The difference generated by the action is the coordinate.
I saw a NASA news story that was both interesting and inspiring, so I wanted to share it with you today.
In simple terms, NASA has a mission called Starling, which recently released the results of a flight experiment. They developed a FALCON system that enables a small satellite to calculate its own location without using GPS at all.
We've all seen those space sci-fi movies, where you can imagine a lone satellite floating in the vastness of space. In the past, for this satellite to know its own location, it never relied on its own calculations, but rather on external navigation information.
In low Earth orbit, the most common method is GPS; a more traditional approach relies on ground stations for tracking. What NASA did this time was to remove the real-time navigation input from GPS, allowing the satellite to navigate on its own through autonomous observation and calculation.
To be honest, it's not just about navigating in space, even in a city, I habitually use navigation for a road that has been open for a year.
Even if you're familiar with a route, having to navigate it multiple times can sometimes trigger inexplicable panic, leaving you wondering: "Is this the right turn? Why does it look off?"
I'd like to say at the outset that, in my view, this is not just a space-related news story, it will ultimately have an impact on things that we all experience.
Space Junk Could Become Road Signs
As for small satellites, they rely on vision to navigate in the vast space.
The satellite had something called a star sensor, which can be roughly described as a camera that looks up at the sky. Its original mission was to identify stars and determine its own orientation.
But this time, NASA had it look at other things, such as other spacecraft floating by, as well as orbital debris, also known as space junk.
An object passes through its field of vision, and the small satellite recognizes what it is and knows its approximate location, so by backtracking, it can also determine its own location.
It's like navigating a wilderness without road signs, where the only landmarks you can use are the large rocks in front of you and the crooked-necked tree that has been dead for decades in the distance.
Of course, space debris is a bit different from those rocks, as it's not stationary, but rather orbiting the Earth at high speed. However, NASA has a catalog that keeps track of what they are and where they are predicted to be at any given time.
So, the small satellite sees it, and then matches it with this catalog, allowing it to reverse-engineer its own location.
For this satellite, even space debris can be turned into a landmark. People have been complaining about space debris for decades, but this time, its despised attribute has actually come in handy.
It's just there anyway, and can't be cleared away for the time being.
Simulation vs Live Observation
Up to this point, you might think this is quite romantic. I initially stopped here too, but what really made me sit up straight was the second thing that the FALCON system did.
The ground station uploaded the catalog to the small satellite: 20,000 space objects, each with a predicted orbit, from the official US public catalog of space objects.
This is its textbook, equivalent to the most professional people on the ground telling it: all the answers are here, just follow it to find them.
When the FALCON system's live camera feed was checked against the reference document, discrepancies emerged.
In three days, it autonomously corrected the orbit data of over 200 objects without any intervention from ground personnel. Moreover, for the objects it had directly observed, the accuracy after correction was higher than the catalog provided by the ground station.
This is equivalent to when a teacher has just handed out the exam papers, and the class genius glances at them, then calmly raises their hand and says, "Teacher, this question is wrong." The whole class erupts in surprise, and the teacher looks at the paper in embarrassment, only to find that the question indeed contains an error.
I won't ask why I was so deeply moved, but it was for this reason that I decided to go to Beijing to be an internet worker rather than a teacher.
But back to this news story, if you think about the process again, you will find it terrifying.
In theory, the ground station is the center, the authority, the place with the most computing power and the most experts. It has a large amount of historical data and orbital models in its hands, and its role is to tell the satellite: where you are.
As a result, this small satellite, which came equipped with a camera and a bit of edge computing power, discovered that the data provided by the center was incorrect after conducting its own analysis. It then revised the directory in its possession while still in orbit.
The most interesting aspect of this matter lies right here.
The catalog was predicted using a large amount of known parameters to extrapolate forward, and the logic can be said to be very rigorous. In contrast, the camera on the small satellite simply observes, and what it sees is what it records.
One involves extrapolating from the past, while the other involves observing what has just happened in reality. When the two collide, observation wins.
Isn't this just what we always say: "what is heard is false, what is seen is true"?
The Entire AI Industry is Lost in Space
Speaking of which, I have to mention the name FALCON.
It is actually an abbreviation, with the full English name being Fast Autonomous Lost-in-space Catalog-based Optical Navigation.
There's a phrase here called "Lost-in-space." A classic science fiction work shares this exact name, translated into Chinese as 《迷失太空》.
However, the name of this system is not romantic, nor is it meant to pay homage to classic films.
In space exploration, it originally referred to a specific and troublesome situation: equipment has just been activated, with no prior information about its location or orientation, and can only re-identify itself based on what is visible in its field of vision.
See, isn't this exactly what FALCON is doing?
Without GPS directly telling it its coordinates, it looks up, observes its surroundings, and then uses what it sees to compare with its database, gradually calculating: oh, I'm probably around here.
But what I want to say is, when I see this term, the first thing that comes to my mind is not a satellite, but rather all the people currently working on AI.
The 88% and 1% Divide
Two days ago, I read a report from Tencent Research Institute about super individuals, which began by citing two numbers.
An annual AI Index from Stanford HAI found that 88% of global organizations have applied AI in at least one business function, while a separate global survey by McKinsey noted that only 1% of companies believe they have truly achieved AI maturity.
Look at these two numbers, 88% and 1%, how striking they are when placed together.
The report refers to this as a huge gap. As for "having implemented" it, in many cases this means simply purchasing an account, establishing a committee, or drafting a policy document, with two or three people in the department figuring things out on their own, while the organization as a whole remains unchanged.
To be honest, this is completely consistent with what I've seen.
I went to Silicon Valley and met with some of the companies that appear to be the most radical from the outside. Many people think they have a clear roadmap in hand, but after talking to them, they say: We're also still testing and don't know if tomorrow's model iteration will overturn everything.
So, what is the current state of the entire industry?
Everyone has started up, everyone is moving, but no one has a GPS in hand, they are all lost in space.
Why the Old Roadmap Failed
At this point, I know some of you may start to lose focus.
You may think that you're not an entrepreneur and don't need to think about strategy every day. You're not the boss, and you're not the one who decides which direction the company will take at the beginning of each year.
I'm just an ordinary person working in a company, and my work style is as follows: at the beginning of the year, I set goals, break them down into KPIs, write reports every week and every month, and review my progress quarterly, then execute accordingly.
You are waiting for a clear plan, waiting for the top to determine the direction, allocate resources, and pave the way, and then you will be responsible for moving quickly and accurately.
I completely understand this feeling, as it has been repeatedly proven to be an effective way of working over the past twenty years. In the past twenty years, the pace of environmental change has been relatively predictable, and the plans made at the beginning of the year have generally remained on track by the end of the year.
The central authority indeed has a broader view than the periphery, and headquarters indeed has more comprehensive information than the front lines. The directory issued from top to bottom is indeed more accurate than what you can see by looking up. Following it will be the most efficient, and you won't have to take responsibility.
So in the past, you would wait for a plan or wait for instructions from your superiors, and there would be no problem at all.
The problem is with the present. The "catalog" currently in use is no longer accurate.
Because all previous plans were extrapolated based on known conditions, moving forward with what was already known. However, the current known conditions, such as model capabilities, costs, and tool forms, are changing every three months.
The December you envisioned in January may no longer be the same as the actual December. Consider the lobster restaurant OpenClaw, which was all the rage in March this year - it's been less than half a year and it already feels like a distant memory.
Give me a clue, no matter how vague
What can be done then?
FALCON's answer is quite straightforward: use whatever you see. Even space debris can be used as long as it can provide positioning functionality.
For us, it's about taking action first. We build an AI tool, conduct a process transformation experiment, and create a rough but working prototype.
It's probably useless, probably broken after three months, and probably your colleagues will take one look and say, "What's this?" Fine, then you've got a piece of junk.
But this garbage has coordinates.
Trash can serve as a landmark not because it is useful in itself, but because it has a fixed location and can be recognized by a camera.
However, I must make one thing clear first, otherwise this conversation may become overly simplistic and give everyone the impression that doing something perfunctory is enough.
It should be noted that the FALCON system is not just looking blindly, it is looking with a directory in hand. Therefore, every glance can produce a reading: I thought you were here, but you are actually there, and the difference is this much.
This difference is what it actually gets.
If you ask AI to create a "Hei Sheng Hua: Wukong" and what comes out is not trash, but noise, then you have no expectations for it, and it's unclear what it will turn out like. In that case, it's impossible to discuss how it deviates from anything or what it tells you.
If you do this kind of thing 200 times or 500 times, your capabilities will still be stuck in the same place.
So, before you actually get started, write yourself a catalog, in one sentence: what do I think AI can do for me.
After you finish, run the results against the baseline. The discrepancy—the part that doesn't line up—is the coordinate you've picked up. That's also why this should sit on the small tasks you genuinely repeat every week, small enough that when it fails, you can pinpoint exactly which step it broke at.
From Super Individuals to Super Teams
Speaking of this, there's a follow-up to the news that I think is the most interesting aspect of the whole thing.
Starling itself is a constellation of four satellites, and the aforementioned experiment mainly involves one satellite observing and calculating on its own. However, NASA plans to take it a step further by having the four satellites share what each of them sees.
If FALCON can operate in coordination on multiple satellites, with four satellites each observing and sharing what they see, and using the combined observations to jointly correct each satellite's position, what changes will occur?
A satellite that observes itself, determines its own position, and even dares to alter ground station data—that is a super individual.
Four people laying out what they each see on the table, correcting each other, and becoming more accurate together, that's a super team.
The report from Tencent Research Institute also listed "influence spillover" as a necessary characteristic of a super individual, stating that
If someone uses AI to increase their output by tenfold without their colleagues noticing, they have not yet become a super individual, but rather an excellent employee who knows how to use AI.
At the time, I thought this statement was correct, but I didn't understand why it was necessary, and the satellite gave me the answer.
If a satellite's observations are not shared, they are merely its own private experience. It alone becomes more accurate, while the other three continue to use the outdated catalog. Only by making its observations public can it transform from a single entity's perspective to a public landmark for the entire constellation.
So, being visible is not about giving face or motivating employees, it is a prerequisite for whether the entire system can be standardized.
It's like when someone in your company develops an AI tool on their own and boosts their work efficiency by three times - if no one else knows about it, the organization hasn't learned anything from the experience.
Doubao Experience Written Back to Same Directory
Moreover, looking back at FALCON's three days, the most valuable aspect was not the correction of over 200 objects' orbits, but rather that each correction was written back to the same catalog, making the catalog in their hands more accurate.
In other words, the 200 demos you casually asked an AI to generate—each one isolated, with nothing written back to any system—don't amount to having done something 200 times. They amount to having done 200 first times.
If you have an AI-related matter at hand, and you've been waiting for a more mature opportunity, a more complete plan, or a clear direction from the top, my suggestion is:
This week, take action and solve the small issue in your own hands, don't try to change others' processes right away.
After completing the task, write down the learned sentence and accumulate it in the same place, then let people see it. Note that what people see is not the tool, but the pit you touched.
Even if what you make turns out to be garbage, that's fine—garbage is still a coordinate on your path forward.