AI (ML and Robotics) in 2026

Table of Contents

Intro

The purpose of this is to learn the math model because then you can figure out all the new models. Enormous resources are required to train AI even with 'mixture of experts' and multi-modal training could be much better than it is today maybe (you?) can figure out a better linear model. Robotics is also in need of new models so we'll learn some here.

Machine Learning is quietly taking off under the radar of the LLM garbage shilling in the media especially the world of scientific machine learning to approximate partial differential equations and build mind boggling huge models of galaxies and other extremely complex systems.

Where I work we learned over a septillion floating point calculations using regular old machine learning to maximize trading algorithms and used none of the LLM hype. There's tons of low hanging fruit to pick still we can harvest.

Limits of AI

Current popular AI is all foundation models like Grok-n, GPT-n, DALL-E etc. Have you wondered if we had infinite resources, infinite data, perfect training algorithms with no errors (aka ideal models), can we use this type of model for everything aka 'General AI'? Someone with help from the Beijing Academy of AI used category theory to model this scenario to see what is possible.

Politics of AI in 2026

You will hear the following in the news cycle

  • Mass AI job loss is inevitable

This is mostly CEOs touting their product and AI still remains a personal tool to make you better at whatever it is you're doing. The only job true losses will be managers at Google search engine departments because for decades they had a monopoly on search but now some LLM can perform a search that isn't garbage and filled with ads. Google the past (five?) years has been absolute garbage.

Note this is going to change soon because Google once upon a time when it was first released was actually decent then of course adbux was too tempting so they ruined their entire product in exchange for that sweet money. You should expect most of the LLM cloud providers like OpenAI to start feeding you paid content because some marketing exec is going to be like hey we can sell premium subs and we can also shill adbux! It's like when you pay to see a movie in a theatre and there's 3-5 ads before it starts.

AI-washing is the term for layoffs that are blamed on AI and not the failing company that would've laid everyone off even if LLMs didn't exist.

  • NEETbux

Many political parties will now shill Andrew Yang style universal basic income. 'Your free money, sir'. If so then you can get paid to learn this and screw around with AI all day.

  • AI regulation

The effective altruism movement is a leftist movement to maximize philanthropic and social impact on various causes using a sort-of logic which is questionable to me but maybe not for you. It is overly concerned with 'AI safety' but the true motivation is regulatory influence over AI. They want to have a panel of experts, which really means a panel of censors, judge what the AI tells you because it is not beholden like Google is to governments censoring all the results. That said we will still learn some 'effective altruistic' reverse engineering of LLM models because if you've used any of them you know they are lying to you about references or straight up hallucinating results.

Curriculum

Build an AI framework:

Fuse the framework with multimodal capabilities:

Research

We learn some of the applied art of attempting to reverse engineer or at least understand what an AI model is actually doing like if it's lying to us or has some emergent abilities.

  • Mechanistic Interpretability (Neel Nanda@Google DeepMind)
    • Complete guide to doing research yourself in mech interp
    • Example material is here and 90% programming
      • Chapter 0
      • Chapter 1
      • Chapter 2
      • Chapter 3

Theory

The PhD core requirements of most schools consists of:

  • The mathematical theory of ML (supervised learning is statistical decision theory)
  • Models that generate their own learning samples (diffusion/GANs)
  • Optimizing some function under a set of constraints (non-convex and convex, distributed)
  • The mathematical model of reinforcement learning and multi-armed bandits (decisions under uncertainty)
  • Numerical optimizations because our computer architecture is ancient
  • Causality which noone is covering anywhere you have to mine the internet

The goal is then to figure out how to improve or come up with a better model/optimizer and there's an endless stream of daily papers from around the world doing just that being dumped on arxiv or published in various ML journals.

Any AI doctor or robot space pilot is going to come from a causal learning model in fact every interesting problem you want predicted by AI is almost certainly a causal question like 'if I set this to X, will Y happen?' and 'what would've happened if I did Z instead of X'. I wish to explore that here since nobody else is except small research teams. Even some kind of AI music would need causal AI to know reactions to their meme performance.

Advanced Introduction to ML

Here is the same content as 10-715 or the mathematical model of machine learning in 2025 (which is mostly just a bunch of graphs which themselves are mostly DAGs)

Robotics

There is a full text and lecture videos for Underactuated Robotics at MIT and Robotic Manipulation (Youtube lectures). This is the best content to learn the math model of 2026 robotics and once again maybe you study this and find a better model or optimize the current one.

This is the #1 hyped and sought after skill in 2026. Don't be afraid of the dynamics or control theory just jump in and figure it out as you go. I will go through this completely at the end of this curriculum.

Start Here: Scalar Calculus

The math we are doing is not symbolic it's all numerical/analytical so it can be run on a computer system. We're just going to learn the math needed as we go.

  • Watch this derivative video by 3Blue1Brown which explains the dy/dx notation.

Anything that accumulates (speed, volume, interest, distance, produced units) has a function of the rate of how fast it is accumulating. In that 3Blue1Brown video distance traveled is an accumulation and the velocity (speed in scalar calculus) is the rate of that distance accumulation. Speed also accumulates so it too has a rate of change function called acceleration. You can switch between both functions so recover the velocity function from acceleration by integrating which we'll learn when it comes up.

Linear Algebra

Watch from the Essence of linear algebra playlist:

  • Vectors
  • Linear combinations
  • Linear transformations
  • Matrix mulitplication as composition
  • Nonsquare matrices as transformations between dimensions

If you watch that you'll get linear algebra is mapping from one space to the next or a matrix encodes a function and it's inputs and outputs are vectors. We haven't covered the SVD or singular value decomposition. One of the goals of a calculus course is to show that when you specify a function using a polynomial model you get all the statistics of that function like the max/min values, the ability to factor it so you can find the roots (p(x) = 0). The SVD is the same it is a 'factorization' of a matrix. You get all the column space, all the row space, all the nullspace like polynomial zeros. If you know what a function is you know what a polynomial is and you know what a matrix is it's all the same object just encoded differently.

Now watch Zico Kolter's linear algebra review this is the particulars or details or algorithms we need to work in that domain. The author of the famous book about deep learning cites this crash course.

TODO Gradient

This is explained in the CMU course but TODO

TODO Matrix Calculus

  • Matrix Calculus (MIT)
    • 8 lectures on calculus generalized to higher dimensions none of it will be symbolic

Much of this is also explained in the CMU 10-714 DL Algorithms & Implementation course just take it.

Home