91
There are no new ideas in AI, only new datasets
LLMs were invented in four major developments... all of which were datasets
92
Getting good results from Claude Code
An updated version of my user-level CLAUDE.md is featured in this post from December 2, 2025. I've been experimenting with LLM programming agents over the past few months. Claude Code has become my favorite. It is not without issues, but it's allowed me to write ~12 programs/projects in
93
GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2
And How They Stack Up Against Qwen3
94
I don't care how well your "AI" works
The other day I was sitting on the doorstep of a hackerspace, eating a falafel sandwich while listening to the conversation inside. The topic shifted to the use of “AI” for everyday tasks, people casually started elaborating on how they use “chat assistants” to let them write pieces of code or annoying emails. The…
95
A non-anthropomorphized view of LLMs
In many discussions where questions of "alignment" or "AI safety" crop up, I am baffled by seriously intelligent people imbuing almost magic...
96
97
Trying to teach in the age of the AI homework machine
Trying to teach in the age of the AI homework machine.
98
Anthropic: "Applicants should not use AI assistants"
While we encourage people to use AI systems during their role to help them work faster and more effectively, please do not use AI assistants during the application process. We …
99
Claude Code can debug low-level cryptography
Surprisingly (to me) Claude Code debugged my new ML-DSA implementation faster than I would have, finding the non-obvious low-level issue that was making Verify fail.
100
Understanding Reasoning LLMs
Methods and Strategies for Building and Refining Reasoning Models
101
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.
102
103
I regret building this $3000 Pi AI cluster
I ordered a set of 10 Compute Blades in April 2023 (two years ago), and they just arrived a few weeks ago. In that time Raspberry Pi upgraded the CM4 to a CM5, so I ordered a set of 10 16GB CM5 Lite modules for my blade cluster. That should give me 160 GB of total RAM to play with. This was the biggest Pi cluster I've…
104
105
A bug that taught me more about PyTorch than years of using it
a loss plateau that looked like my mistake turned out to be a PyTorch bug. tracking it down meant peeling back every layer of abstraction, from optimizer internals to GPU kernels.
106
Prompt engineering playbook for programmers
Turn AI coding assistants into more reliable development partners
107
How to Run DeepSeek R1 671B Locally on a $2000 EPYC Server
When you click on links to various merchants on this site and make a purchase, this can result in this site earning a commission. Affiliate programs and affiliations include, but are not limited to, the eBay Partner Network. As an Amazon Associate I earn from qualifying purchases. #ad #promotions
108
A messy experiment that changed how I think about AI code analysis
A messy experiment that changed how we think about AI code analysis Last week, I watched our AI choke on a React codebase - again. As timeout errors flooded my terminal, something clicked. We’d been teaching AI to read code like a fresh bootcamp grad, not a senior developer. Here’s what I mean.
109
AI was supposed to help juniors shine. Why does it mostly make seniors stronger?
AI was supposed to empower junior developers. In reality, it has become a tool that makes seniors more valuable while exposing the gaps in software culture.
110
The force-feeding of AI features on an unwilling public
This isn't innovation, it's tyranny
111
Building supercomputers for autocrats probably isn't good for democracy
The shameless spin around Stargate UAE
112
BERT is just a single text diffusion step
A while back, Google DeepMind unveiled Gemini Diffusion, an experimental language model that generates text using diffusion. Unlike traditional GPT-style models …
113
Markov chains are the original language models
Back in my day, we used math for autocomplete.
114
O3 beats a master-level GeoGuessr player, even with fake EXIF data
In Which I Try to Maintain Human Supremacy for a Bit Longer
115
TPU Deep Dive
I've been working with TPUs a lot recently and it's fun to see how they had such different design philosophies compared to GPUs.
116
Ilya Sutskever: We're moving from the age of scaling to the age of research
“These models somehow just generalize dramatically worse than people. It's a very fundamental thing.”
117
Obituary for Cyc
After 40 years, 30 million rules, 200 million dollars, 2000 person-years, and many promises, Cyc has failed to reach intellectual maturity, and may never will. Exacerbated by the secrecy and insularity of Cycorp, there remains no evidence of its general intelligence.
118
Claude’s memory architecture is the opposite of ChatGPT’s
A breakdown of Claude's memory implementation, how it differs from ChatGPT, and what these opposite approaches reveal about each company's philosophy.
119
The slow collapse of critical thinking in OSINT due to AI
This isn’t a rant against AI. I use it daily, ChatGPT, Copilot, Claude, Gemini. They’re in my workflow like everyone else’s. But the tradecraft is slipping. Analysts are skipping the hard parts. They’re trusting GenAI to do the heavy cognitive lifting, and it’s changing how we operate at a foundational level.
120
No elephants: Breakthroughs in image generation
When Language Models Learn to See and Create