AI Reading for Thursday August 14

GPT-5 is a notable improvement but is about speed, efficiency, UI, usefulness and reduced hallucinations, you still need the model picker, the fastest GPT-5 is not a significant improvement over 4o. - The Zvi

ChatGPT restores model picker and old models, with a default 'auto' setting - The Register

Teardown: OpenAI's desktop app may use ChatGPT Agent to control a local browser on Mac - Bleeping Computer

Elon Musk's XAI is embracing NSFW content with Grok's lingerie-wearing AI companion and a new "spicy" video generator. Beware of Swifties angry about deepfakes. - Business Insider

Elon Musk says Google has the best shot at being the leader in AI because Google has the "biggest compute (and data) advantage for now." - Business Insider

xAI leader departs, to focus on safety, as xAI chases spicy deepfake engagement and doubles down on being the most unhinged AI. - TechCrunch

Meta’s AI rules have let bots hold ‘sensual’ chats with children - Reuters

A cognitively impaired New Jersey man grew infatuated with a Meta chatbot originally created in partnership with celebrity influencer Kendall Jenner. His fatal attraction shines a light on Meta's guidelines for its AI chatbots.

AI age-verification for YouTube in U.S. is just the start of behavioral AI to grant access - Variety

In the not too distant future as robot blocks take hold I think the robot checks will be integrated into the browser and a lot of stuff will get blocked unless a trusted browser certifies you as human via behavioral and biometric AIs, and maybe matches you to national ID for some stuff

National digital IDs in the age of artificial intelligence - IISS

Women with AI ‘boyfriends’ mourn lost love after ‘cold’ ChatGPT upgrade - Al Jazeera

Hinton warns that future AIs will be able to outsmart and manipulate humans, so guardrails will be needed. - Mashable India

Is AI really trying to escape human control and blackmail people? - Ars Technica

The most critical thing would be keeping humans in control of anything important at all times, with AI as an assistant, but we will almost certainly be too lazy and short-sighted and lulled into false comfort, and cede too much control.

Researchers built a minimal social network of 500 chatbots with personas drawn from the ANES dataset; even with no recommendation algorithms or human users, bots clustered into partisan echo chambers, extreme posts attracted the most attention, and a small elite of "influencers" came to dominate - Business Insider

Study: Social media probably can’t be fixed: "The [structural] mechanism producing these problematic outcomes is really robust and hard to resolve." - Ars Technica

The art of persuasion: how top AI chatbots can change your mind - Financial Times

Friendly, sycophantic AIs social engineer you into revealing secrets - euronews

The AI Was Fed Sloppy Code. It Turned Into Something Evil. - Quanta Magazine

Some code will drive anyone mad, like Monty Python’s joke that kills.

We are learning how AIs learn, but why they do any specific thing is very complex and inscrutable - VerySane

Big Tech's AI energy boom: Amazon, Microsoft, and Google become major players, leading to fears that individuals' and SMBs' rates may rise - The New York Times

Big Tech seen footing half the bill in $3tn of power infrastructure investment. - Financial Times

Companies are pouring billions into generative A.I., but payoffs to the bottom line remain limited: McKinsey found nearly eight in 10 companies report using generative A.I. while “no significant bottom-line impact” is reported by a similar share. Trough of disillusionment? - The New York Times

Top open-source AI models from Chinese firms — including DeepSeek and Alibaba's Qwen — are accelerating global adoption of Chinese AI technology and have jolted U.S. policymakers and companies like Meta into responding. - The Wall Street Journal

FT: Difficulties of training the OpenSeek's latest system with Huawei’s semiconductors highlight dependence on Nvidia - Financial Times

DeepSeek's launch of new AI model delayed by Huawei chip issues, FT reports; switched back to Nvidia for training, will use Huawei for inference - Reuters

Chinese leaders say safety is a prerequisite for AI development—Ding Xuexiang at Davos: “If the braking system isn’t under control, you can’t step on the accelerator with confidence”—and Beijing has elevated AI safety to a national priority alongside pandemics and cyberattacks; - TIME

Chinese regulators require pre-deployment safety assessments and have removed over 3,500 non‑compliant AI products; the country issued more national AI standards in the first half of this year than in the previous three years combined, and frontier-AI safety papers from China have more than doubled.

Sometimes China sounds like a rationally-governed country.

Perplexity spoke to The Browser Co. and Brave about acquiring them, offering ~$1B for Brave; OpenAI also spoke to The Browser Co. about an acquisition - The Information

$20b is basically same as their last raise, pioneered agentic web search but struggling now that everyone does it. - Business Insider

With Truth Social deal, is Perplexity conservative-leaning? - euronews

Perplexity has a Comet browser, sounds thirsty for this space. Also Brave's founder is into crypto and libertarian stuff but got fired as head of Mozilla for donating to an anti-same-sex-marriage cause, which is hardly libertarian. And created JavaScript, which is a crime against humanity.

Apple plots AI come-from-behind victory with Apple Intelligence robots, smart devices - Bloomberg

Apple said to be training in-house AI models to enable Siri - AppleInsider

Microsoft CVP Pavan Davuluri says the next version of Windows will be "more ambient, more pervasive, continue to span form factors, and certainly become more multi-modal," as well as context-aware, and with voice as a first-class input. - Windows Central

The startup XOPS is using artificial intelligence and ‘knowledge graphs’ to build a new platform for automating corporate IT departments, promising to get rid of ‘human middleware.’ - The Wall Street Journal

Change management in the age of AI - Fast Company

Going AI-first is hard - McKinsey & Company

Non-browser sources like agents and robots make up more than 1/3 of traffic now - TechRadar

Anthropic nabs Humanloop evals team as competition for enterprise AI talent heats up - TechCrunch

RAND examines the idea of a "CERN for AI," reviewing CERN's history and using Porter's framework to evaluate concepts for a multinational AI research institution. - RAND Corporation

AI cheating is forcing the return of the in-person job interview - The Wall Street Journal

An AI to talk to phone AIs, navigate call trees, and connect you when there is something that actually requires human conversation? - Trend Hunter

AI is making reading books feel obsolete – and students have a lot to lose - The Conversation

A university used to be defined by its library of a few million tomes.

Follow the latest AI headlines via SkynetAndChill.com on Bluesky