🍭 Chips, Chaos, and ChatGPT 🤖

Good morning. I tried an AI browser last night.

It ordered a pizza and subscribed me to Duolingo Plus.
Only wanted the Pizza, not to learn italian…

Let’s dig in. 👇

🍭 What’s Cookin’:

• AI browsers promise convenience and invite chaos

• OpenAI’s chip binge piles pressure on TSMC

• Anthropic just booked a million TPUs

• Palantir quietly rewires U.S. AI infrastructure

• 5 new AI tools worth your attention

_ToolBox™
⚒️ 5 AI Tools to Find Signals and Win Customers

📱 v0 for iOS
Your AI dev cofounder, now pocket-sized. Design, iterate, and scale full-stack apps from your phone.

🎨 Alai
AI slides that actually look good. Pick from 4 smart options per slide and edit without touching a single pixel.

🚀 AI CMO by Lindy
Automate your marketing brain. Launch 1,000+ AI-driven experiments in minutes—Sora, Veo, GPT, all on autopilot.

🧠 Presentation 2.0 by CubeOne
Forget slides—CubeOne gives you the script, visuals, and story so you’re stage-ready in 5 minutes flat.

🍭 Snack Prompt
The prompter’s playground. Discover, upvote, and remix the best ChatGPT prompts from the community that actually uses them.

AI Browsers
🌐 Agentic Browsing: Powerful, Inevitable, And (For Now) Fragile

The Bite: AI browsers promise hands-free web work.

Clicking links, filling forms, and stitching workflows across your accounts.

The leaders right now:
Perplexity’s Comet (early mover) and OpenAI’s ChatGPT Atlas (the new heat), with solid newcomers like Dia.

They’re useful for simple tasks until the security risk enters the chat.

The headline risk:
Prompt injection, where a booby-trapped page quietly instructs your agent to leak data or act on your behalf.

Snacks:

• Agents often ask for deep access: email, calendar, contacts.
  

• Prompt injections can hide in text or inside image metadata.
  

• Brave’s research: this is a systemic problem across AI browsers.
  

• OpenAI added logged-out mode to shrink the blast radius.
  

• Perplexity says it flags injections in real time; still not foolproof.
  

• Experts: models struggle to tell who is giving instructions.

Why it Bites: Agentic browsing probably will become the default.
You’ll delegate more of the boring web to machines.

But these tools get more helpful the more keys you hand them, and right now, provenance of instructions is unsolved.

Treat agents like a smart intern with your passwords:
Use unique creds + MFA/2FA, limit scopes, and keep them away from banking and health until defenses mature.

Steal This Prompt
🎃 Halloween Helper

Your personal Halloween consigliere — from cursed costume inspo to party planning chaos and pumpkin perfection.

• Plan a “zombie Gatsby” dinner party with eerie elegance

• Generate an AI-haunted house itinerary (complete with fake Yelp reviews)

• Summon trick-or-treat safety tips written like an Edgar Allan Poe poem

1. Click this link (Halloween Helper)

2. Paste into GPT

3. Replace the # with your Halloween goal

4. Watch it conjure magic — or mild terror

   
   🍭 Try the prompt on Snack Prompt 🍭

Answer.AI
👨‍🏫 Learn a different way to use AI

Ever tried to build something with AI and gotten stuck?

You're not alone.

Join the Solve It With Code course with Jeremy Howard (Inventor of the LLM) & Eric Ries (The Lean Startup). Their method teaches how to think with AI, not just prompt. 

Break challenges into small pieces, then build iteratively. 

The 5-week course includes:

✅ 10 lessons taught live
✅ Access to the Solveit platform 
✅ Starts Nov 3

_{Everything Else}
🧠 You Need to Know

🇨🇳 Trump and Xi Set to Finalize TikTok Deal

→ The U.S. and China will “consummate” a TikTok agreement on Thursday, putting the app’s U.S. operations under a new board led by Oracle and Fox.

📉 Big Tech Earnings Clouded by AI Bubble Fears

→ Microsoft, Alphabet, Amazon, and Meta report this week amid warnings that AI valuations may be outpacing reality.

🧑‍💻 Google AI Studio Launches ‘Vibe Coding’

→ Google unveiled a new AI Studio experience where users can describe an app and watch Gemini build it—no APIs, no code. It turns the prompt itself into a dev tool, complete with an “I’m Feeling Lucky” brainstorming mode.

⚙️ Palantir Quietly Rewires America’s AI Infrastructure

→ A new partnership with Lumen gives Palantir access to one of the largest U.S. fiber networks, pairing its models with real-time data pipes.

☁️ Anthropic Expands Compute With Google Cloud TPUs

→ Anthropic is adding up to one million Google TPUs, worth tens of billions, to scale Claude and research. The deal boosts its 300K business users and deepens ties with Google, even as it keeps AWS and NVIDIA in the mix.

_{OpenAI <> TSMC}
⚙️ The World Runs on One Factory. And It’s Getting Crowded…

The Bite: OpenAI just locked in chip deals with AMD and Broadcom to build the next generation of AI silicon.
Tens of billions of dollars’ worth.

Ok that sounds great for people like you and me so what’s the problem?

Whether it says OpenAI, AMD, or NVIDIA on the label, they all roll off the same line: TSMC:

The world’s most important (and most overworked) factory.

The more AI we build, the tighter that bottleneck gets.

Snacks:

• OpenAI’s AMD deal = 6 gigawatts of GPUs by 2029.
  

• Broadcom partnership adds 10 gigawatts of AI accelerators + networking.
  

• Both projects depend entirely on TSMC’s 3nm and 2nm fabs.
  

• TSMC already makes chips for Apple, NVIDIA, AMD… even Intel.
  

• 75%+ of TSMC’s business = North America.
  

• Experts call it the “single point of failure” in the global economy.

Why it bites: OpenAI’s hardware ambitions make sense.
More chips, more control, less NVIDIA tax.

But TSMC isn’t infinite.

Every new AI contract crams more compute into an island already stretched thin by Apple and NVIDIA.

If Taiwan sneezes politically, geologically, or economically the whole AI boom catches a cold.

We built the future on one factory floor, and it’s starting to creak.