Welcome to this week's edition of Overclocked!

In this issue, the U.S. Senate just tore out a blanket AI-regulation ban, Meta waved eye-watering cheques to lure top scientists, Oracle agreed to power the next generation of OpenAI models, and much more. Let’s dive in ⬇️

In today’s newsletter ↓
🗳️ Senate lets states craft their own AI laws
💰 Meta’s mega-bonuses score three elite hires
⚡ OpenAI locks in a $30 billion Texas power deal
🏢 Microsoft trims staff to double-down on Copilot
🩺 AI hits new highs in hospital diagnostics
📚 Weekly Challenge: Master Google AI Mode in 5 quick tasks

🌠 Perplexity Comet Changes Everything About How We Use Browsers

When Perplexity teased its own browser in April, skeptics asked, “Why build Chrome again?” On July 9th, the company answered with Comet—an AI-powered browser that flips the relationship between search, chat, and webpages. Instead of opening a blank bar, you’re greeted by a conversational panel on the left and a live preview pane on the right. Type a question and Comet visits pages, scrolls sections, and cites every quote in real time.

🛠️ Built-In Agent, Not an Extension

Unlike Arc’s AI sidebar or Microsoft’s Copilot add-on, Comet’s agent is the browser. It uses Perplexity’s in-house LLM plus live crawling; each answer is an annotated mash-up of pages it just scraped. In a demo for TechCrunch, CEO Aravind Srinivas asked, “Compare Nvidia’s Blackwell Ultra to AMD’s MI325 X.” Comet opened six tabs behind the scenes, generated a table, and let him click a row to deep-link the source paragraph.

🔑 Why Power Users Care

• Zero tab debt. Comet calls its snapshots “cards.” They’re searchable later so you can close everything without losing breadcrumbs.
  

• Citations you can’t fake. Hover on any sentence to jump to the exact scroll position in the source article. No more phantom links.
  

• Focus modes. Press Ctrl + K to ask follow-up queries that stay inside the current page set, perfect for literature reviews.
  

Early testers say Comet feels like “having a junior analyst who pre-reads every link.” The Verge notes memory is local for now; cloud sync arrives later with enterprise controls.

⚠️ Caveats and Skepticism

Comet sometimes “over-summarizes,” flattening nuance in legal docs. And because it scrapes on the fly, paywalled sites remain off-limits unless you log in. Privacy auditors also want clarity on how long crawl logs persist.

🌟 Big Picture

If Comet delivers speed without sacrificing source transparency, it could push mainstream browsers to bake AI deeper than toolbars. As one beta user put it: “Google feels like dial-up after 10 minutes in Comet.”

🖼️ Google Veo 3 Adds Image-to-Video Magic

Google just pushed a quietly stunning upgrade to Veo 3: drop a single photo into the Gemini app and it spools out an eight-second, 1080p video with sound in about a minute. Ars Technica found early clips packed with parallax zooms, gentle lens flares, and lifelike ripples on water—all inferred from pixels that never moved in the original shot.

🧰 How It Works

Veo triangulates depth from shadows, edges, and color gradients, then generates “tween” frames to simulate camera motion. A landscape photo can morph into a slow dolly-in, while a street snapshot becomes a subtle pan that reveals extra storefront detail. 

ZDNet’s hands-on says the motion feels “smooth but slightly dreamy,” perfect for B-roll or social teasers, less so for dialogue-heavy scenes where lip sync would matter. Behind the curtain, Veo leans on a diffusion-based frame interpolation model similar to Google’s “Imagen Video,” but fine-tuned for short clips and smartphone latency.

🤌 Why It Matters

Creators accustomed to MidJourney’s budding V-video tool or Runway’s Gen-4 must juggle multiple prompts or seed images. Veo’s one-photo simplicity could be a game-changer for marketers who need quick motion posts without storyboarding.

Limitations remain: free-tier users get five conversions per day, and each clip tops out at eight seconds unless you subscribe to Gemini Advanced. Google also watermarks output by default—a plus for authenticity, a minus for premium branding.

The Weekly Scoop 🍦

📏 Ex-Intel CEO unveils “AlignBench” to score AI values

🩺 Google open-sources MedGemma for healthcare imaging

🚀 xAI debuts Grok 4 and $300/mo “Pro” plan

💻 Goldman Sachs says its autonomous coder cuts dev time 45%

🔢 Robinhood CEO’s AI math startup nears $900M valuation

💰 Nvidia becomes the world’s first $4 trillion publicly traded company

⏱️ Weekly Challenge: Turn One Photo Into a Three-Scene Video With Sound

Here are your step-by-step instructions + a few AI video generation pro tips:

1. Pick a high-res photo with clear foreground, midground, and background elements (e.g., a city street or beach sunset).
   

2. Open the Gemini mobile app → Veo and select “Photo-to-Video.”
   

3. Scene 1 prompt: “Slow zoom-in on the foreground object, add gentle lens flare.”
   

4. Scene 2 prompt: “Tilt up to reveal midground activity, slight parallax.”
   

5. Scene 3 prompt: “Pan wide to sunset background, animate waves and birds.”
   

6. Download each 8-second clip (Veo lets you regenerate until it feels right).
   

7. Use any editor—CapCut, iMovie, or Gemini’s built-in stitcher—to trim each clip to ~2.5 seconds and chain them together.

Pro tips

• 📏 Keep prompts under 20 words; Veo rewards specificity but penalizes verbosity.
  

• 📹 If motion feels jittery, toggle “Stabilize” in advanced settings.
  

• 🔆 Light is your friend: high-contrast photos yield richer depth cues.
  

Show us what you create and share with your followers, family, and friends!

Zoe from Overclocked