991
The Default Trap: Why Anthropic's Data Policy Change Matters
A quick note on the importance of reading the terms of service!
New
The latest blog posts to take off on Hacker News. 3,091 posts.
992
From multi-head to latent attention: The evolution of attention mechanisms
From Multi-Head to Latent Attention: The Evolution of Attention Mechanisms What is attention? In any autoregressive model, the prediction of the future tokens is based on some preceding context …
993
Flunking my Anthropic interview again
I didn't misclick any buttons. My best wasn't good enough. I'm not good enough.
994
An LLM is a lossy encyclopedia
Since I love collecting questionable analogies for LLMs, here's a new one I just came up with: an LLM is a lossy encyclopedia. They have a huge array of facts …
995
The Bitter Lesson Is Misunderstood
Together, the Bitter Lesson and Scaling Laws reveal that the god of Compute we worship is yoked to an even greater one — the god of Data.
996
Some thoughts on LLMs and software development
a short post
997
Building your own CLI coding agent with Pydantic-AI
How to build a CLI coding agent
998
Will AI Replace Human Thinking? The Case for Writing and Coding Manually
Learning to Think Again, and the Cost of AI Dependency.
999
Rendering a game in real time with AI
Discover the process of transforming an ASCII game into real-time AI-rendered graphics with fal.ai
1000
The Math Behind GANs (2020)
Generative Adversarial Networks refer to a family of generative models that seek to discover the underlying distribution behind a certain data generating process. This distribution is discovered through an adversarial competition between a generator and a discriminator. As we saw in an earlier introductory post on…
1001
Important machine learning equations
A reference guide to fundamental machine learning equations with correct implementations and clear explanations of how they connect to each other.
1002
Are OpenAI and Anthropic losing money on inference?
Deconstructing the real costs of running AI inference at scale. My napkin math suggests the economics might be far more profitable than commonly claimed.
1003
I Am An AI Hater
I am an AI hater. This is considered rude, but I do not care, because I am a hater.
1004
AI coding made me faster, but I can't code to music anymore
AI coding has made me much more productive—but at a strange cost. I can't listen to music when I code anymore.
1005
Why I'm declining your AI generated MR
Sometimes a merge request (MR) doesn't merit a code review (CR) because AI was used in a bad way that harms the team or the project. For example:
1006
In Search of AI Psychosis
Folie a deux ex machina
1007
Will Smith's concert crowds are real, but AI is blurring the lines
Will Smith is being accused of generating fake fans with AI, but it's complicated: the crowds are real, but the videos were manipulated by Smith's team and YouTube itself.
1008
A bubble that knows it's a bubble
Railway Mania, dot-com, AI. Technology bubbles follow a pattern older than any of them. What's different this time is that everyone knows the script.
1009
Making games in Go: 3 months without LLMs vs. 3 days with LLMs
After 15 years as a software engineer, I realized I had never actually built and published a game.
1010
Writing with LLM is not a shame
Do we have to disclaim how we use AI ?
1011
DeepWiki: Understand Any Codebase
How I use DeepWiki to speed up real coding work
1012
How to build a coding agent
It's not that hard to build a coding agent. 300 lines of code running in a loop with LLM tokens. You just keep throwing tokens at the loop, and then you've got yourself an agent.
1013
Evaluating LLMs for my personal use case
My life is not a math Olympiad
1014
AGI is an engineering problem, not a model training problem
LLM models are plateauing, but true AGI isn't about scaling the next breakthrough model—it's about engineering the right context, memory, and workflow systems. AGI is fundamentally a systems engineering problem, not a model training problem.
1015
Writing Speed-of-Light Flash Attention for 5090 in CUDA C++
In this post, I will walk through how I learned to implement Flash Attention for 5090 in CUDA C++. The main objective is to learn writing attention in CUDA C++, since many features are not available in Triton, such as MXFP8 / NVFP4 MMA for sm120. I also feel this is a natural next step after learning about matmul…
1016
My experience creating software with LLM coding agents – Part 2 (Tips)
moved to https://blog.efitz.net/posts/my-experience-creating-software-with-llm-coding-agents-part-2-tips/
1017
Sprinkling self-doubt on ChatGPT
I replaced my ChatGPT personalization settings with this prompt a few weeks ago and promptly forgot about it: Be extraordinarily skeptical of your own…
1018
A guide to Gen AI / LLM vibecoding for expert programmers
I get it, you’re too good to vibe code. You’re a senior developer who has been doing this for 20 years and knows the system like the back of your hand. Or maybe you’re the star individual contributor who is the only person who can ever figure out how to solve the hard problems. Or maybe you’re the professor who…
1019
In the long run, LLMs make us dumber
The comfort we get when offloading our cognitive load to LLMs is bad for us. Cognitive load should exist, and if we reduce it too much – if we stop thinking – we can actually unlearn how to think. Kids who always choose the easy route and copy their homework from other students eventually find […]
1020
Building AI products in the probabilistic era
AI turns products from deterministic functions into probabilistic systems. That requires expanding old playbooks (SLOs, funnels, siloed finance), and reasoning in terms of trajectories, Minimum Viable Intelligence thresholds, and data as company operating system.