Technology happened quickly. During this time two years ago, few people heard ChatGPT, and few investors knew what NVIDIA did. On January 6, 2023, I wrote a note entitled “TALOS”, and I said …
“The most recent development is the arrival of ChatGPT. This is an interactive” smart “robot developed by OpenAI (the establishment of a social constructive AI for the establishment of a social constructive $ 30 billion seven years ago). Students found that it could write half of the article. “The problem responded, and as a result, my clumsy opinion was a good response to the” two aid economists “that may give ('on the other …”). I think I can do it almost better. If there are any lessons, it can make human writers more self -righteous, weird or stylish, driven by their writing. I haven't withdrawn from work yet.
I'm still writing, don't use ChatGPT. In January 2023, my assumption was that because ChatGPT had already existed for some time, investors knew this and incorporated it into the stock price (the financial effective market theory said that all public available information quickly quickly and available. Reflected in market prices). This is not the case. It seems that afterwards, I realized that students in economics and finance in the 1990s spent too much time to absorb the “high -efficiency market assumptions”. This was a major event at the time. Era. I suspect that more people today think that the market is effective, not effective, and vampire squid type funds and central banks have made some culprits.
The performance of AI -centric stocks (especially NVIDIA) shows that AI's early commitments have been underestimated and may be underestimated today. Indeed, the concentration of the top ten companies in the US stock market (40 % of the market value) is as high as 1930. In this case, Deepseek obviously replaced the ChatGPT App Store in the Apple as a real examination.
This is also an effective market inspection. I have seen the original model test results of Deepseek early this year (Azeem Azhar's blog mentioned it as early as November, and it was mentioned as early as December 2023). To this end, the technology community has been conscious of DeepSeek, so this is a surprising thing, and for the efficient market assumption, another shooting face (I am still obsessed with this).
Regardless of the valid market, the appearance of DeepSeek reminds people how to operate the innovative cycle in the economy, and some commentators hurriedly shot for their Schumpeter. Austrian economists created the term “creative destruction”, but his view of capitalism is worth reading because he believes that it will destroy himself (different associates or companies will effectively prevent the system) and collapse. His work in the business cycle is also overdue.
我们不太知道DeepSeek(声称与Chatgpt和Claude AI相比,他们以一小部分成本产生了卓越的模型结果),中国国家是否涉及任何间谍活动,还是他们拥有的程度背靠在美国团队已经完成的Work. The bandwidth of the model is limited in China and is unlikely to be widely used in the West.
The important development is that the production threshold of large AI models is low. Other cheap models are on the way. The Dibao-1.5 or Moonshot Kimi K1.5 of Bontedance is two emerging examples.
To this end, the first stage of AI prosperity is now over. Just like other key technical infrastructure breakthroughs-Railway (railway company (railway company accounted for 50 % of the stock market in 1900). Automobiles (France occupied the automobile market from 1903 to 1929, and sometimes half of the world cars will be manufactured. (Anyone) (Anyone) Remember AOL or NETSCAPE “) -The initial innovation has obtained a lot of capital and is expected to change the world, but it is usually not expected by investors. The heritage is usually a new infrastructure network.
In the next stage of AI, investors are unlikely to pay for model developers. Instead, they focus on unique data sets, AI (Health) applications and AI industrial supply chain applications See last week's “Humphrey” “Humphrey”).
The view that the stock market rebounds quickly is that consumers are the winner-Deepseek opened the cheap prospects, and better AI for joint use. I think that investor dialogue will turn to how to affect the consumer model and how people work (from the police from the police (from the police Go to the teacher). The facts of the stock market's positive view of DeepSeek News show that the AI bubble in the stock still lives well.
In other parts of AI, the model development accelerates in a dangerous manner. In the paper entitled “The Border AI system has surpassed the red line of self -copy”, the scientists of Fudan University highlight how AI builds its own copy. When this process falls into obstacles, the survival instincts are displayed (for example, restart the hardware to repair and repair Hardware to repair and repair hardware errors). To my surprise, this is the real AI development we need to pay attention to.