Apple's iPhone 16 has been released, OpenAI's "Strawberry" model has finally arrived, and Microsoft has unveiled Copilot 2. Indeed, the path for generative AI is becoming increasingly convergent, with everyone moving toward the "correct path":
The task of "compressing" knowledge via Large Language Models (LLMs) is essentially complete. Transformers have proven invincible, moving from text to voice, images, and video...
Reasoning (rather than mere inference) is the critical path from large models to AGI. OpenAI's o1 demonstrates the "broad prospects" of reinforcement learning built atop LLMs.
A consensus has largely been reached: generative AI is a tool. Therefore, Copilot should be positioned as a toolmaker. AI-powered phones should offer more imaginative application scenarios.
As we enter "deep water," there is a growing scarcity of computing power and data. The threshold for large-scale cluster computing power has exceeded $1 billion (and in two years, the threshold may be an order of magnitude higher). As for data, what is needed is massive amounts of "synthetic data"; only those with the best models will have access to the best data.
Consequently, we see a vast number of traditional enterprises launching AI transformation strategies, and more countries viewing this as a new opportunity for "leapfrog" development, despite increasing skepticism: how can this be compared to the "Industrial Revolution"?
In reality, as we move forward, we will become accustomed to it and start questioning the next "new thing."
However, current skepticism must still face the question: has generative AI entered "garbage time"?
Office workers who start their day by processing hundreds of "unread emails" would not think so: they need features like "auto-summary" or "Priority Notification."
Artists and designers constantly seeking inspiration would not think so: more and more people are using generative AI to break through various "possibilities."
Data analysts would not think so: generative AI is a bridge, filling the gap between finite energy and infinite information.
...
Therefore, no enterprise will think so either. As long as competition exists, generative AI remains a top-priority strategy.
At this moment, however, I am in a library, typing on an Asus dual-screen AI PC—a supposed exemplar of hardware innovation (Wintel struggles to make a good AI PC). But thanks to the collective efforts of Windows Update, Asus background processes, and Intel drivers, my computer is running at the speed of the 386-486 era...
I killed a bunch of processes and glanced at my Samsung Fold, which just lost its Wi-Fi functionality two days ago. Upon opening it, the screen suddenly went black. Yes, the familiar folding screen ribbon cable failure is back...
So, this segment was written after I confirmed the Windows system failure and realized my phone could no longer fold; an entire morning was killed.
My reason for liking folding screens is simple: an 8-inch screen is perfect for reading documents anytime, anywhere. My reason for liking dual-screen laptops is the same reason many people like having two monitors at their desks.
Apple is seen as uninspired—no folding, no dual screens, no innovation at all...
And yet, I have a Mac (the so-called "Trash Can") nearly ten years old that still serves smoothly, and a five-year-old iPhone SE that always plays the role of the "super sub" (ready to be promoted from backup at any time). Thus, having been thoroughly "brainwashed" by environmentalism, after seeing the prices of new Samsung and Apple phones, I immediately chose to use my current ones for another year.
After one year, there can be another...
For me, once open-source models reach a passing grade, hardware specs become redundant. Our work and life problems are clearly software problems—and personalized software problems at that. For the last mile, outside help is useless; you must solve it yourself.
Just as everyone knows the Fed will cut interest rates on the 18th, no one dares to judge whether a 25 or 50 basis point cut will lead to a rise or fall. This is because no one knows how others will interpret whether the "rate cut magnitude" meets expectations versus how they will interpret whether the "recession exceeds expectations."
Almost everyone can make accurate judgments about certain future events, yet no one can judge the market. The market absorbs all incremental information in the shortest possible time.
This is the "perfectly efficient market" found in textbooks;
yet it is the market's "garbage time."
How many people can truly align knowledge with action and "focus only on the long term, ignoring the short term"? How many can escape the speed of information explosion and hardware update cycles that have already exceeded human physiological limits?
99% of people have no choice. It isn't "garbage time"; it is simply the self-awareness of corporate drones in a makeshift stage production: "Forget the stars and the sea, just keep your head down and keep moving."