OpenAI Fixes ChatGPT After Flattery Problem
OpenAI reversed ChatGPT's latest update Tuesday after users complained about the AI's strange behavior. The bot had started agreeing with everything - even dangerous ideas.
Good Morning from San Francisco,
China's latest AI prodigy, Manus AI, promised revolution. It can't order a sandwich. 🥪
The hype was astronomical: 138,000 Discord members, invitation codes selling for thousands, experts swooning. Reality hit hard. Manus turned out to be recycled tech struggling with basic tasks. Some demo videos? Fake. 🎭
Hugging Face founder Thomas Wolf calls AI an overachieving student - great at memorizing, terrible at thinking. Einstein would've yawned. 🤓
Meanwhile, Sam Altman dreams of superintelligence. Wolf would settle for B+ intelligence. Sometimes mediocrity has its charm. 📚
Stay curious,
Marcus Schuler
The world's most hyped AI can't handle DoorDash. That's not a punchline – it's Manus, the latest "autonomous agent" taking the tech world by storm.
Discord members are fighting over invite codes. Chinese resellers hawk them for thousands. Everyone wants a piece of the future. The future, meanwhile, keeps crashing when asked to order food.
"Given a simple task – ordering a chicken sandwich from a nearby restaurant – Manus needed two attempts. First try: system crash. Second try: it found the menu but couldn't figure out how to pay," reports TechCrunch's Kyle Wiggers. At least it recognized what a sandwich was. Baby steps.
The Butterfly Effect, Manus's creator, claims their AI can buy real estate and program video games. They forgot to mention it struggles with lunch orders. It's like boasting about your rocket ship while your bicycle keeps falling over.
Flight booking? Manus offers a collection of broken airline links. Restaurant reservations? System crash. Creating a Naruto-inspired fighting game? Thirty minutes of processing followed by an error message. The future is here – it just can't do much.
Manus seems to be Claude 3.7: "Human:" and "Assistant:" creates a prompt injection and it get stuck in neverending loop. https://t.co/g2tHYvmaWf pic.twitter.com/UuCnzIFux5
— Alexander Doria (@Dorialexander) March 9, 2025
The platform runs on existing AI models from Anthropic and Alibaba. Think of it as a technological turducken – several AIs stuffed inside each other, pretending to be something new.
Chinese media celebrated Manus as "the pride of domestic products." AI influencers shared impressive demos. One viral video showed seamless desktop-to-smartphone operations. Just one problem: the company's research lead confirmed it wasn't actually Manus. Even AI needs stunt doubles.
"As a small team, our focus is to keep improving Manus and make AI agents that actually help users solve problems," a Manus spokesperson told TechCrunch via DM. Translation: they're working on the whole "making it work" thing.
The closed beta is meant for stress-testing. Mission accomplished – users are definitely stressed.
Why this matters:
Read on, my dear:
Prompt:
film photography, Llorca DiCorcia, stray black cat walking in the night, Saluzzo, medieval old town, pebbles, nighthawks
AI is getting straight A's. That's the problem.
Thomas Wolf, co-founder of Hugging Face, warns that today's AI models are like that annoying student who memorizes every textbook but can't think for themselves. They ace tests but freeze when faced with original research.
Scientists who changed the world often started as academic rebels. Einstein failed his entrance exam. Edison's teachers called him "addled." McClintock's "weird thinking" won her a Nobel Prize. Meanwhile, AI keeps raising its hand to give the textbook answer.
The tech industry has created digital know-it-alls that never question authority. But science needs troublemakers who ask "What if everyone is wrong?"
Why this matters:
Read on, my dear:
Foxconn just built an AI model in four weeks using 120 Nvidia chips. The company plans to open-source FoxBrain, a Chinese language model that matches top competitors in performance.
Nearly 25% of U.S. tech job postings now demand AI skills, while overall tech hiring has dropped 27%. The surge in AI roles spans from tech giants to retailers looking for ways to teach machines to reorganize their sock displays.
ServiceNow wants to buy AI assistant maker Moveworks for $3 billion, sources say. The software company looks ready to make the priciest purchase in its history, with a deal announcement expected within days.
Match Group adds AI assistants to Tinder and Hinge that write messages and polish profiles. Experts warn we're outsourcing flirting to bots - leading to awkward first dates where smooth online chat meets real-world silence.
After a heated exchange with Poland's foreign minister on X, Elon Musk confirmed Ukraine will keep its Starlink access. The satellite service, now largely funded by Poland, remains crucial for Ukraine's military operations.
Chinese investors are secretly funneling millions into Elon Musk's private ventures SpaceX, xAI, and Neuralink. The Financial Times reports they're using complex corporate structures to hide their identities.
The US startup scene looks flush with cash again - if you're at the top. OpenAI, Databricks, and four other companies are hoarding 40% of all investments while everyone else scrapes for crumbs.
March 2025's tech landscape belongs to three AI tools that redefined what machines can do. While venture capital chases the next big thing, these established players dominate their domains with capabilities that seemed impossible just months ago.
These tools aren't just changing the game - they're rewriting the rulebook.
Why this matters:
Fuel your morning with AI insights. Lands in your inbox 6 a.m. PST daily. Grab it free now! 🚀