Claude Banned, Zhipu Surges, Google Nobel Talent Flees, DeepSeek Funding Excludes Alibaba, and Codex's Comeback!
The data coming out of the AI world recently is genuinely terrifying. Zhipu's market cap skyrocketed past one trillion Hong Kong dollars overnight — a number so staggering it makes you gasp. The Americans had just arrogantly pulled the plug on Claude 5, and Zhipu immediately announced that GLM 5.2 would be free and open-source. It was a perfect one-two knockout, shoving the caught-off-guard Silicon Valley giants face-first into the mud. This wave of blood-and-sweat dividends has left everyone dumbfounded.
What Does the Trillion-Dollar Surge Really Cover Up?
You think the Claude bans were just routine account suspensions? Wrong! This ruthless blade ripped off the hypocritical mask of the entire AI industry. Think about it: I was using Claude 5 to code the night before, smooth as silk. The next morning, my account was gone! The whole internet was in mourning. Even AI godfather Karpathy got kicked off because he wasn't a U.S. citizen.
Then Zhipu stepped in to fill the void. We used to think Zhipu's models were lacking. But when they rolled out that insane million-token context window, helping you pinpoint bugs like finding a needle in a haystack, with token usage so efficient it feels endless — it's like you were queuing at an unaffordable Michelin-starred restaurant, got kicked out, and then a street stall owner laid out a full imperial feast, patting his chest saying, "It's all on the house!" How could you not go crazy?
Let me give you the most vivid example. With other top-tier models, debugging a brutally hard bug would burn through 50% of your limit after just a few queries. You're barely warmed up, and the referee blows the whistle, benching you for five hours. With Zhipu, I ran ten intensive tasks and only used 2% of my quota! Even if I have to redo some work, I'll take it!
Google's Nobel Laureate Flees Overnight — How Fractured Are the Big Tech Internals?
While Zhipu is celebrating, former titan Google's talent pipeline is collapsing. Just last week, the lead inventor behind AlphaFold, who won a Nobel Prize, packed up overnight and jumped ship to archrival Anthropic! You think it's just a desk change? Dead wrong — it's a full-blown exodus! Following him were Character.AI's founder and Transformer co-author Noam Shazeer.
Inside Google, the battle over direction has become life-or-death. One faction of old-school academics keeps shouting, "We need to build deep strength slowly!" Another faction of execs, watching the stock price plummet, is screaming, "Forget deep strength! Crank out a cheap Flash version to make quick cash!" It's like a martial arts novel where one guy only trains internal energy and gets knocked out, while another only trains external moves, goes berserk, and ends up castrating himself!
These Nobel-winning brains are already financially free! Do you think a few million more in salary matters? They want unfettered resources and insane compute power. Endless bickering at Google? Might as well go to Anthropic, where their equity is worth several Zhipus. This loss of core talent is ten thousand times more deadly than a hundred-billion-dollar market cap drop.
Liang Wenfeng Drops $2B to Buy In — AGI on His Terms?
If you think only Silicon Valley giants are bleeding, you're naive. The talent war in China is just as brutal. DeepSeek's funding terms are circulating, and they blew my mind. Founder Liang Wenfeng personally threw in a full $2 billion, with ironclad rules: investors (except state-owned) get no voting rights, and poaching even one researcher is strictly forbidden. Alibaba wanted in? Sorry, you're out!
It's like Liang Wenfeng, alone with a machine gun, screaming at capital: "Leave the money, get lost, and if you touch my core team, I'll destroy you!" Why so ruthless? Because in this zero-sum game, top talent is the lifeline.
Don't let DeepSeek's current success fool you — they're walking a razor-thin tightrope. If a competitor doubles the offer and poaches your core team, the whole system collapses. That $2 billion isn't for expansion; it's a Great Wall of flesh and blood against poaching! If you're not tough today, you'll be a sitting duck tomorrow, devoured by capital sharks.
Codex, the Great Benefactor, or a Free Labor Scam?
At the end of the day, what are the big players fighting for? Absolute dominance in AI coding. LLMs have evolved; pure chat models are worthless. Whoever can deliver production-ready code will make real money.
But I've uncovered a sinister truth. People brag about agents running 7-8 hour long coding tasks. But after burning thousands of dollars in API fees — the cost of a used car — you open your laptop the next day to find it ran in circles, hit a logic dead end, and popped up: "Boss, I'm stuck." You're left cleaning up the mess.
Enter Codex, the year's biggest "benefactor." It's compatible with major open-source models and even records operation steps. You think it's lightening your load? Naive! These AI giants used to spend tens of millions hiring African workers for embodied AI training. Now they've outsourced it for free to millions of users who think they're getting a deal! We happily feed them premium operational data. The highest form of business warfare is making you willingly work for free — getting played and counting the money!
No Vision, No Closure? The Ultimate Multimodal Kill Shot
Even funnier: everyone's hyping multimodal models, but without strong vision capabilities, how can you achieve code closure? It's absurd! It's like going to a brothel, paying a fortune, and being told, "Sir, drinks only, no touching!" What kind of fun is that? Isn't that a scam?
You ask an AI to write front-end code, but if it can't see what the rendered page looks like, how can it match the design pixel-perfectly? Those single-modal models without visual verification are fundamentally crippled.
It's like a street performer doing somersaults in the cold, smashing rocks on his chest, sweating buckets, with the crowd cheering. Then he pats his pocket — damn, forgot the money bowl! If you're not grounded, if you haven't integrated the toolchain with visual verification, no matter how fancy your code, it's worthless garbage.
That's the brutal truth of the current battlefield: only Codex, the silent "benefactor" that's playing the long game, will end up with a bowl full of coins.
So, do you see it now? The AI battlefield for the next five years isn't a warm tech revolution — it's a naked zero-sum game. From Zhipu riding the wave, to Google's strategic collapse, to DeepSeek's $2 billion all-in gamble, and Codex's cunning free labor scheme — no one is safe. At this trillion-dollar table, you either hold talent and compute to corner your rivals, or you become free fuel for the giants. Don't harbor any illusions about "tech for good." Under capital's butcher knife, only the fast survive in this bloody jungle.
Think about it: we stare at model leaderboards until our eyes are bloodshot, worrying like we're buying drugs. But even if Zhipu hits ten trillion, I don't own a single share. What does this insane wealth have to do with me?
Meeting adjourned. Tomorrow, I'm back to working for AI.
[Subscribe to JustSayAI Daily Brief]
🌈 New friends! Subscribe to the JustSayAI Daily Brief · Two emails daily · Listen with one tap
🌟 Old friends! Join as a member to get the JustSayAI Daily Brief + in-depth reports + text columns
[Host's Xiaohongshu]
📕Call Me Xiao Su
[Follow Us]
▶️YouTube ｜ 📺Bilibili ｜ 📕Xiaohongshu ｜ 🎵Douyin ｜ 📻Podcast
📱 Business cooperation/listener group: justsayai666