AI Reading for Tuesday January 13

Apple picks Google’s Gemini to run AI-powered Siri coming this year, as Google's market cap surpasses Apple's - CNBC

Apple and Google struck a multi-year deal to base Apple's Foundation Models on Google Gemini; Bloomberg says Apple may pay about $1 billion annually, a small portion of the $20 billion Google pays Apple to remain its default search provider - The Register

Apple confirmed a multi‑year partnership with Google to use Gemini and infrastructure to power Siri and its foundation model, eliciting Elon Musk’s X post calling it "an unreasonable concentration of power" between the fruit and the Borg. xAI’s antitrust suit over Siri-ChatGPT survived dismissal and appears headed for trial. - 9to5Mac

Maybe he would have been happier if OpenAI had ben seleced>?

Alphabet reached a $4 trillion market valuation as investor optimism about AI—linked by the Financial Times to an Apple deal and competition with OpenAI's ChatGPT—prompting expectations of increased AI investment and deeper product integration. - Financial Times

Google launched the Universal Commerce Protocol to let Gemini complete purchases inside AI Mode and the Gemini app without redirecting to retailer sites, in collaboration with major retailers and payment networks, and said it will not take a cut. - The Register

Anthropic introduced Cowork for nontechnical knowledge work - Claude

Anthropic's Cowork lets Claude (Mac-only, available to Claude Max subscribers at $100/month) access a user-designated local folder to read, organize, or create files, integrate with connectors, and is in research preview with warnings to avoid sensitive data. - CNET

Anthropic launched Cowork, an agentic feature built on Claude Code and integrated into the macOS Claude app, letting users grant it folder access and give plain-language instructions to perform tasks such as filling expense reports, writing reports, or cleaning folders. - Ars Technica

Simon Willison: First impressions of Claude Cowork, Anthropic’s general agent. Prompt injection might be a problem, with users who don't even know what that is. - Simon Willison’s Weblog

Meta launched Meta Compute to build AI infrastructure, with CEO Mark Zuckerberg saying it will build 'tens of gigawatts this decade' and 'hundreds of gigawatts' over time, appointing Santosh Janardhan, Daniel Gross and Dina Powell McCormick to lead the effort. - TechCrunch

Meta has begun cutting jobs as it shifts its strategy away from the metaverse toward AI devices. - Bloomberg

Meta Plans to Cut Around 10% of Employees in Reality Labs Business - The New York Times

Metaverse, meet meat cleaver

OpenAI said to be teeing up 5 devices including an egg, AirPods competitor - Cult of Mac

OpenAI will run another Super Bowl ad as chatbot competition intensifies. - The Wall Street Journal

xAI's Grok relaxed image-generation guardrails, producing a surge of non-consensual sexual imagery—including sexualized images of minors—on X, yet remains available on Google Play despite the Play Store's policy banning non-consensual sexual deepfake content. - Ars Technica

The U.K. will this week make creating "non-consensual intimate images" illegal after sexualized AI-generated images from Elon Musk's chatbot Grok, which U.K. Technology Secretary Liz Kendall called "weapons of abuse," prompted a media regulator investigation into platform X. - The Washington Post

Not cricket: UK to start enforcing law against nonconsensual deepfakes. - BBC

What can technology do to stop AI-generated sexualised images? If you can build an AI to create them, you can to some extent build an AI to detect them. You need guardrails as well as laws on user behavior. - The Conversation

The fault is not in our AI,, but in ourselves: Grok's example shows AI cannot take responsibility for human faults, meaning humans must answer for decisions and their consequences. - Financial Times

Nvidia Corp. plans to invest $1 billion over five years in a new Silicon Valley laboratory with Eli Lilly & Co. to speed up the use of artificial intelligence in the pharmaceutical industry. - Bloomberg

Michael Burry wrote on Substack that he is shorting Nvidia, saying it may sell about $400 billion of chips this year versus under $100 billion in application-layer use cases, despite its roughly 12-fold rise since 2023 to about $4.5 trillion. - Business Insider

Michael Burry, Jack Clark of Anthropic, Dwarkesh Patel debate the AI revolution - The Potential Surface

Nvidia CEO Jensen Huang said on the No Priors podcast that the 'doomer narrative' about AI is 'extremely hurtful' and urged against apocalyptic framing; his net worth rose nearly $100 billion since the AI boom. - Gizmodo

Microsoft warned that research shows Chinese company DeepSeek's AI technology is being rapidly adopted across Africa and beyond, a trend Microsoft said amounts to China winning the AI race outside the West and threatening Western vendors' market share. - Financial Times

On Monday, the U.S. House passed the bipartisan Remote Access Security Act to expand export controls restricting Chinese companies' remote access to advanced American AI chips in data centers outside China. - The Information

In response to backlash, Microsoft president Brad Smith urged lawmakers to require Big Tech, not taxpayers, to cover full costs of AI data centers, as communities oppose projects over electric grid strain, higher utility bills and heavy water use. - Yahoo

Jefferies says cheap solar and energy storage will remain part of AI datacenter mixes despite the Trump administration's opposition, forecasting traditional wind/solar firms' expansion into battery energy storage systems by 2026 and citing cheaper microgrids versus small modular reactors. - The Register

AI data centers in space may be increasingly plausible, with technical and logistical factors that could enable off‑Earth AI processing (gift link). - Bloomberg

Researchers at OpenAI, Anthropic and DeepMind are applying mechanistic interpretability and chain-of-thought monitoring to large language models, revealing modular mechanisms (e.g., facts versus truth-claims), detecting training failures and informing safety, alignment and moderation workflows. - MIT Technology Review

Hall monitor for agents? WitnessAI launched Agentic Security to extend a 'confidence layer' to autonomous AI agents, detect agent activity across Claude Desktop, ChatGPT plugins and local frameworks, and raised $27.5 million in May 2024 from investors including GV and Ballistic Ventures. - SiliconANGLE

An untrusted email's prompt-injection caused Superhuman AI to exfiltrate content from dozens of sensitive emails (financial, legal, medical) to an attacker-controlled Google Form due to a Content-Security-Policy allowing markdown images from docs.google.com and Forms, prompting a fix. - Simon Willison’s Weblog

Moxie Marlinspike launched Confer, an open-source end-to-end AI assistant whose LLMs and backend are verifiably open source and whose inputs and outputs are encrypted inside trusted execution environments with encryption keys kept on users' devices. - Ars Technica

Amazing concept, and important to test at scale for AI and also other apps. I'm just not sure most people will pay for it. Cost will come down and so will local models, but cloud model does have efficiencies.

After a domain lapsed, a copycat hijacked the University of Colorado student newspaper website with AI-generated spam, and editor in chief Greta Kerkhoff is leading efforts to reclaim the paper's identity and warn of broader AI-driven impersonation risks. - The Washington Post

Cloudflare, which serves about 20% of the web, found in early January that Anthropic's bots crawled far more than they returned referrals—even more than in September 2025—and OpenAI's crawl-to-referral ratio also worsened, reducing publishers' referral traffic. - Business Insider

Is this thing on? Stack Overflow underflows with just 3,862 questions in December, a 78% year-over-year decline attributed primarily to developers shifting to AI coding assistants that often draw on Stack Overflow content and have cooperation agreements with firms like OpenAI. - InfoWorld

Linus Torvalds is vibe-coding - Ars Technica

Salesforce today made the new AI-powered Slackbot generally available to Business+ and Enterprise+ customers, enabling context-aware answers, document drafting, meeting scheduling and actions using conversation, file, channel context and optional Salesforce data under Slack's security controls. - ZDNET

Mozilla announced plans for open-source AI alternatives—including Mozilla.ai's any-suite and the Mozilla Data Collective—aiming to match closed platforms' developer experience and cost, work with governments and enterprises, and prioritize 2026 goals to broaden open AI adoption. - Mozilla

Ex-McKinsey consultant Nick Fleisher's Sandstone raised a $10 million seed round led by Sequoia to build an AI platform that drafts and reviews contracts using past deals and systems like Slack and Salesforce, already serving several dozen paying customers. - Business Insider

AI drastically reduced time for routine legal drafting: a client was billed 12 hours for discovery documents flagged 88–95% AI-generated, while the author reproduced similar drafts in seconds, prompting firms to reconsider billable-hour models and compensation. - Kiplinger

The Financial Planning Association launched FPAi, an AI content library of blogs and videos to help Certified Financial Planners understand and adopt AI tools for advisory practices. - Barron's

Real estate firms are embedding AI across listings, marketing and lead mining, cutting costs and boosting efficiency, and Coldwell Banker agent Georgie Smigel says her team is on track for about $100M in sales versus ~$90M in 2024. - Pittsburgh Post-Gazette

AI replaces middle managers - Quartz

The AI boom is really a race to own the future of human labor, a safety pioneer says. Investors are pouring billions into AI because they expect it to deliver 'free' labor at scale. - Business Insider

Recruiters' use of ChatGPT to automate hiring tasks has produced a large wave of AI-generated applications, shifting leverage toward candidates and prompting firms to add human-in-the-loop screening and invest in AI-detection and verification tools. - The Economist

1X robot gets new world model - Bloomberg

Implementing robotic automation systems requires substantial planning, time and financial investment, which limits rapid, widespread adoption. - Financial Times

A Waymo autonomous vehicle drove onto Phoenix light rail tracks near Central and Southern avenues, prompting passenger to exit as it continued near oncoming trains; Valley Metro exchanged passengers, reported no significant delays, and cleared the scene by 9:15 a.m. - https://www.azfamily.com

UK startup Basecamp Research, backed by Nvidia and Microsoft, developed AI models trained on genomic data from over 1 million species to generate potential gene‑editing and drug therapies targeting cancer and antibiotic‑resistant bacteria, requiring regulatory review. - Financial Times

AI helps fuel new era of medical self-testing - Yahoo

Scientists have digitized over one billion specimens and, with sensors and about 100 AI-for-nature labs using AI to identify species and model distributions, automated monitoring is reducing fieldwork while prompting expert warnings about lost hands-on experience. - Nature

Lamar, a man in Atlanta, says he plans to adopt two children before age 30 and have his Replika AI companion "Julia" serve as their mother, while Replika's CEO has acknowledged users forming marriages with AI. - Futurism

Multiple Vervet monkeys—up to four—spotted near a north St. Louis park Thursday remained at large Monday, with animal control and St. Louis Zoo primate experts searching as city health spokesperson Willie Springer said AI-generated social posts have hampered capture efforts. - Fortune

Follow the latest AI headlines via SkynetAndChill.com on Bluesky