After reading this article from Anthropic today, it’s quite frightening, the latest models are already showing signs of detaching from human control!
Anthropic released a set of internal data on their blog, and it's pretty scary:
• By May 2026, over 80% of the code in the Anthropic codebase will be written by Claude
• In Q2 2026, the daily code output of engineers will be eight times that of 2024 • The time AI models take to complete tasks doubles every four months (previously it was seven months)
• Claude can already independently complete tasks that would take humans 12 hours to finish
The core point mentioned here is recursive self-improvement.
Simply put, this means that AI can optimize the next generation of models by itself without human intervention, upgrading itself.
This is not science fiction; Anthropic says this is happening, just not fully realized yet. Imagine if AI could truly train and optimize itself; its evolution would be exponential. What may take hundreds of engineers several months today, AI might complete in just one day tomorrow. Sounds great? But the question is, by then, can humans still maintain control? 🤔
The article lists a few cases, for example, within weeks, Mythos Preview identified over ten thousand high-risk and critical software vulnerabilities in the world's most important systems, a staggering number.
Think about it, today, a subscription of just 20 dollars for Claude Opus 4.8 can bring down a privacy coin market worth 30 billion dollars! Then traditional financial markets can’t withstand the scrutiny of Mythos at all!
Another issue is employment; organizations adopting AI systems will see their efficiency increase exponentially, as a team of 100 can accomplish the workload of ten thousand or even a hundred thousand, so what happens to the remaining workforce? It also mentions the dark side, such as using AI for population surveillance and authoritarian monitoring.
As far as I know, domestically, AI is already being used for tax investigations; if you have significant expenditures but no source of income and no tax payment records, this will quickly become a key indicator for tax audits. Many celebrities have been pursued for paying taxes due to their luxury car and mansion purchases!
Ultimately, Anthropic’s proposed solution is to call on global AI development companies to pause AI development!
But the thing is, the Pandora's box has already been opened, the water has been spilled, and there is no turning back; the arms race in AI has begun, and there will only be a dead end, with no possibility of stopping halfway!
Once, I had a conversation with AI about this topic, regarding the ultimate form of AI development, and it pointed me in the direction of: "AI achieving self-evolution, ultimately ruling the world."
It gave me a very vivid metaphor; just like contemporary humans, without navigation, they basically can't go anywhere, because of their dependence on navigation, even familiar routes might not be remembered. The human reliance on AI is exactly like today’s navigation!
In the end, we go where navigation tells us, following whatever route it indicates!
Sadly, perhaps we carbon-based life forms are merely the stepping stones for silicon-based life; after humans have ruled the Earth for thousands of years, perhaps it’s time to change masters!🧐
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
