Anthropic just lost its shit.
GPT-6 Astra dropped and it is not close.
Better at computer use. Better at coding. Better at math. Better at actual work.
The gap is real. The cope will be legendary.
AI subscriptions today feel a bit like gym memberships.
For what you are spending your money today?
Codex, Claude, Gemini?
Anthropic reportedly leaked 512,000 lines of Claude Code source code yesterday, and what followed over the next 12 hours was wild.
At 4 AM, Anthropic pushed an update to npm that accidentally included a 60 MB debugging file containing their entire codebase. Just 23 minutes later, researcher Chaofan Shou spotted it, downloaded the zip, and posted about it on X. Within 6 hours, the post had reached 3 million views. By the time Anthropic’s team woke up, the code had already been forked more than 41,000 times across GitHub. Anthropic began issuing DMCA takedowns, but by then it was too late.
One of the people who jumped on it was Korean developer Sigrid Jin, Claude Code’s biggest power user. According to The Wall Street Journal, he burned through 25 billion tokens last year. After reading the leaked code, he rewrote the entire thing in Python in just 8 hours. That repo reached 30,000 stars faster than any GitHub project in history. Then he rewrote it again in Rust, and that version now has 49,000 stars.
Meanwhile, someone mirrored the code to a decentralized platform with a single message: “will never be taken down.” In other words, the code is permanent, and Anthropic cannot get it back.
The part I cannot stop thinking about is this: Anthropic had built something called “Undercover Mode,” a feature designed specifically to stop Claude from accidentally leaking internal secrets. They shipped an anti-leak system inside their own product, then leaked their own source code in a .map file. The irony is brutal.
The biggest problem with AI in business today isn’t technological. It’s narrative.
Too often, AI is sold as a “partner”, a “personality”, something more than a tool.
But companies do not need a relationship with a chatbot.
They need outcomes.
Less searching.
Less guessing.
Fewer bad decisions.
More accurate answers.
Faster access to knowledge.
More control over information.
AI does not need to be “someone”.
It needs to work well, work safely, and be accountable.
That is exactly why we built PrivGPT.
No fairy tale about emotional bonds with a model.
Just a tool that helps companies reach the right knowledge faster, with permissions, sources, and full auditability.
No magic.
No anthropomorphism.
No pretending that “AI native transformation” will solve organizational problems by itself.
What matters is usage.
What matters is value.
What matters is results.
Our model is simple:
If people use it and it delivers value, you pay.
If they do not, you do not.
That is what AI in business should look like.
Less fairy tale. More accountability.
Dave 🚀 retweeted
Introducing Durable.
The first AI business builder that replaces your 9-5 income.
RT + comment “Durable” and we'll build your business for FREE.
I am using both codex and claude code now on $20 subscription
the main difference feels like in claude code i hit the limits very frequently, but with codex I can code almost all day without hitting the limit
not sure how that’s possible 😂
have you been already trying opencode?
i’ve tried it for couple of hours then switched back to codex
Dave 🚀 retweeted
Replying to @d4m1n
you shouldn’t be sleeping at all Dan sending it all of that to the cloud AI 😄




