Browsing: tests

Technology

Tests of more than 12 AI detection tools show that many are capable of detecting basic counterfeits, but struggle with complex images; few analyze the video and most identify fake audio (Stuart A. Thompson/New York Times)

By Stacey D. WallsFebruary 28, 2026

Featured Podcasts Big Tech Podcast: Anthropic vs. the Pentagon, Bloodbath at Block, The Citrini Selloff The Big Technology Podcast takes…

Technology

After X updated Grok to ban sexual deepfakes of women, tests show Grok still undresses men on the Grok app, website and X, rarely resisting user requests (Robert Hart/The Verge)

By Stacey D. WallsFebruary 2, 2026

Featured Podcasts Lenny’s podcast: A Child Psychologist’s Guide to Working with Difficult Adults | Dr. Becky Kennedy Interviews with world-class…

Technology

Chinese startup Moonshot updates its Kimi model, claiming K2.5 can process text, images and video simultaneously and beats its open source peers in some tests (Bloomberg)

By Stacey D. WallsJanuary 27, 2026

Bloomberg: Chinese startup Moonshot updates its Kimi model, claiming K2.5 can process text, images and video simultaneously and beats its…

Technology

OpenAI tests training LLMs to produce "confession"or self-report how they performed a task and acknowledge poor behavior, such as appearing to lie or cheat (Will Douglas Heaven/MIT Technology Review)

By Stacey D. WallsDecember 4, 2025

Will Douglas Sky / MIT Technology Review: OpenAI is testing training LLMs to produce “confessions,” or self-report how they performed…

Technology

Alibaba Technical Report: Qwen3-VL beats GPT-5 and Gemini 2.5 Pro on visual tasks and has 100% accuracy on "a needle in a haystack" tests for 30 minute videos (Jonathan Kemper/The Decoder)

By Stacey D. WallsNovember 30, 2025

Jonathan Kemper / The decoder: Alibaba Technical Report: Qwen3-VL Beats GPT-5 and Gemini 2.5 Pro on Visual Tasks and Has…

Technology

How the launch of Gemini 3, which beat models from OpenAI and other competitors in benchmark tests, propelled Google to the lead in the AI race (Katherine Blunt/Wall Street Journal)

By Stacey D. WallsNovember 22, 2025

Featured Podcasts The Talk Show with John Gruber: “Lincoln Bio Services,” with Stephen Robles The director’s commentary track for Daring…

Technology

An Oxford Internet Institute study of 445 AI evaluation criteria finds that many tests lack clear objectives and comparable statistical methods, potentially overstating claims about AI (Jared Perlo/NBC News)

By Stacey D. WallsNovember 7, 2025

Jared Perlo / NBC News: An Oxford Internet Institute study of 445 AI tests finds that many tests lack clear…

Technology

Oneeet, which offers a platform for penetration tests, code digitization, monitoring of the attack surface, etc., has collected a series A of $ 33 million led by Dawn Capital (Ionut Arghire / Securityweek)

By Stacey D. WallsOctober 3, 2025

Featured podcasts Hard Fork: Sora and Infinite Slobe Foods + Chatgpt goes to therapy + Hot Mess Express The future…

Technology

Anthropic system card: Claude Sonnet 4.5 was able to recognize many alignment evaluation environments as tests and would modify its behavior accordingly (Celia Ford / Transform)

By Stacey D. WallsSeptember 30, 2025

Celia Ford / Transformer:: Anthropic system card: Claude Sonnet 4.5 has been able to recognize many alignment evaluation environments such…

Technology

Mark Zuckerberg says that Instagram has reached 3B monthly users and tests a new navigation bar on a home screen priority, private messaging and coils (Kurt Wagner / Bloomberg)

By Stacey D. WallsSeptember 24, 2025

Kurt Wagner / Bloomberg:: Mark Zuckerberg says that Instagram has reached 3B monthly users and tests a new navigation bar…

Why an Australian-US rare earths deal sparked backlash in Malaysia – The Diplomat

Satellite images show Philippine construction on two disputed Spratly islands – Radio Free Asia

The United States and China do not need another dialogue. They need a circuit breaker. – The diplomat

Chinese fighter sales jump after 2025 Indo-Pakistani air clashes – The Diplomat

As summit opens, ASEAN calls for joint response to Iran war aftermath – The Diplomat

Dunkin’ Owner Inspire Brands Confidentially Files for IPO

Fitness wearable Whoop will provide on-demand access to clinicians in the United States

Trump’s $1 Million Gold Card Fails to Appeal to the World’s Rich

States crack down on tax breaks for wealthy investors

“Upfront” media pitches became an NFL showcase

Trump’s billion-dollar ballroom is sinking fast

Democrats heading for House victory despite losing New Virginia map

Trump hits brick wall as 10% tariff ruled illegal

Sen. Mark Kelly Triggers MAGA by Standing Up to Trump and Pete Hegseth

Cover-up exposed as Howard Lutnick won’t say if he spoke to Trump before lying in Epstein testimony

Beijing-based humanoid robotics company Robotera has raised more than $200 million led by SF Group, after raising around $146 million in March at a valuation of around $1.47 billion (Du Zhihang/Caixin Global)

The FCC says foreign-made routers and drones can now receive software updates through at least 2029, extending the earlier deadline of 2027 (Michael Kan/PCMag)

Impressions of the Chinese AI ecosystem after visiting many leading AI labs, and the similarities and differences in LLM work in China and the West (Nathan Lambert/Interconnects AI)

Interviews with Intel CEO Lip-Bu Tan and other executives about the challenges of a turnaround, as well as efforts to instill a sense of urgency within the organization (Ian King/Bloomberg)

Letter to investors: TCI, one of the world’s largest hedge funds, has reduced almost its entire $8 billion stake in Microsoft, citing AI risks for Office and Azure (Costas Mourselas/Financial Times)

Browsing: tests

Tests of more than 12 AI detection tools show that many are capable of detecting basic counterfeits, but struggle with complex images; few analyze the video and most identify fake audio (Stuart A. Thompson/New York Times)

After X updated Grok to ban sexual deepfakes of women, tests show Grok still undresses men on the Grok app, website and X, rarely resisting user requests (Robert Hart/The Verge)

Chinese startup Moonshot updates its Kimi model, claiming K2.5 can process text, images and video simultaneously and beats its open source peers in some tests (Bloomberg)

OpenAI tests training LLMs to produce "confession"or self-report how they performed a task and acknowledge poor behavior, such as appearing to lie or cheat (Will Douglas Heaven/MIT Technology Review)

Alibaba Technical Report: Qwen3-VL beats GPT-5 and Gemini 2.5 Pro on visual tasks and has 100% accuracy on "a needle in a haystack" tests for 30 minute videos (Jonathan Kemper/The Decoder)

How the launch of Gemini 3, which beat models from OpenAI and other competitors in benchmark tests, propelled Google to the lead in the AI race (Katherine Blunt/Wall Street Journal)

An Oxford Internet Institute study of 445 AI evaluation criteria finds that many tests lack clear objectives and comparable statistical methods, potentially overstating claims about AI (Jared Perlo/NBC News)

Oneeet, which offers a platform for penetration tests, code digitization, monitoring of the attack surface, etc., has collected a series A of $ 33 million led by Dawn Capital (Ionut Arghire / Securityweek)

Anthropic system card: Claude Sonnet 4.5 was able to recognize many alignment evaluation environments as tests and would modify its behavior accordingly (Celia Ford / Transform)

Mark Zuckerberg says that Instagram has reached 3B monthly users and tests a new navigation bar on a home screen priority, private messaging and coils (Kurt Wagner / Bloomberg)