Frequency and Spectrum in Investing | Oasis Signals

Not long ago, we hosted an in-person conversation with 42章经 (42 Chapters), which we recorded as a podcast, sharing the trends we've observed over the past year and the reflections they've sparked. The transcript of the keynote portion has been reorganized and published as a WeChat article. Enjoy.

The full article is 5,892 words — estimated reading time: 15 minutes.

Since the famous paper Attention Is All You Need was published, the Transformer architecture has not only endowed machines with attention for the first time, driving a leap forward in AI technology, but has also advanced human research into attention mechanisms, prompting us to re-examine our own perception and understanding of information from an entirely new perspective.

What's interesting is that artificial intelligence, through continuously evolving attention mechanisms, is becoming more focused, more goal-oriented, and more capable of long-term thinking — while humans, amid the flood of generative content, are growing increasingly lost, fragmented, and disjointed.

The reason for this divergence is that the attention mechanisms of AI and humans are fundamentally different.

The denser the information, the higher the demands on attention, and the greater the energy consumption. Yet AI consumes computing power, which can be scaled through addition — simply add more processing units. Humans consume mental energy, which can only be managed through subtraction: focus and restraint.

The former is easy. The latter is hard.

Especially as the world spins faster, maintaining focus and consolidating attention grows ever more difficult. Our attention mechanisms are gradually failing, even collapsing, causing many to lose their capacity for independent judgment.

The result: the world is becoming increasingly polarized.

This polarization manifests at every level of society — from macro-level geopolitics, to mid-level wealth inequality, down to micro-level marital relationships.

In the past, people could exercise independent judgment in the face of volatility, then reach consensus, and work together to weather storms — a dynamic of "we rise together, we fall together." Now, scenarios of "your loss, my gain" or "my loss, your gain" are becoming more common.

Therefore, it could be said that the root of many problems in today's world lies in our loss of control over attention.

To better illustrate this, let's try to understand attention mechanisms from the perspective of signal processing.

Any signal system can be simplified into a basic structure of "input — processing — output."

Humans are no exception. Our sensory organs — eyes, ears, nose, tongue — function like sensors, transmitting external information to our brains. Once inside, this information is processed and eventually converted into some form of action output. Our outputs can be visible movements, speech, or expressions; or they can be invisible thoughts, judgments, or preferences.

So humans can be further abstracted as a signal system of "perception — processing — action."

But among these three capabilities, there exists a vast chasm.

The human retina receives roughly 10^9 bits of information per second — equivalent to a two-minute 480p video. The brain can process only 10^6 bits per second, about the capacity of a novel. And the actual bandwidth for action output is lower still: merely 10^1 bits, the level of a single instruction.

So our perceptual capacity far exceeds our processing capacity, which in turn far exceeds our output capacity.

What to do?

In signal processing, to address this class of problems, we introduce filters to "denoise" — that is, to extract signals and eliminate noise.

For humans, to solve the problem of information overload, we have evolved an active filter: the attention mechanism. This mechanism helps us ignore vast amounts of irrelevant noise in the torrent of information, and flexibly adjusts as our goals change, ensuring we remain focused on what matters most.

For example, when we focus on a particular person, almost nothing else exists in our vision; surrounding sounds and images naturally fade away. This is the attention filter at work.

A truly focused person can filter out the most important signals in any given situation. These signals are what we call the principal contradiction — the key variables in the evolution of things.

Conversely, someone who lacks focus is easily swept along by information or carried away by others' rhythms, becoming an echo chamber.

This is because when our attention mechanism collapses, we abandon active perception of the world and instead process high-stimulation information, or simply rely on others' behavior to form judgments.

Thus, the phenomenon of "huddling together for warmth" becomes increasingly prevalent. For instance, hot sectors become extraordinarily hot — even when bubbles are clearly forming, crowds still rush in headlong — while cold sectors are completely abandoned. Or a cold sector may suddenly heat up, and a hot sector may suddenly cool down.

This resembles the "Larsen effect" in signal theory, which refers to the screeching feedback that occurs when an input and output port are too close together, causing the signal to loop and amplify.

For example, when you place a microphone near a speaker, the mic picks up the signal and feeds it to the speaker, and the speaker's output is picked up again by the mic — cycling repeatedly, causing the signal to continuously amplify itself, producing that piercing screech.

In truth, the world today is screeching every moment, in every industry.

In such a world, "focus" becomes increasingly valuable.

And "focus" is becoming ever more critical for founders. The reason people feel increasingly mentally exhausted is precisely because scattered attention creates high energy consumption, while excellent founders typically know exactly when and where to concentrate their attention.

When most people are directing their attention toward others and following the crowd, if you can quiet your mind and manage your own attention — focus on perceiving the world — you'll discover that opportunities in this era are actually multiplying, not diminishing.

For example, in 2023, when we began paying attention to embodied intelligence, there wasn't even a dedicated term for it in the market, and there were skeptical voices internally. To verify the market opportunity, we organized a team to conduct field research. The market feedback we received was largely negative. At the time, some friends even told us that robot projects they had invested in back in 2015 still hadn't exited, urging us not to enter this "bottomless pit" (laughs).

But this kind of market feedback actually convinced us that this direction held opportunity. Because many fundamental changes had already been quietly unfolding — they simply hadn't received proper valuation because no one was paying attention.

If you can notice such domains before others do, congratulations — you've found your opportunity.

I know an exceptionally outstanding founder who completed his bachelor's, master's, and PhD at one of the top institutions in his field. After graduation, to many people's surprise, he chose to start a company in an extremely obscure,底层 R&D direction.

I was quite surprised at the time and asked him why he chose this path.

He said: "Because my starting conditions aren't particularly advantaged, I need to find a field that receives no attention, where I don't need to compete with others for resources, but that has the potential for major breakthroughs in 5-10 years. The fewer people paying attention to this direction, the better — because only then will I have the time and opportunity that belong to me."

I pressed further: "But do you know? The direction you've chosen is still subject to policy restrictions — the government has never issued relevant licenses."

He replied: "I know. But precisely because no one is doing it, the technology hasn't broken through, so the government doesn't issue licenses. I'm betting that once I do it, the government will issue licenses sooner or later."

So this founder, with a team of fewer than ten people, buried themselves in R&D for seven full years.

By 2023, the government finally issued the first batch of licenses — only four globally, and his company received one. Moreover, their costs were substantially lower than those of other companies worldwide.

So most human competition, at its core, is a competition of attention.

Whether you're an entrepreneur or investor today, if you want to achieve your desired results, one crucial thing is managing your attention. Where your attention goes, your world manifests there. Because the world is not an objectively existing entity — it is the result of what we perceive and process. In other words, there is no so-called objective world, only the world as perceived by us.

So how exactly do you get results?

It's simple: clarify your needs. A filter is a device whose energy consumption decreases as its target becomes clearer. So the clearer your needs and the more focused your attention, the easier it becomes to fulfill them.

Going further, this means knowing yourself and being yourself.

This sounds simple, but people typically encounter two problems when actually putting it into practice:

First, many people don't actually understand themselves. They simply slap labels on themselves based on others' evaluations or market feedback. They have no clear idea of what they truly like, or they feel that what they like isn't valued highly enough, doesn't fit the conventional definition of success, and therefore isn't sophisticated enough. But what you like is sophisticated, by definition.

Second, many people, even after knowing themselves, don't dare to be themselves. They don't believe that doing so will lead to good outcomes, so they prefer to change themselves, aligning with mainstream narratives and others' strengths.

Returning to AI: as mentioned at the beginning, humans have been using technology to help AI build attention mechanisms, giving it the ability to identify and focus on what matters.

After introducing attention mechanisms, AI's capabilities have indeed leaped forward, and the scope of work it can handle continues to expand. We believe AI will keep breaking through. It's only a matter of time before artificial intelligence surpasses humans in virtually all tasks — and that time won't be long.

Because humans are actually quite limited. This limitation manifests not only in the brain's processing capacity, but also in our ability to perceive the world.

As we know, eyes can see images, ears can hear sounds, skin can sense pressure, temperature, and humidity.

Undoubtedly, humans are multimodal large models.

So, is there a dimension that can unify different modalities of input — images, sounds, scents, temperature?

We can approach this from a different angle. Suppose information is a function. As we shared in High and Low Frequencies in Investing, any function can be decomposed through Fourier transform into the superposition of different sine waves, and each sine wave corresponds to a stable frequency. For example, each string on a guitar corresponds to a frequency, and thus to a piece of sound information. When we apply different pressure to a string, its frequency changes, and so does the sound information.

So frequency can be simply understood as the number of times a sine wave changes per second. For instance, if a heart beats 60 times per minute, its frequency is 1 Hz; after exercise, the heartbeat rises to 120 beats per minute, making the heart's frequency 2 Hz.

Different types of information can all be unified along the dimension of frequency. Visible light, for example, has a frequency of roughly 10^15 Hz, while sound and scent frequencies do not exceed 10^4 Hz.

The range of frequency spans from low to high is wide — we call this range the spectrum.

And different organisms have different perceptual spectra. Dogs, for example, have a wider olfactory spectrum than humans, allowing them to access more scent information.

In fact, perceptual spectra vary considerably even among humans. Some people naturally have wider spectra or finer perceptual granularity — what we call greater sensitivity. So we shouldn't judge others hastily, especially not highly sensitive individuals, because their world may contain experiences we are completely incapable of perceiving.

If we map out common spectra, we can see that the portion humans can perceive is extremely limited.

The horizontal axis in the figure represents the known electromagnetic spectrum; the color-coded portions indicate human perceptual senses and modalities. The latter accounts for only 1/10^9 of the former.

Precisely because of this limitation, we have continuously strived to expand our capacity to perceive and store spectra.

We've invented MRI machines, X-ray machines, weather radar, gravitational wave detectors, and other tools to assist us in sensing information bands far beyond human physiological limits. At the same time, we record information across modalities. For instance, we invented musical notation to store sound information as images, achieving cross-modal information storage.

However, these methods are essentially mappings of information — transforming information from other bands into forms we can perceive. This process involves substantial information loss, and therefore has significant limitations.

Now look at Agent again — doesn't it strike you as terrifyingly powerful?

Because theoretically, an Agent's perceivable spectrum is infinite.

Today's multimodal large models are still in their infancy. Currently we only have language models, speech models, image models, and video models — but in the future, there could very well be tactile models, temperature models, scent models, and other full-spectrum models.

From this perspective, we don't agree with the view that "we're running out of data." Some believe we've exhausted all available text, image, audio, and video data on the internet — but this data merely represents the three modalities that humans can perceive and record. Have we used X-ray data? Gamma rays? Alpha waves in the brain?

The world generates massive amounts of data every day in countless forms — it's just that much of it lies beyond our perception, so we haven't been able to make good use of it, and entrepreneurs haven't even begun to think about how to use it.

Looking further ahead, we believe the future will bring an explosion of sensors. All kinds of sensors will be manufactured to help Agents perceive bands that humans cannot — from gamma rays and X-rays to infrared. Then, Agents will interpret this data through their full-spectrum processing capabilities, integrate information across modalities through language models, and finally act through embodied intelligence.

Ultimately, we believe future AI will be a trinity of "sensors + Agent + embodied intelligence." The world it will be able to perceive is beyond our imagination. And this system should run on blockchain, not the traditional internet. In other words, blockchain will be the internet of Agents in the future.

Based on this conviction, many startup directions today may simply be wrong.

Many projects are still organized around a "human-centric" model. Whether copilots or deep research tools, they all attempt to empower humans through various AI integrations.

But faced with a signal system that may possess full-spectrum perception, processing, and action capabilities, we should be thinking about how to organize work in an "AI-centric" manner.

We shouldn't focus on "how to integrate AI into our workflows." We should invert the question: "How can we integrate into AI's workflow?" and "What can we provide for AI?"

At this point, you may have a question: If Agents know more than me and compute faster than me, will humans be completely useless in the future?

No. Humans have two unique sources of value.

First, creating demand.

If the Industrial Revolution was defined by "whoever creates supply, creates value," then the AI era will be defined by "whoever creates demand, creates value" — because in the AI era, supply will approach infinity. Going forward, we need to know ourselves better, explore ourselves more, understand ourselves more deeply, and clarify what problems we want to solve.

Second, aesthetics.

Aesthetics is, first and foremost, a predictive algorithm we evolved to conserve brain computing power. We typically pre-judge new information, then only compute the "variations" — the parts inconsistent with our predictions — rather than computing the entire information set.

We often say some scientists have excellent technical aesthetic sense, referring to their ability to find the right direction by "feel" even when the technical path is unclear. This capability is actually the result of long-term practice and accumulation.

However, this energy-saving mechanism has side effects. Especially among entrepreneurs, many have a mode of thinking called "wishful thinking" — prettified as "idealism." But in reality, wishful thinking is a form of laziness: when your brain detects "variation," it should recompute, but chooses not to, instead rendering a delusion of conviction. From this angle, "seeking truth from facts" is a form of diligence.

Moreover, aesthetics is not merely an algorithm — it is a processing result that transcends single modalities. For instance, you meet a potential partner; the conversation goes well, their thinking is rigorous, yet something just feels off. This "feeling off" may not come from conscious analysis, but from an integrated synthesis of multimodal information — touch, scent, and more.

If something produces in us a sense of "beauty," it typically signals positive feedback; conversely, it may signal danger. We find smiles beautiful because they usually represent goodwill and acceptance. We instinctively avoid someone with a ferocious appearance because we intuitively perceive their violent tendencies.

Humans developed aesthetics precisely because our perceptual and processing capacities are limited — we need aesthetics to help us pursue benefit and avoid harm. And this aesthetic may transcend multimodal processing results, even transcend binary opposition, becoming some kind of unified mapping.

AI, because of its infinitude, does not possess aesthetics. This means that in the AI era, aesthetics will become the core value that humans can hold.

So everyone should ask themselves: What do you have an aesthetic sensibility for that clearly exceeds others? This can take any form — art, code, mechanical structures. As long as you have above-average judgment and expression of beauty for something, you possess the most scarce capability in the AI era.

Today we've discussed much about human "finitude" and AI's "infinitude." It may seem humans are at a disadvantage facing AI, but in fact, it is precisely human finitude that drove the evolution of unique capabilities — attention, subjective prediction, aesthetics — and gives us the motivation to render infinite richness from our finite lives.

Imagine: if you had unlimited resources and capital, would you still want to start a company? Probably not. You might pat your co-founder on the shoulder and say: "Go have fun, we lack for nothing." Just as when you use cheat codes in a video game, the game suddenly becomes utterly boring.

So in the AI era, we need to do two things well.

First, protect your "mental energy." The strength of your mental energy directly determines your attention performance. If the Industrial Revolution shifted humanity's core value from physical strength to intellect, then we believe AI will shift humanity's core capability from intellect to mental energy.

Second, return to aesthetics. This process of return is essentially a process of removing noise, and also a process of demonstrating individual unique value. We need to redirect our attention back to ourselves, trust our own intuition and judgment, rather than relying on others' information and actions.

Over the years, we've spoken with many entrepreneurs, and ultimately found that truly outstanding founders aren't necessarily smarter or more hardworking than others — but they always choose persistence in the most difficult moments, maintain equanimity and self-restraint in the most glorious moments, and demonstrate courage and resilience in the most agonizing dilemmas.

The above is shared with you all in mutual encouragement. May we all maintain our vitality in the AI era, continuously evolve, and become individuals with aesthetic sensibility and inner stability.

You're also welcome to subscribe to the Oasis-produced podcast "Signal and Noise". In this information-cluttered society, we hope to remove noise and return to purity together with individuals full of vitality.