Sign in

Mike X Cohen PhD, Sincxpress Education

@mikexcohen.bsky.social
403 followers 189 following 194 posts

Check out my longer posts on Substack (mostly detailed tutorials about data science and LLMs). mikexcohen.substack.com Explore my video-based courses and books ➔ sincxpress.com

PostsRepliesMedia
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 06/09/2026
I nerded out and wrote a Substack post on Givens angles and Jacobi rotations. It's a tutorial post with Python code to demonstrate everything. By the end of the post, you will accidentally discover the eigenvalues of a matrix. It's fun. mikexcohen.substack.com/p/givens-ang...
mikexcohen.substack.com
Givens angle, Jacobi rotations, and eigenvalues
Join me for a meander down linear-algebra lane
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 17/07/2026
Fun movie idea: Time travel exists but you can only travel to the year 6174. mikexcohen.substack.com/p/a-kink-in-...
mikexcohen.substack.com
A kink in the universe at 6174
I’m gonna do something different: I’m not going to talk about AI.
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 15/07/2026
Are you a neuroscientist considering a transition to AI? I wrote some thoughts about it. mikexcohen.substack.com/p/transition...
mikexcohen.substack.com
Transitioning from neuroscience to AI?
Some tips from someone who's done it (sort of)
2104
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 28/01/2026
I'm almost finished my new book, titled "50 ML projects to understand LLMs". I call this figure the "mitten plot."
050
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 19/11/2025
How to live a happy life, according to the most elegant equation in humanity. (Hint: it's the math version of "omit needless words.") mikexcohen.substack.com/p/eulers-ide...
040
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 06/11/2025
In my latest substack post, I show how to measure gender bias in open-source LLMs. The post comes with detailed explanations and a Python notebook file that you can easily adapt to any other biases. Curious? Come check it out :) mikexcohen.substack.com/p/gender-bia...
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 27/10/2025
No comment.
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 22/10/2025
Statistician: Someone who has their head in an oven, their feet in an ice bucket, and says "on average I feel fine."
160
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 17/10/2025
If I had a million euros (that's 1.170686 million USD), I would do Daily Data Challenges to master machine-learning. Like this one, which involves running a t-test in Python to compare two random samples. Act like a millionaire. Solve Daily Data Challenges. mikexcohen.substack.com/p/ddc-15-two...
mikexcohen.substack.com
DDC-15: Two-samples t-test
A data challenge a day helps you master machine learning
030
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 13/10/2025
“King – man + woman = queen” sounds smart until you actually check the math. Spoiler: it’s fake news. mikexcohen.substack.com/p/king-man-woman-queen-is-fake-news
mikexcohen.substack.com
"King - man + woman = queen" is fake news
On analogies and arithmetic inside LLM embeddings spaces.
040
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 09/10/2025
Do straight lines exist? open.substack.com/pub/mikexcoh...
open.substack.com
Do straight lines exist?
Do lines even exist outside our imaginations?
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 08/10/2025
I don't know what's going on in this AI image. But I do know this: Daily Data Challenge #25 is short and very practical for data science. You'll get to create a symmetric matrix and extract its unique elements. Can you solve it in less than 6 minutes? mikexcohen.substack.com/p/ddc-25-uni...
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 07/10/2025
Here’s a picture of Thalorion The Destroyer reading my calculus textbook. I hope he likes it! Otherwise he might unleash doom on humanity. www.amazon.com/dp/B0DX6BPPZ8
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 04/10/2025
Karl Gauss (yeah, *that* Gauss) sent me a fax this morning to say that he really enjoyed working through today's Daily Data Challenge. mikexcohen.substack.com/p/ddc-23-dis...
mikexcohen.substack.com
DDC-23: Distributions of MNIST pixel values
A data challenge a day helps you master machine learning
040
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 02/10/2025
Why I left academia and neuroscience. This post on Substack has gained a lot of traction. I think many people identify with it. (Most of my posts are technical tutorials on machine-learning and LLM-mechanisms.) mikexcohen.substack.com/p/why-i-left...
mikexcohen.substack.com
Why I left academia and neuroscience
Don't worry, this isn't yet another story of rage-quitting.
4327
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 01/10/2025
Daily Data Challenge 8: Test for Pythagorean triplets A data challenge a day helps you master machine learning. More exercises and solutions at mikexcohen.substack.com youtube.com/shorts/LlqS2...
youtube.com
Daily Data Challenge 8: Test for Pythagorean triplets
YouTube video by Mike X Cohen
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 27/09/2025
Bootstrapping: because sometimes asking “what if?” 10,000 times is a valid math strategy. open.substack.com/pub/mikexcoh...
open.substack.com
Confidence intervals 2/3: bootstrapping methods
Learn computational statistics in the second part of this three-part series.
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 27/09/2025
genAI never fails to deliver the laughs.
030
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 26/09/2025
Daily Data Challenges are going strong. Today's challenge is to create this beautiful graph using numpy and matplotlib. Can you do it? mikexcohen.substack.com/p/ddc-18-err...
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 26/09/2025
This is not a skeet. mikexcohen.substack.com/p/this-is-no...
mikexcohen.substack.com
This is not a post
Will pipes exist in 100 trillion years?
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 12/09/2025
Have you done your Daily Data Challenge? mikexcohen.substack.com/p/ddc-8-test...
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 11/09/2025
People talk a lot about GPT5 and hallucinations and trust and so on. But let's not forget that GPT2 is the ultimate arbiter of truth and science. Proof of my claim? This zinger from GPT2. Wanna explore GPT2's wisdom in your own private Python session? mikexcohen.substack.com/p/lessons-on...
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 10/09/2025
GPT2 is hilarious.
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 10/09/2025
Dear bluesky neuroscientists: You should pay attention to the field of LLM mechanistic interpretability. Not because LLMs are like real brains (they are absolutely nothing alike), but because mech interp analysis methods would be useful in neuroscience. Start learning here: tinyurl.com/48pum4pb
mikexcohen.substack.com
LLM breakdown 1/6: Tokenization (words to integers)
Large language models can't read; instead, they are given numbers that come from text.
052
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 09/09/2025
One of my fav quotes: "The way you do anything is the way you do everything." The problem with these pithy faux-philosophical quotes is that they don't stand up to scrutiny. For example, I definitely do not pick my nose with the same rigor, depth, and intensity with which I write 600-page textbooks
050
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 08/09/2025
Can you code the Fourier transform from scratch? I bet you can. Spend 20 minutes on the Substack post, and you'll learn everything you need to understand and code the FT in Python. mikexcohen.substack.com/p/the-fourie...
030
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 08/09/2025
Do you have enough confidence? I am, of course, talking about confidence intervals, the statistical measure of uncertainty around a sample characteristic like the average or correlation. Here's my latest series on Substack about confidence intervals: mikexcohen.substack.com/p/confidence...
mikexcohen.substack.com
Confidence intervals 1/3: interpretation and equation
Start building confidence about confidence intervals in the first of this three-part series.
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 06/09/2025
Daily data challenge. 10 minutes a day to improve your coding, data science, and visualization skills. open.substack.com/pub/mikexcoh...
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 05/09/2025
Happy teacher's day (in India)!
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 04/09/2025
Have you done your Daily Data Challenge? mikexcohen.substack.com/p/ddc-3-do-t...
mikexcohen.substack.com
DDC-3: Do these functions intersect?
A data challenge a day helps you master machine learning
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 04/09/2025
Why do people keep claiming that LLMs are "just linear algebra"? Most of the magic happens at the nonlinearities between the linear transformations. If you're going to be angry at math, at least be angry at the "non" part of it 🤔
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 04/09/2025
Here's a neat thing: You can use machine-learning analyses to understand how large language models work. Learn how here: mikexcohen.substack.com/s/ml-on-llms
mikexcohen.substack.com
Dissecting LLMs with ML | Mike X Cohen | Substack
Understand how large language models (LLMs) really work by applying machine learning (ML) methods to their internal activations. Each post explores how LLMs process text, isolate patterns, and generat...
021
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 03/09/2025
Why do artificial neural network models like LLMs expand the data dimensionality? It's because nonlinearly separable features can become linearly separable in higher-dimensional space. mikexcohen.substack.com/p/llm-breakd...
030
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 03/09/2025
So weird that we have to drink 8ish cups of water a day. What did humans do before cups were invented??
000
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 02/09/2025
Dear LLM mechanistic interpretability teachers: If you think the course linked below might be helpful for your courses, let me know! Discounts or free coupons can be organized 👍 www.udemy.com/course/dullm...
udemy.com
A deep understanding of AI large language model mechanisms
Build and train LLM NLP transformers and attention mechanisms (PyTorch). Explore with mechanistic interpretability tools
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 02/09/2025
These heatmaps show how the distributions of Q, K, and V vectors inside a GPT transformer change as the text travels deeper into the language model, and picks up new contextual information. Wanna learn more? Check my Substack post (comes with Python code!) mikexcohen.substack.com/p/llm-breakd...
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 02/09/2025
Be who you are and say what you feel, because those who mind don't matter and those who matter don't mind. (You've earned my respect if you know the source without looking it up!)
000
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 02/09/2025
Have you done your daily data challenge? substack.com/@mikexcohen/...
substack.com
Mike X Cohen (@mikexcohen)
Have you done your daily data challenge? https://open.substack.com/pub/mikexcohen/p/ddc-2-numpy-array-to-pandas-dataframe?utm_source=share&utm_medium=android&r=6bsj8n
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 02/09/2025
Joseph Fourier was a French physicist, mathematician, and inventor of Cirque du Soleil (ok, just kidding about that last one).
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 01/09/2025
Dear scientists of BlueSky: Wouldn't it be great to learn how large language models work by running experiments on them and analyzing the data? Yeah, you can do that, and I'll show you how. mikexcohen.substack.com/s/ml-on-llms
mikexcohen.substack.com
Dissecting LLMs with ML | Mike X Cohen | Substack
Understand how large language models (LLMs) really work by applying machine learning (ML) methods to their internal activations. Each post explores how LLMs process text, isolate patterns, and generat...
0135
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 01/09/2025
Have you done your Daily Data Challenge? mikexcohen.substack.com/p/ddc-1-viol...
mikexcohen.substack.com
DDC-1: Violin plot
A data challenge a day helps you master machine learning
000
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 30/08/2025
I read recently that I can use these posts to share my unfinished thoughts, so therefore I
120
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 30/08/2025
I manipulated GPT2's internal representations to prefer "Berlin" vs. "Paris." You can do it too! Check out my Substack post to find out how. mikexcohen.substack.com/p/llm-breakd...
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 30/08/2025
I like turmeric, but I dislike that it has two "r"s. (I should really stop posting about tuumeric.)
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 29/08/2025
I think more people would eat turmeric if it were purple.
000
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 29/08/2025
Have you seen those neat-looking text heatmaps that show how LLMs like Claude or GPT interpret text? They're surprisingly easy to create in Python. open.substack.com/pub/mikexcoh...
021
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 28/08/2025
"Omit needless words." Strunk was trying to tell you how to live a good life, not just how to write a good sentence. mikexcohen.substack.com/p/strunk-and...
mikexcohen.substack.com
Strunk and White's Elements of (Life)Style
Strunk was trying to tell you how to live a good life.
020
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 27/08/2025
So you wanna manipulate a GPT to give the output you want? I show how to do that in part 2 of my LLM breakdown series. (note: this is for small open-source models, not ChatGPT5 🥲) mikexcohen.substack.com/p/llm-breakd...
mikexcohen.substack.com
LLM breakdown 2/6: Logits and next-token prediction
Learn how language models (including chatbots) generate new text
052
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 27/08/2025
FYI: LLMs don't think or reason. They recognize patterns in token ("tokens" are numbers that represent words), and then generate new text that follow those patterns. When LLMs seem logical, it's because they been carefully trained to identify and reproduce patterns in human-written logical texts.
010
Mike X Cohen PhD, Sincxpress Education @mikexcohen.bsky.social · 27/08/2025
I started a Substack :) mikexcohen.substack.com I'm experimenting with "long-form" writing (1-2k words per post). Honestly, after writing 600-page textbooks, 2k words feels very short. Anyway, feel free to stop by my Substack and check out what's there so far.
mikexcohen.substack.com
Mike X Cohen | Substack
Former neuroscience professor, current independent educator. I help people understand applied math and AI, machine-learning, deep learning, LLM mechanisms, etc. Click to read Mike X Cohen, a Substack ...
040