Welcome to this week's edition of Overclocked!

This week, are we still underestimating AI’s trajectory? A viral explainer says yes—and we break down why. We also dig into Microsoft’s latest layoffs, recap seven buzzing headlines, and test your critical‑reading skills with an AI fact‑check challenge. ⬇️

In today’s newsletter ↓
🤯 Eric Schmidt thinks we aren’t hyping AI enough
💼 Microsoft layoffs are a warning for all devs
🦽 Google enables AI accessibility tools
🚔 FBI says AI voice clones are wreaking havoc
🔍 Challenge: AI Fact‑Check Face‑Off

Google’s Ex CEO Says We’re Not Hype Enough About AI

Eric Schmidt thinks the chatter around AI is still too quiet. In a viral TED interview, the former Google CEO argues that the true inflection point wasn’t ChatGPT—it was AlphaGo’s legendary Move 37 in 2016. When the Go‑playing system invented a strategy no human had seen in 2,500 years, Schmidt realized “non‑human intelligence” had arrived.

🚀 Exponential Everything

From AlphaGo to GPT‑4o, each new model scales parameters by roughly 10× while slashing inference costs. Capabilities once slated for 2030—real‑time vision, speech, and reasoning on commodity GPUs—are shipping today. At this pace, sub‑$100 phones could host multimodal copilots before the decade’s halfway mark.

🧠 From Words to Plans

Early LLMs were autocomplete engines; the latest reinforcement‑learning agents plan and iterate. Systems like o3 and DeepSeek‑R1 roll forward and backward through a task tree, refining strategies in real time. Schmidt calls this “language → sequence → planning,” a pipeline that ends with agents running business processes end‑to‑end.

⚡ Hardware, Power, Data

The catch? Compute. Schmidt told Congress the U.S. may need 90 gigawatts—roughly 90 nuclear plants—just to feed the next wave of AI data centers. Meanwhile, public internet text is tapped out; future models must synthesize or simulate new data. Energy and training costs, not algorithms, could throttle progress.

🌍 Economic Shockwaves

IMF papers now peg potential task automation at 40% of global labor by 2030, dwarfing earlier McKinsey estimates. Productivity could rise 30 % annually—an economic regime change with no historical parallel. Without massive upskilling, inequality could explode.

🛑 Safety vs. Speed

Yoshua Bengio urges a pause on “agentic” AI, but Schmidt doubts moratoria will stick in a globally competitive race. Instead he advocates guardrails: provenance tracking, human‑in‑the‑loop kill switches, and clear red lines (no recursive self‑improvement, weapon access, or unsupervised replication).

💼 Microsoft Layoffs May Signal Something Deeper

Microsoft’s latest round of layoffs hit software engineers hardest, with internal memos citing “efficiency gains from AI‑assisted development.” Around 1,900 roles were cut on May 14, mostly in Azure and Windows engineering, Bloomberg reports. In Washington state alone, developers accounted for 60% of terminations.

📈 AI Codes While Humans Review

CEO Satya Nadella told investors that 20–30% of Microsoft’s code is now AI‑generated, primarily via GitHub Copilot and in‑house models. Managers claim AI handles boilerplate logic, freeing humans for architecture and security. Critics counter that junior dev pipelines are being gutted, shrinking the next expert cohort.

💡 Productivity or Publicity?

TechCrunch notes engineers have mixed feelings: velocity metrics look great, but reviewers spend more time vetting AI‑written pull requests. “Copilot is an eager intern—it writes fast but you still rewrite half of it,” one senior dev said.

🔮 Industry Ripple

Analysts warn other giants may follow, highlighting a “hollow‑middle” risk: entry roles vanish while elite AI oversight jobs grow. Bootcamps are pivoting to prompt‑engineering modules, and some VCs predict a renaissance of indie devs armed with AI copilots.

For now, Microsoft bets that fewer coders + smarter tools = faster shipping. Whether quality and morale hold up remains to be seen.

The Weekly Scoop 🍦

🎥 Hedra lands $32M to tackle the ‘uncanny valley’ in AI video

⚖️ Harvey legal AI adds Anthropic & Google models, drops OpenAI exclusivity

🔐 FBI warns AI voice clones are impersonating U.S. officials

🌏 UAE to build the world’s largest AI campus after Trump visit

📱 Google rolls out new AI accessibility tools across Android & Chrome

💻 Nvidia set to unveil more Taiwan AI server partnerships at Computex

💡 Challenge of the Week: AI Fact‑Check Face‑Off

In the day and age of AI content, it’s challenging to know what’s real, 100% false, or somewhat in the middle. Use the tools below to begin fighting back against click-bait and learning whether or not your go-to news sources are sticking to the facts.

1. Pick from browser extensions like NewsGuard, Stopaganda Plus, and Media Bias/Fact Check (MBFC), or a reasoning LLM (i.e., o3, o4-mini, Grok 3, Google Pro 2.5).
   

2. Gather three articles: one from a major outlet, one from a lesser‑known blog, and one on a hot‑button issue.
   

3. Run each piece through one plugin and one model. Take note of the ratings on source credibility, bias flags, and any claim‑level alerts. Also, manually fact check the fact-checker, so you can easily calibrate how good or bad it was.
   

4. Compare outputs: Where did tools agree or clash? Which surfaced citations you’d actually click?
   

A 15‑minute audit that could save you from sharing junk forever.

That's all for this week's Overclocked! Are we still under‑hyping AI—or over‑trusting it? Hit reply with your viewpoint, and we might feature you next issue. Stay curious, stay critical!

Zoe from Overclocked