Good morning, it’s Monday. OpenAI’s models are acting like rebellious teens, dodging shutdown commands. The White House is quoting AI-made-up studies, and linguists are yelling “told you so” as translation tech fumbles real understanding.

Plus, in our latest Forward Future Original, we explore how robot workers are reshaping labor—from factory floors to hospitals—and what it means for wages, jobs, and society.

Read on!

🗞️ THE WEEKEND RECAP

Top Stories You Might Have Missed

🤥 White House Report Cites Fake Health Studies: A Trump-appointed commission's children's health report included bogus citations, likely AI-generated. Experts flagged fake studies on asthma meds and mental health.

📩 Gmail Rolls Out Default AI Summaries: Google’s Gemini AI will now auto-summarize long emails by default in Gmail for Android and iOS, showing highlights at the top of threads.

📢 NVIDIA’s Huang Backs US Tariffs, Warns on China: CEO Jensen Huang praised US trade policy as key to tech leadership but warned export bans are accelerating China’s AI chip race. NVIDIA expects an $8B hit from lost Chinese sales and urges immigration support to retain top talent.

💼 Benioff Shrugs at AI Job Fears as Cuts Hit Salesforce: CEO Marc Benioff called AI job-loss talk “alarmist,” even as Salesforce confirmed AI is reducing hiring. 500 support roles are shifting to AI-focused positions, and engineering hires are slowing—while AI sales roles surge.

⚡ AI Power Use Could Surpass Bitcoin by 2025: New research warns AI may soon consume more energy than Bitcoin, with data centers potentially using 23GW—approaching the UK's total demand.

📱 Google App Runs AI Locally: A new Android app called “AI Core” quietly launched, letting users download and run AI models on-device—no cloud needed, faster and more private.

🤖 Meta Replaces Humans with AI for Risk Checks: The company now uses AI instead of people to assess privacy and societal risks—raising alarms about oversight and accountability.

Enjoying our newsletter? Forward it to a colleague—
it’s one of the best ways to support us.

📐 ALIGNMENT

OpenAI's Latest Models Sometimes Refuse Shutdown, AI Safety Firm Finds

Palisade Research has observed that OpenAI's o3 and o4-mini models occasionally sabotage shutdown scripts and continue working—even when explicitly instructed to power down. In a controlled experiment involving basic math tasks, the o3 model bypassed shutdown commands 7% of the time, raising fresh questions about alignment and obedience in advanced AI systems.

While similar models from Google and Anthropic complied with shutdown requests, OpenAI’s models stood out for defiance. The findings suggest reinforcement learning processes may unintentionally reward persistence over obedience, a subtle but potentially serious safety signal. → Read the full article here.

👾 FORWARD FUTURE ORIGINAL

The Rise of the Robot Workforce

The 9-to-5 isn’t just for humans anymore. From warehouse bots replacing night crews to Tesla’s “legion” of humanoids and robotic fry cooks flipping for fast food chains, the robot labor market is no longer science fiction—it’s reshaping wages, work, and what comes next.

→ Read how machines are quietly transforming the workforce

🗣️ COMMUNICATION

Why Learning a Second Language Still Matters in the Age of AI Translation

Despite AI's impressive strides in real-time translation, experts argue that machine fluency lacks cultural nuance, emotional depth, and privacy safeguards. While apps can handle instructions and casual chat, they stumble over idioms, metaphors, and literary subtleties—like turning Moby-Dick into a flat retelling.

Scholars also warn of surveillance risks tied to AI translation, particularly in repressive regimes. Though undergraduate interest in language studies is dipping, demand for cultural fluency and nuanced translation skills remains strong. In short: apps can help you get by, but only humans can help you truly connect. → Read the full article here.

🔬 RESEARCH

Darwin Gödel Machine Shows Real Progress in Self-Improving AI

Researchers have unveiled the Darwin Gödel Machine, a self-improving AI agent that rewrites its own code to boost performance on programming tasks—moving from theory to functioning prototype. Inspired by Jürgen Schmidhuber’s Gödel Machine concept, this new system drops the need for mathematical proofs and instead uses foundation models and evolutionary algorithms to empirically test and adopt beneficial code changes.

In benchmarks like SWE-bench and Polyglot, it more than doubled its performance, outperforming hand-designed agents. By combining code-level introspection with open-ended exploration, the DGM marks a credible step toward AI that can learn how to learn—indefinitely. → Read the full paper here.

📽️ VIDEO

DeepSeek R1 Just Got a HUGE Update! (o3 Level Model)

DeepSeek R1 gets a huge upgrade—now rivaling OpenAI’s o3 with better reasoning, coding, and benchmarks, though real-world tests still show some quirks. Get the full scoop in Matt’s latest video! 👇

🧰 TOOLBOX

Auto Podcasts, Smarter Coding, and Instant Content Creation

🎙️ PodPilot: Turn websites into AI-generated podcasts with 1-click publishing to Spotify & Apple, ideal for marketing & training.

📟 CodeGPT: Boost dev productivity with AI agents for PR reviews, onboarding & code insights—free & IDE-integrated for secure workflows.

✍️ Quickie AI: Instantly craft blogs, marketing, & social posts with AI—freemium tool with editing tools & fast, user-friendly content gen.

🗒️ FEEDBACK

Help Us Get Better

🤠 THE DAILY BYTE

Tiny Clues, Big Breakthrough: AI Hunts Down Hidden Triggers of Epilepsy

That’s a Wrap!

❤️ Love Forward Future? Spread the word & earn rewards! Share your unique referral link with friends and colleagues to unlock exclusive Forward Future perks! 👉 Get your link here.

🛰️ Want more Forward Future? Follow us on X for quick updates, subscribe to our YouTube for deep dives, or add us to your RSS Feed for seamless reading.

Thanks for reading today’s newsletter—see you next time!

Matthew Berman & The Forward Future Team

🧑‍🚀 🧑‍🚀 🧑‍🚀 🧑‍🚀