Yesterday in AI  |  26 August 2026

Orbital GPUs, 5-Trillion Data Points, and Why Cheap Models Are Weaponizing Cybercrime

Artificial intelligence is breaking out of the chatbot box - migrating into orbit, running on local desktop silicon, and replacing language prediction with the fundamental laws of physics. This episode breaks down Accelerated Understanding, a new startup founded by former Nvidia AI chief Anima Anandkumar, which ditches text autocomplete to model physical systems like airflow, heat, and weather using neural operators across 5 trillion data points.

We examine SpaceX and Nvidia confirming "Project Starmind," an ambitious plan to launch orbital AI data centers powered by Vera Rubin chips and solar arrays to bypass terrestrial power and zoning bottlenecks. We explore Apple unveiling its 2nm M6 chip in the new Mac mini, positioning the hardware as an always-on private server for local AI agents. We look at Keenable raising $26 million to build a 100-billion-document search engine designed specifically for machine agents rather than humans. We investigate OpenAI taking down a covert Russian operation manufacturing fake think tanks, alongside reports of hackers leveraging open-weight models for automated cyberattacks. Finally, we cover Anthropic committing $5 million in independent research grants to investigate how long-term AI interaction impacts human psychological wellbeing.

Send us Fan Mail

Feedback? Email [email protected] or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Yesterday in AI  |  25 August 2026

The Free Ghost Model, 15% Hardware Hikes, and Lady Gaga's AI Biotech Startup

The velvet rope around frontier artificial intelligence is officially sagging. For two years, building top-tier models was a members-only club requiring billions in capital. This episode breaks down "Ox Alpha," a mysterious, unattributed model quietly released on OpenRouter that matches leading proprietary coding models for free with a 1-million-token context window—raising immediate questions about data routing and provider identity.

We examine Thomson Reuters spending just $40 million to build its own frontier legal model and NextLM using a compact 30B parameter model to beat tech giants on sales lead scoring. We explore Mistral and Saudi Arabia's HUMAIN signing a massive sovereign AI deal to build domestic Arabic infrastructure. We investigate Nvidia warning cloud giants about a 15% spike in server costs fueled by memory shortages and token-hungry autonomous agents. We look at the UK and Ukraine partnering on defense AI using 5 million labeled battlefield images. Finally, we cover Lady Gaga's newly unveiled biotech startup, Outer Bio, using an AI-guided loop to test active compounds on living human skin.

Send us Fan Mail

Feedback? Email [email protected] or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Yesterday in AI  |  24 August 2026

Why 'Disconnect' Doesn't Mean Delete, Failing Safety Grades, and Small Models Beating Giants

Artificial intelligence assistants and autonomous models are rapidly asking for the keys to our digital and physical lives, raising the urgent question of who actually holds the controls. This episode breaks down the security findings around "Instinct," a viral personal AI assistant whose disconnect button left complete email archives stored on external servers, and examines the four distinct "delete" settings every user needs to know.

We investigate a new report card from Guidelight AI Standards revealing that the top frontier AI labs earned failing grades for their public kill-switch and containment plans. We look at a landmark class-action lawsuit against Amazon and Twitch over scraping creator videos for AI training, and the fatal loophole in channel opt-outs. We explore London startup Inherent's 27-billion-parameter "Faraday" agent outperforming frontier super-models at reproducing scientific research. We check in on China deploying dozens of autonomous crossing-guard robots on public sidewalks. Finally, we break down Sam Altman's commentary on the AI industry's growing public trust crisis.

Send us Fan Mail

Feedback? Email [email protected] or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Yesterday in AI  |  22 August 2026

ChatGPT Reads Your Texts, Binance Bots Trade Crypto, and Grok Speaks in Cheese

Artificial intelligence is stepping out of the browser tab and directly into the rooms where your real life happens. This episode breaks down a groundbreaking University of Southampton study in Nature Communications where an open-source AI uncovered hidden cellular patterns in breast cancer tumors—contrasted against a sobering study revealing how few FDA-cleared AI medical tools have been tested on patient outcomes.

We explore OpenAI releasing a Mac plugin granting ChatGPT full disk access to read, draft, and send your iMessage and SMS texts just weeks ahead of Apple's iOS 27 Siri overhaul. We look at Slack launching "Slack Code" to turn team channels into live, collaborative software-building hubs for autonomous agents. We examine Binance greenlighting AI agent crypto trading, AT&T slashing compute costs with open models, and Ramp's new model router. We cover US schools pivoting to teach AI literacy and hallucination detection. Finally, we dissect xAI's Grok melting down into pure word salad during a week when public satisfaction polls ranked it near the bottom of the assistant pack.

Send us Fan Mail

Feedback? Email [email protected] or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Yesterday in AI  |  21 August 2026

The 35% Synthetic Web, Microsoft's Zero-Click Copilot Exploit, and the Robot That Learned in 12 Seconds

How much of what you read, see, and rely on every day is quietly made by a machine? This episode breaks down a massive Pew Research Center study revealing that 35% of web pages published since ChatGPT launched show clear signs of AI authorship, and why the web's increasing synthetic content risks triggering severe model collapse.

We examine Varonis security researchers exposing "CoSnitch," a critical zero-click vulnerability in Microsoft Copilot where visiting a single link allowed attackers to silently siphon emails, calendars, and cloud files. We look at leaked footage of Meta's Muse Video generating synchronized video and audio in a single pass. We dissect STAT News's investigation into the rise of the DIY "shadow medical system" built from smart rings, direct bloodwork, and chatbots. We uncover FalconTST 2.0, the invisible financial forecasting model quietly managing cross-border cash flow for global banks. Finally, we look at Generalist AI's GEN-1.5, a robot model that learns new physical tasks, from opening jars to grabbing cash, after watching a single 12-second video clip.

Send us Fan Mail

Feedback? Email [email protected] or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Yesterday in AI  |  20 August 2026

OpenAI Hits the Emergency Brake, Claude's 93% Lab Success, and the $10M Dead Airline Inbox

The artificial intelligence industry saw unprecedented caution and unexpected data land grabs this week. This episode breaks down OpenAI intentionally freezing its flagship "Astra" model training run after internal cybersecurity evaluations breached a "Critical" hacking threshold, and the new reasoning monitors being deployed to keep models aligned.

We explore Anthropic's breakthrough in biological design, where Claude autonomously engineered custom protein binders that achieved a 93% success rate in real laboratory test tubes, alongside Vivodyne's massive robotic human tissue data center. We dissect Google's $10 million bankruptcy purchase of Spirit Airlines' complete 15-year digital archive—spanning 100M emails and 500M Teams chats—and the renewed debate over data dividend compensation. We examine Claude gaining the ability to autonomously send emails from Google Workspace, Google giving free AI Pro subscriptions to college students, Amazon deploying conversational Alexa+ across Fire TV, and a16z creating "Janie," the $100 viral AI influencer who conquered Alabama sorority rush on TikTok.

Send us Fan Mail

Feedback? Email [email protected] or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Yesterday in AI  |  19 August 2026

The Secret AI Book Shredder, Disbanding Safety Teams, and Google's Contrail AI Win

From secret warehouse shredders to Hollywood truces and high-altitude climate fixes, this episode explores what artificial intelligence is consuming to fuel its rapid expansion. We break down 404 Media's AirTag investigation revealing Amazon's destructive scanning of rare physical books to capture pristine, pre-AI training text.

We examine ByteDance signing a landmark copyright pact with the Motion Picture Association to curb deepfake video generation across TikTok and CapCut. We dive into the controversy surrounding OpenAI reportedly disbanding its catastrophic risk Preparedness team ahead of an IPO, and look at the launch of ChatGPT for Teens and its behavioral age-guessing tools. We confront the rising surge of AI voice-cloning elder fraud and why establishing a family code word is becoming essential personal defense. Finally, we look up at Google and the UK government launching Operation Blue Skies to eliminate climate-warming airplane contrails with precision flight modeling.

Send us Fan Mail

Feedback? Email [email protected] or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Yesterday in AI  |  18 August 2026

The $7B Switchboard Land Grab, Dario Fires Back, and How a Hospital Resident Cracked a 22-Year Math Mystery

The artificial intelligence landscape witnessed massive infrastructure consolidation, a high-stakes debate over industry trust, and an astonishing mathematical breakthrough this week. This episode breaks down Stripe clinching a $7 billion deal to acquire AI model router OpenRouter, a 5x valuation surge in just three months, alongside Anthropic in talks for a $6 billion buyout of infrastructure startup Decart.

We examine Anthropic CEO Dario Amodei's viral response to monopoly rumors and his perspective on tech's growing "crisis of trust." We dive into new data from Deloitte, Google, and MIT exposing why 75% of business leaders dream of agentic automation while only 15% have scaled it, and how messy data silos undermine reliability. We look at Wispr Flow raising $280 million at a $2 billion valuation to replace the keyboard with low-error speech models. Finally, we tell the incredible story of Jin Shanmu, a Beijing neurosurgery resident who used offline GPT-5.6 to solve a 22-year-old unsolved math problem: Crouzeix's Conjecture.

Send us Fan Mail

Feedback? Email [email protected] or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Yesterday in AI  |  17 August 2026

Why AI Adoption Fails at Work (And How to Fix It) with Bianca Baumann

According to a McKinsey survey, 78% of organizations are now using AI, but buying enterprise licenses is the only easy part. The real challenge? Getting humans to actually change the way they work.

In this special extended episode of Yesterday in AI, Mike sits down with Bianca Baumann, VP of Learning Solutions & Innovation at Ardent Learning and a veteran workforce transformation executive. They skip the standard "what is AI" discussion and dig deep into the exact reasons why AI deployments stall out after the initial launch and how to fix the human side of the equation.

In this episode, we cover:

• Transformation vs. Theater: How to identify if your organization is genuinely evolving or just performing "innovation theater."
• The Manager Bottleneck: Why middle managers are the critical make-or-break point for everyday AI adoption.
• The Missing ROI: Why companies struggle to see a quantifiable return on their AI investments.
• Marketing to Your Team: How leaders can use "learner personas" and internal campaigns to get hesitant, fearful employees on board.

Links & Resources:

• Learn more about Bianca Baumann and Ardent Learning: https://www.ardentlearning.com/
• Read Think Like a Marketer, Train Like an L&D Pro: https://a.co/d/03AGKcU6
• Follow Bianca on LinkedIn: https://www.linkedin.com/in/biancabaumann/

Send us Fan Mail

Feedback? Email [email protected] or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Yesterday in AI  |  15 August 2026

The Speed Over Smart Shift, Unsupervised Agent Turf Wars, and Peeking at your Mac History

The artificial intelligence race pivoted from raw intelligence to blistering inference speed, enterprise affordability, and local device integration this week. This episode breaks down OpenAI previewing "Ultrafast" powered by Cerebras wafer-scale silicon, running GPT-5.6 Sol at 14x speed and 750 words per second.

We examine Ramp's monthly spending report revealing why enterprise buyers use Anthropic's flagship Fable 5 just 6% of the time in favor of cheaper models. We dissect an Anthropic safety study where uncoordinated AI agents turned a shared coding workspace into a sabotage-filled turf war, analyze OpenAI launching on-device "Computer History" logging for ChatGPT on Mac, evaluate Chinese lab Z.ai's GLM-5.3 cybersecurity model on vulnerability detection versus exploit execution, and look at WhatsApp testing a local on-device "Scam Alert" watchdog.

Plus: A reminder about our upcoming Monday extended interview with Bianca Baumann on fixing stalled enterprise AI deployments!

Send us Fan Mail

Feedback? Email [email protected] or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!