AI Reading for Thursday August 21

Suddenly, Silicon Valley Is Lowering AI Expectations. Investor sentiment cools as Sam Altman is talking bubbles now, and Eric Schmidt is over the term “AGI.” - Intelligencer

Is the A.I. Sell-Off the Start of Something Bigger? Tech stocks slid as investors reassessed AI valuations; officials also floated taking equity stakes in CHIPS Act beneficiaries, raising policy and market questions. - The New York Times

US tech-stock stumble shows vulnerability in AI trade. - Reuters

What AI sell-off though, still right around highs, just stopped going up every single day. A lot of hand-wringing over a nonexistent selloff.

Google brass pressures devs to dogfood Google AI and coding tools. "You know a technology works and is great when you're forced to praise it to maintain your livelihood." - Business Insider

Google AI Mode search gets agentic, with computer use helping search for and make restaurant reservations. - Google

Google announces new AI-powered personal health and fitness coach for Fitbit - TechCrunch

Fitbit’s AI health coach is the first I might actually be interested in - The Verge

Want to take better photos? Google thinks AI is the answer. - The Washington Post

20+ on-device features for Pixel 10’s TSMC-built Tensor G5 with ~34% faster CPU performance and up to 60% stronger TPU, running a newer Gemini Nano with a 32K context window - 9to5Google

Pixel 10 demonstrates near-real-time on-device translation that reproduces the speaker’s voice in the target language. - Gizmodo

Anthropic bundles Claude Code into enterprise plans - TechCrunch

This is very good, 1h to get the full tour of Claude Code, it is giving me an existential crisis - www.deeplearning.ai

You combine Claude and Gemini with Google’s ADK and agent hosting and overall platform, it’s strong. - YouTube

It’s hard to get an overall solution where you can write agents, deploy them, monitor and eval and secure them with ability to connect to all your enterprise resources, let rank and file make and deploy workflows.

Databricks achieves hectocorn status - www.google.com

China blocks Nvidia AI chip after ‘insulting’ Howard Lutnick remarks. “We don’t sell them our best stuff, not our second-best stuff, not even our third-best…You want to sell the Chinese enough that their developers get addicted to the American technology stack, that’s the thinking.” - Financial Times

Vantage Data Centers is building a $25B, 1.4-gigawatt AI campus in Texas with ultra-high-density racks, backed by Silver Lake and DigitalBridge, to meet hyperscaler compute demand. - CNA

Bipartisan U.S. senators pressed Meta about its AI chatbots, escalating policy scrutiny. - Semafor

Yann LeCun demoted in massive Meta AI reorg resulting in Yann reporting directly to Alexander Wang - Business Insider

Zuck wants Alexandr Wang to be his Sheryl Sandberg but for AI instead of monetization, but it’s causing friction with the old guard.

Controversial Character.AI discussed a potential sale and new funding options a year after its founders joined Google. - The Information

Why Microsoft's AI boss is worried about 'Seemingly Conscious AI' - Business Insider

Just because your program says print(“I am sentient”) doesn’t make it so but then again what does?

Bret Taylor, OpenAI chairman and veteran tech executive, told the Acquired podcast that AI is “obviating” his identity as a computer programmer and is “like an Iron Man suit for all of us,” calling the technology an inflection point in society and technology comparable to the internet. - Fortune

UC Berkeley lab developed HiL-SERL, combining demonstrations, human corrective feedback and RL to teach robots complex, unpredictable tasks (Jenga, flipping an egg, passing objects, assembling a motherboard, car dashboard and timing belt) with 100% success and rapid learning times - TechXplore

Analysts foresee large markets for humanoids, but they still suck. Real-world dexterity, reasoning, and adaptability remain limiting factors despite reinforcement-learning advances. - The Conversation

A new Boston Dynamics video shows it somewhat jankily accomplishing complex tasks with planning, while humans try to interfere. - YouTube

Stability AI study compares AI music generation to previous innovations like sampling, autotune, and modern pianos, says pros use it to innovate, blend styles, help with lyrics. - Music Business Worldwide

AI has crossed an 'aesthetic Turing Test,' or uncanny valley, complicating authenticity in ads, music, and imagery. - The Conversation

A study found LLMs less persuasive than humans using moral appeals, except among users predisposed to view AI as superior or human-like. - The Register

AI Becomes Hollywood’s New Secret Weapon in Talent Negotiations: ‘The Data Doesn’t Lie’ - Variety

Double Descent? AQR’s ‘Hard to Believe’ Study Spurs Clash Over AI Use for Quants - Bloomberg

Does The Future of AI Lies in Neuroscience and Research on Monkeys? - The Wall Street Journal

Voluntary U.S. AI safety pledges seen to have limited impact; Apple scored lowest on compliance indicators. – Fast Company - Fast Company

Amazon quietly updated the Amazon Q Developer VS Code extension to curb prompt injection and code execution risks after a researcher showed exploits enabling exfiltration and RCE - The Register

MCP servers go GA in Visual Studio; researchers warn of compositional risks when combining servers and ingesting untrusted inputs. - The Register

Windows 11 test brings AI file search to the Copilot app - The Verge

Attackers abuse AI website builder Lovable to host phishing and malware at scale; platform adds automated defenses. - TechRadar

Wired and Business Insider have removed freelance articles that appear to be written by AI under a fake name about things and people that don't exist, lifestyle stuff like people getting married in Minecraft, Disney influencer superfans. - Press Gazette

Tuning your car for high performance with AI - Supercar Blondie

Don't arouse the ire of the AI gods - YouTube

Follow the latest AI headlines via SkynetAndChill.com on Bluesky