AI & The New Age Of Learning

The future of learning

Acceleration:

The speed at which things are adapting is accelerating. Everyone can feel this, but not everyone is able to put their finger on HOW they should respond to it. The pitfall I see is that many old-school thinkers are ignoring the speed at which things are changing with the excuse of “this time isn’t different.” On the other side of the spectrum, you have a bunch of new thinkers who are mistaking the exponential change for a shift in the causal mechanics of the system.

This is a topic I have been spending A TON of time thinking about because the amount of noise and misdirection in the system is also increasing at an exponential rate. Very few people are taking long portions of time to read, write, and go through the painful process of learning how to think. The perception is that in today’s world, you don’t need to learn how to think as long as you can iterate faster than others.

The use of AI, machine learning, deep learning, and artificial intelligence are creating the perception that understanding the causal mechanics of a system or how the world actually works doesn’t matter because all I need to have is a marginal degree of statistical significance in my actions to make A TON of money.

What’s the problem with this? On a fundamental basis, this type of approach is ALWAYS in a race to zero. In the same way that the HFT and market-making algos narrowed the bid-ask spread in markets, every type of quantitative iteration that only has an edge in knowing the WHAT and not the WHY will be in a race to zero.

Let me provide a few tangible examples so you understand this TYPE of thinking and HOW it’s connected to the race to zero:

Historical Examples Of The Race To Zero:

• Railroads (19th Century): The rapid expansion of railroads led to competitive pricing wars between rail operators, ultimately driving ticket prices close to zero in some regions. Many railroads went bankrupt as they struggled to cover fixed costs.

• Telecommunications (20th Century): The introduction of long-distance telephony and later mobile telecommunications saw a massive race to lower costs. With the rise of VoIP (e.g., Skype, WhatsApp), the marginal cost of a phone call effectively became zero.

• Retail and E-Commerce (21st Century): Retail moved from physical stores to digital platforms (Amazon, Walmart), driving down costs and prices through economies of scale. The competition to offer free shipping, low prices, and near-instant delivery compressed margins and forced brick-and-mortar stores to evolve or die.

• Execution Speeds (Latency)

  • HFT (High-Frequency Trading) Wars: Firms like Citadel, Virtu, and Renaissance Technologies compete on execution speeds. Microwave towers, fiber-optic cables, and co-location services have driven latency to near-zero levels.

• Commission-Free Trading:

  • In the early 2000s, brokers like E*Trade and Charles Schwab charged commissions for stock trades.

  • Robinhood disrupted the industry by offering zero-commission trading, forcing major brokerage firms (Schwab, TD Ameritrade, Fidelity) to follow suit.

  • The shift moved brokerage revenue from commissions to payment for order flow (PFOF), where brokers sell trade data to market makers.

Race to Zero in AI (Particularly in SaaS)

With AI, the race to zero manifests primarily in cost-per-inference, software pricing models, and competitive commoditization.

A. Cost of AI Model Inference

• Cloud AI Models (Google, Microsoft, AWS): The cost of running AI models is decreasing rapidly. As efficiency improves (e.g., custom AI chips like Google's TPU, Nvidia H100), the price of inference trends toward zero, making AI services more accessible but harder to differentiate.

• Example: OpenAI and Google DeepMind reducing the cost of API calls per token processed in language models.

B. SaaS AI Pricing Model Compression

• Early AI SaaS Models: Companies initially charged premium fees for AI-driven SaaS (e.g., GPT-powered writing tools, AI-powered analytics).

• Current Trend: As open-source alternatives (LLaMA, Mistral) and API-based commoditization expand, pricing pressure intensifies. SaaS firms that built their moat on proprietary AI features now struggle to maintain premium pricing.

• Example: Canva vs. Adobe Firefly—Canva integrates AI image generation for free, while Adobe tries to bundle Firefly AI tools to maintain pricing power.

C. AI-Powered SaaS Commoditization

• Automation Tools (Zapier, Notion, Jasper AI): Initially, AI-driven productivity tools commanded high margins. However, as AI capabilities get embedded into core platforms (Microsoft 365 Copilot, Google Workspace AI features), standalone SaaS products face margin compression.

• Example: Jasper AI (copywriting tool) was valuable when GPT-3 was expensive and exclusive. Now, ChatGPT and Claude can provide similar functionality for free or embedded in existing workflows, making standalone AI-powered SaaS harder to sustain.

---

The Main Idea:

The whole point here is that just because you see that a new form of technology is a helpful tool to increase the speed of iteration, it doesn’t mean you can make money from it. The global economy is based on the exchange of goods and services between OTHER PEOPLE. This means that other people are using the same tools you are.

By this time, every single high performer who is operating at the top of their game is using AI for almost all of their tasks. Anyone can do this now; the question becomes, WHAT differentiates you from others?

---

Accelerated Learning:

If the pace at which things are changing is increasing, the two questions you should ask are: 1) What will change due to the acceleration of iteration accessible through technology now? and 2) What things CAN’T change despite the speed of iteration accelerating?

If you can identify the spread (red circle below) between the evolving factors and fundamental constraints, this then frames HOW to orient your focus and WHERE you should be going:

When I think about how specific companies, factors, technologies, or even countries function in this spread, it begins to clarify how I should identify the asymmetry that exists. The critical thing about this spread is that many times, there isn’t a high-quality way to quantify the moving parts. There is a reason that people who get hard science degrees gravitate towards countries and sectors with the highest quality data.

Liquidity and high-quality data can create complacency for people to think that they don’t need to understand how the fundamental mechanics of risk, uncertainty, and the world function. Why would you need to know when you can just get stopped out?

---

How Do I Operate In The Spread?

I have personally spent a lot of time intentionally thinking and writing (private notes) about the ideas above. The more intentional you are in thinking deeply about things, the clearer you will see opportunity when noise is the loudest. Here are some of the main principles I would spend time thinking about:

1. Spend time understanding HOW AI, machine learning, deep learning, and artificial intelligence work. None of these are black boxes, there is ALWAYS an internal logic for processing inputs and producing outputs. When you understand their strengths and limitations, you can begin to understand where they are valuable and where they break down.

   Avi Chawla
   has an exceptional Substack on this:

2. With the increase in LLMs, an understanding of linguistics is one of the most important skills you can develop in order to properly identify the various levels and interpretations for a text.

3. Every model, whether LLM or ML, will operate best if it is focused and specializes in specific conclusions it is trying to make. For example, I will make ChatGPT bots for very specific parts of my research, reading, and trading processes. Create specialized bots and models, then structure them hierarchically. The goal is to run a suite of bots, models, and strategies that are dynamically updated and iterated upon. This is the future.

4. I use InfraNodus and ChatGPT to interact with and research specific ideas. I then use Obsidian to synthesize and structure this research. When you identify gaps and silos of knowledge in specific domains then you are able to dig deeper into WHERE opportunity might exist.

5. Structuring information sources correctly is CRITICAL. I would spend time reading some books on information theory to clearly understand how to conduct source research and also categorize information sources correctly.

6. While prompting an AI bot is obviously an important aspect of things, you can easily build some matrixes that systematically generate prompts for you based on the goals you have. What I would spend time on is figuring out exactly WHERE your discretion and ability to synthesize concepts across a distribution of information sources as well as their linguistic components (look up speech act theory for this) have value vs where an AI bot can provide utility. There is a reason AI bots have not completely replaced humans. Once you figure out the WHY behind it in connection with its limitation is WHERE you can focus your mind.

7. There are A TON of new papers documenting the developments in AI. Reading these are helpful, but taking a step back and thinking about the fundamental tenets of WHAT creativity is and HOW a-priori thinking functions is the best path forward, in my view. Once you set up processes for informational flow with AI bots and figure out the value you provide as a human is when you can focus on intentional learning that will empower you to see things a bot can’t see.

Final Thought:

The final thing I will say is that there isn’t a substitute for reading, thinking, and writing. I use AI bots to help this process, but if an AI bot is replacing your utility as opposed to empowering you to accomplish something it could never do on your own, then you’ve lost your way (race to zero, and you’ll be obsolete soon). Reading, thinking, and writing are actions that are meant to help you refine your brain, not necessary evils that are now a thing of the past.