Posts
- Earth was born dry until a cosmic collision made it a blue planet
- Pavel Durov's interview with Lex Fridman
- Agentic Loop Design - Tool #1: Long-Term Memory that Survives Resets
- Make sense of "Codex produces 13x Merged PR than Claude Code"
- Agentic Loop Design
- Mental models for AI along the ambiguous ladder
- Mental Flexibility as a Software Engineering Maturity Metric
- Codex knows Python
- Thoughts on Richard Sutton's interview on Dwarkesh Podcast
- Less is More for X
- How Prompt Writing Style Shapes GPT-5 and Codex
- Release `pydantic-optuna-bridge`
- Tests for Fit and Tests for Correctness
- Fail fast and noisily creates more resilient software
- Resist the tempatation to anthropomorphizing LLM
- Iberian harvester ant queens produce offspring of their own species and of the builder harvester ant
- Surprising GPT-OSS 20B 2-bit Quantization Performance
- Due to odd JAX issues
- How OpenAI writes prompts for text classification
- Qwen-8B Embeddings: Near-SOTA Performance at 600x the Speed
- Evolutionary Test-Time Compute: trade time & token for creativity
- Three great virtues of an AI-assisted programmer
- Qwen3-30B: The first AI model that is good enough for local deployment
- Two Insights from Tao's Blue/Red Teams Metaphor: Software Testing's Future and AI as a Coach
- This is to have succeeded.
Subscribe via RSS