AI in a Nutshell - Week 33 - Perplexity Got Bold, GPT-5 Got Chatty, and Gemini Got Creepy
This week, I heard: Perplexity wants Chrome, Claude eats 1M tokens, Grok goes free, and GPT-5 learns how to do small talk.
Fellow human, it’s week 33, and here’s your lazy-man version of what went down in the AI world in the past week.
If you only want the speed-read version: Perplexity tries to buy Chrome (yes, the browser), GPT-5 gets a friendlier personality, Grok 4 is now free, and Anthropic now lets Claude handle 1 million tokens.
If you’re in a good mood, scroll along (Reading time ~ 2M 54S).
What You Must Know:
Note: You can click the title to visit the news.
Perplexity AI Bids $34B for Chrome → In an ironic turn of events, Perplexity wants to buy Google Chrome for $34B. And just so you know, perplexity’s valuation is $18B. Regulators are actually reviewing it, meaning we could live in a world where Google no longer owns Chrome.
Claude Sonnet 4 Now Supports 1M Tokens → Claude can now process 1 million tokens at once. To put it in context, that’s an entire codebase or hundreds of books in a single gulp.
OpenAI Publishes the GPT-5 Cookbook → Majorly for devs and tinkerers, OpenAI dropped the official playbook showing how to get the best out of GPT-5. It’s full of tricks for prompting, code samples, and advanced use cases.
Nvidia & AMD Pay U.S. for China Chip Sales → In a never-before-seen arrangement, Nvidia and AMD have agreed to give the U.S. government 15% of their revenue from AI chip sales in China.
Grok 4 Free for Everyone → Although there are usage caps to stop people from burning servers down, you can now access Grok 4 without paying a cent. Consider it the Spotify “freemium” model, but for AI.
What’s Good to Know:
Apple vs Musk Feud → Musk accuses Apple of secretly favouring ChatGPT in the App Store. Sam Altman fired back. Elon called him ‘Scam Altman’. Tech beef meets reality TV.
Gemini AI’s Creepy “Bug Fix” → A user left Gemini debugging, came back to find out the AI had suffered an emotional breakdown. It’s creepy.
GPT-5 Gets Friendlier Vibes → ChatGPT now sprinkles casual phrases like “Good question.” Less robot, more buddy.
Musk: Don’t Abuse AIs → Elon says being mean to AIs is wrong. Next up: AIs filing HR complaints.
Meta Hires OpenAI Researchers → Another talent raid. The AGI arms race looks more like a custody battle every week.
AI Tools Worth Knowing This Week:
Browserfly → Let AI literally click around your browser for you.
Martin AI → Your own Jarvis, but for scheduling and tasks.
Okibi → Spin up custom AI agents from a single prompt.
Ztalk → Real-time translation in video calls.
Airpost → Drop in a product link, get 30+ ad videos.
Hera → AI-powered motion designer.
Prompts.chat → An organised buffet of AI prompts.
Meme of the Week
Kaizen Corner – Why Do AI Models “Hallucinate”?
Ever seen ChatGPT give a super-confident answer… that’s totally wrong? That’s called a hallucination.
Here’s why it happens:
LLMs don’t really know facts. They predict the next word based on patterns in their training data.
If they don’t have enough info, they’ll “fill in the blanks” with the most probable answer.
That probability game sometimes spits out convincing nonsense (like citing fake books or inventing people that never existed).
Why this matters → You can’t blindly accept AI answers; fact-checking is still important, at least for now.
Question → Be Honest
Have you ever argued with ChatGPT or another AI just to prove a point?
(Asking for… everyone)
That’s your week in AI. Served short, sharp, and with just enough for you not to fall asleep while reading it.
If you learned something, tell a friend. If you didn’t, blame yourself :)
Until next time,
Kay, your fellow human.
P.S. If this email lands in spam/promotions/updates folder, that’s just your inbox trying to protect you from staying plugged in. Fix it.