First Principles Thinking In The Age Of AI
Edition #158
Introduction
Welcome to The ContraMind Code.
The ContraMind Code provides you with a system of principles, signals, and ideas to aid you in your pursuit of excellence.
The newsletter shares the source code through quick snapshots for a systems thinking approach to be the best in what you do.
The Code helps you reboot and reimagine your thinking by learning from the best and enables you to draw a blueprint on what it takes to get extraordinary things done. Please share your valuable thoughts and comments and start a conversation here.
Take a journey to www.contraminds.com. Listen and watch some great minds talking to us about their journey of discovery of what went into making them craftsmen of their profession to drive peak performance.
The Growth Maze vs The Idea Maze.
Why the next wave of AI will shift the race towards distribution and growth.
Andrew Chen, General Partner, Andreessen Horowitz, shares his thoughts on how the AI race will evolve and take shape over the next few years. It provides some lovely perspectives on what to expect, learning from the past when the web and mobile started becoming mainstream. It sets the tone for how we must reimagine many businesses and evaluate new ideas and business models that are expected to emerge from this storm.
Here are some points from the article that have the potential to make your mind gallop with breakthrough thoughts and ideas:
A good idea means having a bird's-eye view of the idea maze, understanding all the permutations of the idea and the branching of the decision trees, and gaming things out to the end of each scenario.
How can you evaluate the zillions of ideas you are expected to come across? Try to understand what the founder’s mastery of their corner of the idea maze is. It’s often the strongest sign that someone is an expert in their space, and they will teach you more in an hour than you might learn in months.
The world is constantly changing, and sometimes it takes a beginner’s mind to try something once again in a new era. Previous Idea Mazes that had been marked solved, like search engines, CRM, or email clients, might be reinvented as new doors are unlocked by AI.
Solving the Growth Maze might require precisely timed choices, jumping from one part to another, and a good dose of luck.
In an Idea Maze, you traverse by adding and removing features, but in the Growth Maze you traverse by iterating on new go-to-market initiatives and making decisions about potential starting points, who will you serve dilemmas, timing, dead-ends, shortcuts, hidden paths, trapdoors, continuously understanding the context, etc.
Growth Maze might be very different in the pre-AI age versus in the coming waves of AI innovation.
Read the article here.
The Right(and wrong) Way To Spend Money In Your Startup.
Many startups fail not only because they lack a product-market fit but also because they don’t spend the money they raise prudently. This Y Combinator episode provides valuable insights on managing the money they raise and how to avoid making some mistakes others have made.
Here are some powerful insights that can make you think not only if you are a startup founder but also if you are building a new unit, business, or product in a large company or enterprise too. The lessons and learnings apply across both:
Some of the questions you want answers to, when it comes to money, are:
When is a good time to spend money?
When is a bad time to spend money?
Is there a way to think about spending money?
What are the great things to spend money on?
What are the bad things to spend money on?
Lesson #1: In the starting-up phase, spend money only on a few critical people who can make a difference to make the startup idea come alive. Sometimes, you don't really need full-time roles. Learn to work with part-timers till the concept comes to some shape. The more people, the slower you get to product-market fit.
Lesson #2: Don’t hire just because you don’t like to do something or are not good at it. You have to learn the things.
Lesson #3: No one can sell the product like the founder. Sales and marketing are the last jobs to hire for.
Lesson #4: How can you be disciplined and frugal about the money in the bank? Have two bank accounts, and you put half of the money in the other one, pretend like you don't have it.
Lesson #5: When startups get funded, every dollar and energy should be spent on finding a product-market fit.
You can also listen to the entire episode on:
OpenAI VP On Competing With Deepseek, How ChatGPT ‘Reasons’ and More.
In a conversation with Belle Lin of WSJ, Srinivas Narayanan, OpenAI's VP of Engineering, discussed the use cases for reasoning models and AI agents in the enterprise and the engineering challenges faced by OpenAI.
Here are some key takeaways:
AI is going to be the most significant technological transformation of our lifetimes. Open AI launched o1 and 03 mini, and they are continuously refining the ‘Chain of Thought Summariser’, which is playing a significant role in how AI models will be used to solve complex problems worldwide, such as tax analysis, strategizing, researching an industry or the financial health of a company, assistive coding, etc.
Oscar Health is using the Open AI stack to understand patient outcomes. The Open AI stack has a base reasoning model, on top of which are built sophisticated and specialised AI models, such as operator and deep research.
The Biosciences industry uses reasoning models to better understand clinical trials and determine which drugs to use for drug discovery. Or Berkeley National Lab, which studies mutated genes and the symptoms of some rare diseases.
There are many tasks that would take humans several hours to get done, but Open AI’s Deep Research models can now automate these tasks or search. For example, think of a question like ‘Tell me how the retail industry has changed in the last decade.’
Open AI’s model, Operator is another reasoning model that can automate your tasks within a browser’s context. For example, you can take its help to book a restaurant. It knows your context, then figures out the best restaurants you may like, and books them for you.
Highly capable models are going to get cheaper over time. Distillation is a trend where you build a smaller model that learns from a larger one. Scaling laws will be applied to the training and inference sides.
The key skill you need to build in an agentic future is ‘Take a problem, figure out how you want to think about it, and find a way to evaluate it. If you can express your job to be done well, AI will help you do that better and better through reinforcement learning.’
You can watch the entire video by clicking on the above link.
First Principles Thinking In The Age Of AI
It’s only once in 100 -150 years that people get a chance to be there when revolutionary changes occur. Once the fundamental foundations are created, the next generation of people will work on this foundation to develop products, services and solutions. The conversation with Srinivas Narayan and the article by Andrew Chen triggered some of these thoughts and ideas.
Imagine living a hundred years back when electricity, rail, and oil were not invented or discovered yet. Some of the inventors or creators of those times had to ask first principles questions that others around them were not thinking or even imagining the possibilities. For example:
“Why does it take me several days, months, or years to reach the destination that I want to go to? Can’t I make it there faster?"
“Why do we need to live in the dark after sunset? Can’t we have the same level of brightness and light even after sunset?. Lighting a fire is not good enough.”
“Human beings or animals have only a limited energy source, how do we create an infinite source of energy that can be used?”
“How can we build different machines that make products faster than humans?”
Many of the innovations happened by asking questions from first principles. After that, several decades of incremental innovation happen on top of them. Think about these questions for a minute:
‘What is the next big innovation in travel since building an aeroplane to fly across continents? Why should we travel 24 hours to reach the US from India and vice versa?’
‘What is the next significant innovation that has happened in personal transport since cycles or cars were built?’
‘What causes specific human diseases and how can we proactively prevent them?’
You get to be part of such changes only once in a generation. AI is ushering that change. The industrial age automated ‘demanding manual tasks’ that humans manually performed at that time, shrinking time and expanding product availability. The AI age is trying to automate ‘mental tasks’ that humans have been used to doing over the last century. If the industrial age automated manual labor and work for scale and built an economy of ‘mass production of goods’, the AI age is about working on automation knowledge for ‘ mass production of knowledge’.
Therefore, in the industrial age, the cost of goods or products decreased rapidly over time. In the AI era, the cost of knowledge, which is at a premium today, is getting democratised and cheaper. If knowledge professions like computer programmers, lawyers, accountants, consultants, architects, engineers are expensive today, AI will disrupt some of these professions. Generic knowledge will no longer be the differentiator, specialised knowledge will have a premium. It’s like an operator agent sitting on the topic of a generic AI reasoning model.
Data is becoming a natural resource, such as water, sun, air, metals, etc. The ones who distilled ideas like electricity, transport, machines, IC chips, etc. out of them made money. What you distill and automate with data, will be the next source of professional and business competitive advantage.
When you look at professions in the pre-industrial and post-industrial age, say shoemakers, toolmakers, artists, the demand for shoes or tools did not decrease. The ones who added ‘inventive concepts and design’ to ‘generic production’ methods did well. So, new opportunities will emerge when you add ‘inventive concepts and design’ to ‘generic knowledge’ production.
So, in the age of AI, knowledge will no longer be power, but scaled ‘empowering knowledge’ will become powerful.
Some of the lessons we learnt from this week’s mission:
The ones who can envision and crack the growth maze will be the winners as AI becomes mainstream.
Most startups fail as they don’t prudently spend the money they raise at different phases of their evolution and growth.
If the industrial age shrunk time and availability of goods, the AI age will shrink time and accessibility of knowledge.