Oz Akan
@oz7z.bsky.social
Building meaningful technology at the intersection of AI and systems. 25+ years leading teams. When not coding, I'm running, reading, or pondering free will and quantum mechanics.
My Blog: https://luminary.blog
My Blog: https://luminary.blog
173 million times the performance: The difference between good enough code and optimized code
luminary.blog/techs/05-rec...
#programming #performance #sustainability
luminary.blog/techs/05-rec...
#programming #performance #sustainability
We Need More Data Centers
One function. Four implementations. A 173-million-times performance difference.
luminary.blog
November 11, 2025 at 8:15 PM
173 million times the performance: The difference between good enough code and optimized code
luminary.blog/techs/05-rec...
#programming #performance #sustainability
luminary.blog/techs/05-rec...
#programming #performance #sustainability
This paper says the universe's expansion is slowing down, not speeding up. That will be a remarkable outcome once confirmed!
#bigbang #universe
arxiv.org/pdf/2510.13121
#bigbang #universe
arxiv.org/pdf/2510.13121
arxiv.org
November 11, 2025 at 8:13 PM
This paper says the universe's expansion is slowing down, not speeding up. That will be a remarkable outcome once confirmed!
#bigbang #universe
arxiv.org/pdf/2510.13121
#bigbang #universe
arxiv.org/pdf/2510.13121
November 8, 2025 at 12:54 AM
We need more centers! For the wrong reasons.
luminary.blog/techs/05-rec...
#programming #optimization #algorithms
luminary.blog/techs/05-rec...
#programming #optimization #algorithms
We Need More Datacenters
One function. Four implementations. A 173-million-times performance difference.
luminary.blog
November 7, 2025 at 9:10 PM
We need more centers! For the wrong reasons.
luminary.blog/techs/05-rec...
#programming #optimization #algorithms
luminary.blog/techs/05-rec...
#programming #optimization #algorithms
What have I done?
% git status -s -u|wc -l
38
> touched too many files at once. It is better to do small, incremental changes but refactoring is messy.
% git status -s -u|wc -l
38
> touched too many files at once. It is better to do small, incremental changes but refactoring is messy.
November 6, 2025 at 3:54 PM
What have I done?
% git status -s -u|wc -l
38
> touched too many files at once. It is better to do small, incremental changes but refactoring is messy.
% git status -s -u|wc -l
38
> touched too many files at once. It is better to do small, incremental changes but refactoring is messy.
Can't afford to have a race condition? Master the Single-Flight Pattern. #SystemDesign luminary.blog/techs/04-sin...
Single-Flight Pattern
How the single-flight pattern can prevent race conditions
luminary.blog
November 6, 2025 at 3:33 PM
Can't afford to have a race condition? Master the Single-Flight Pattern. #SystemDesign luminary.blog/techs/04-sin...
Big O still a mystery? Clear the confusion & boost your skills in minutes 💥 #TechInterviewTips luminary.blog/techs/04-big...
Time & Space Complexity Refresher
Refresher on Big O time and space complexity with interview questions
luminary.blog
November 4, 2025 at 8:10 PM
Big O still a mystery? Clear the confusion & boost your skills in minutes 💥 #TechInterviewTips luminary.blog/techs/04-big...
There's a tendency to forget old knowledge. I feel pretty upset when I realize I’ve forgotten some domain vocabulary after not visiting the subject for a while. I will do something about it; I just didn't have time to implement it.
October 29, 2025 at 4:36 PM
There's a tendency to forget old knowledge. I feel pretty upset when I realize I’ve forgotten some domain vocabulary after not visiting the subject for a while. I will do something about it; I just didn't have time to implement it.
To test the REST API I am building, so far, I have used Apidog, RapidAPI (aka Paw), Thunder Client for VS, REST Client for VS, and EchoAPI for VS. I haven't yet decided which one. Any ideas?
October 28, 2025 at 6:34 PM
To test the REST API I am building, so far, I have used Apidog, RapidAPI (aka Paw), Thunder Client for VS, REST Client for VS, and EchoAPI for VS. I haven't yet decided which one. Any ideas?
Can we ever know, or do we just believe? As AI becomes part of our daily lives, our trust shifts from social intuition to algorithmic “truth.” What happens when machines become the authority?
Read the full article on Substack:
ozakan.substack.com/p/love-ai-and-god
Read the full article on Substack:
ozakan.substack.com/p/love-ai-and-god
Love, AI and God
When people believe in AI
ozakan.substack.com
October 23, 2025 at 3:35 PM
Can we ever know, or do we just believe? As AI becomes part of our daily lives, our trust shifts from social intuition to algorithmic “truth.” What happens when machines become the authority?
Read the full article on Substack:
ozakan.substack.com/p/love-ai-and-god
Read the full article on Substack:
ozakan.substack.com/p/love-ai-and-god
Recently, OpenAI executives falsely claimed that GPT-5 had solved previously unsolved mathematical problems. GPT-5 didn't really solve these problems. As LLMs become more advanced, we will lack the intellectual ability to determine if the answer is correct. There will be believers and non-believers.
October 22, 2025 at 8:45 PM
Recently, OpenAI executives falsely claimed that GPT-5 had solved previously unsolved mathematical problems. GPT-5 didn't really solve these problems. As LLMs become more advanced, we will lack the intellectual ability to determine if the answer is correct. There will be believers and non-believers.
Leadership isn’t about controlling outcomes. It’s about creating the conditions where great outcomes can happen naturally.
www.consultingforeveryone.com/p/the-anti-p...
#leadership #satire
www.consultingforeveryone.com/p/the-anti-p...
#leadership #satire
The Anti-Playbook for Project Managers
A Step-by-Step Guide to Killing Morale, Trust, and Productivity
www.consultingforeveryone.com
October 21, 2025 at 3:20 PM
Leadership isn’t about controlling outcomes. It’s about creating the conditions where great outcomes can happen naturally.
www.consultingforeveryone.com/p/the-anti-p...
#leadership #satire
www.consultingforeveryone.com/p/the-anti-p...
#leadership #satire
September 18, 2025 at 8:22 PM
I wish the Claude VSCode extension could also be a right-column app like ChatGPT Codex is.
#claude #claude-code #codex #vscode #code
#claude #claude-code #codex #vscode #code
September 16, 2025 at 5:20 PM
I wish the Claude VSCode extension could also be a right-column app like ChatGPT Codex is.
#claude #claude-code #codex #vscode #code
#claude #claude-code #codex #vscode #code
Ever wonder how ChatGPT gets so smart? It all starts with loss. Dive in👇 #AI luminary.blog/techs/llm-lo...
Which Loss Function Do LLMs use?
Exploring Cross-Entropy Loss in Large Language Models.
luminary.blog
September 10, 2025 at 1:17 AM
Ever wonder how ChatGPT gets so smart? It all starts with loss. Dive in👇 #AI luminary.blog/techs/llm-lo...
September 8, 2025 at 11:33 PM
This AI trick packs multiple brains into one—meet Matryoshka learning 💡 #MachineLearning luminary.blog/techs/matryo...
What is Matryoshka Representation Learning (MRL)?
Nesting Power and Flexibility into ML Embeddings
luminary.blog
September 5, 2025 at 4:12 PM
This AI trick packs multiple brains into one—meet Matryoshka learning 💡 #MachineLearning luminary.blog/techs/matryo...
Forget big numbers—UE8M0 FP8 is small, fast & mighty. AI will never train the same. #AI luminary.blog/techs/deep-s...
UE8M0 FP8 Number Format
Training LLMs without H100 using UE8M0 FP8 number format.
luminary.blog
August 28, 2025 at 3:24 PM
Forget big numbers—UE8M0 FP8 is small, fast & mighty. AI will never train the same. #AI luminary.blog/techs/deep-s...
Unlock ML secrets hidden in your numbers—precision matters more than you think! #MachineLearning luminary.blog/techs/number...
Understanding ML Numerical Formats
Understanding INT4, INT8, FP16, BF16, and TF32 formats in machine learning - their precision, speed, and memory trade-offs for training and inference.
luminary.blog
August 26, 2025 at 5:57 PM
Unlock ML secrets hidden in your numbers—precision matters more than you think! #MachineLearning luminary.blog/techs/number...
LLM inference is what we do all day. If you want to understand how it really works, I wrote about it a while ago.
luminary.blog/techs/07-llm...
#llm #inference #aiml
luminary.blog/techs/07-llm...
#llm #inference #aiml
What is LLM Inference?
Understanding how Large Language Models generate text through the inference process.
luminary.blog
August 19, 2025 at 10:56 PM
LLM inference is what we do all day. If you want to understand how it really works, I wrote about it a while ago.
luminary.blog/techs/07-llm...
#llm #inference #aiml
luminary.blog/techs/07-llm...
#llm #inference #aiml