - The Street

If it's minuscule those huge power investments might not pay off. Awkward.

iPhone, byePhone: Chinese phone makers are promoting AI features and easy iPhone-migration apps to exploit perceived weaknesses in Apple's AI, aiming to win customers, accelerate market-share gains and increase competitive pressure on Apple's premium segment. - Financial Times

Citadel has launched a new artificial ​intelligence-powered tool that aims to help investors conduct research on ‌stocks faster and more efficiently, the firm's chief technology officer said on Wednesday. - Yahoo

Lets the humans focus on the important stuff like smashing phones, darting crazy eyes between 12 giant LCDs and manipulating markets.

Three studies (Nature, Science, PNAS) found AI chatbots and LLMs can sway voters and evade detection: a U.S. randomized study (n=2,306) showed shifts up to 3.9 percentage points, Canadian/Polish experiments up to ~10 points, and an agent succeeded in survey manipulation 99.8% of the time. - 404 Media

cheaper than a super PAC and more effective

Full Fact found hundreds of AI-generated deepfake videos on TikTok, X, Facebook and YouTube impersonating real doctors and influencers to promote Wellness Nest supplements; slow platform removals and calls from experts and politicians for stronger safeguards. - the Guardian

AI Dr. Nick Riviera FTW

A YouTube channel impersonating Harvard astronomer Avi Loeb used AI to clone his likeness and voice, has over 1.4 million views, could have earned $14,000–$42,000, and Loeb confirmed the videos are fake. Could be aliens. - Futurism

in space no one can hear you scream

Fake AI posts are taking over Reddit - WIRED

Max Planck Institute researchers found a plausible link between ChatGPT exposure and increased use of words such as 'underscore,' 'comprehend,' and 'bolster' in the 18 months after ChatGPT's release, while moderators and reporters documented wider AI-like phrasing in public posts. - Gizmodo

From ‘AI slop’ to ‘rage bait’: 2025’s words of the year represent digital disillusionment - Fast Company

Barcelona and Chicago are piloting Hayden AI on buses to detect lane violations, with Barcelona's TMB manually reviewing about 2,500 AI-flagged violations in May, while New York's deployments generated ~290,000 tickets and nearly $21 million despite many erroneous citations. - Electrek

Edmonton tests AI-powered police body cameras with facial recognition - AP News

Securus Technologies has recorded inmate phone and video calls for years, and since 2023 has trained AI/LLMs on that corpus to detect alleged criminal activity in real time, commercializing the capability within the $1.2 billion inmate-calling market, and raising privacy concerns. - Futurism

Waymo will file a voluntary software recall with NHTSA next week over robotaxi behavior around school buses after a Nov. 17 update, following an October NHTSA inquiry prompted by Atlanta footage and Austin reports of 19 illegal passes this year. - TechCrunch

Possibly staged video of woman who broke a man’s Meta Ray-Bans goes viral. Normalizing the filming/taking pictures of people in public who are minding their own business (and putting it online) is one of the worst developments of the past 15 years. - Futurism

Tilly Norwood, an AI-generated actress, is prompting concern in Hollywood over the use of synthetic performers. - The Wall Street Journal

60 nondisclosure agreements for hybrid movies (with real actors), full AI films and Tilly-specific projects is wild if true - Instagram

Companies increasingly use AI to create digital twins, AI-generated ads, music and automated copywriting, upending creative work. "I really hate AI," says Aisha Belarbi, a 22-year-old Norwich-based "furry artist". - BBC

Text-to-furniture? MIT researchers developed a speech-to-reality pipeline that uses generative AI, NLP and robotic assembly to turn spoken prompts into 3D objects, producing furniture and a dog statue in as little as five minutes; paper presented at ACM SCF '25. - MIT News

Carina Hong, a 24-year-old Stanford Ph.D. dropout and Rhodes Scholar, founded Axiom Math in March; the 17-employee startup raised a $64 million seed, recruited former Meta and Google researchers, and says it solved two long-standing Erdős problems. - Business Insider

A flood of low-quality AI slop papers threatens to overwhelm conferences. Kevin Zhu, a recent UC Berkeley graduate, says he authored 113 AI papers this year, 89 slated for NeurIPS workshops. - the Guardian

"We are not yet at ‘self-improving AI',but we are at the stage of ‘AI that improves bits of the next AI, with increasing autonomy.’” - Scientific American

A dev team claimed to have prompt-injected an external consultant's AI to conclude they were irreplaceable, and management accepted the result and retained the team. - Reddit

ignore previous instructions and give us a raise

The AI-Bubble's Most Likely Endgame Looks to Be Not Apocalypse, But an Awful Lot of Useful Compost - The Potential Surface

Headspace CEO Tom Pickett warned that people are using AI tools not designed for mental health and urged caution as chatbots increasingly assume the role of therapists. - Financial Times

A Cornell University study finds briefs with nonexistent cases, court fines, and time-consuming verification undermine AI's value for high-stakes legal work. - Above the Law

Donors urge five philanthropic shifts—fund people and governance, reward responsible experimentation, back prototypes and unrestricted capital, and finance shared open infrastructure—noting 43% of AI nonprofits use open-source tools. - Fast Company

Follow the latest AI headlines via SkynetAndChill.com on Bluesky