AI engineering

How agent harnesses, models and AI tooling actually behave once you build on them.

Playlists

September 2026
6 landmark AI papers you must read
The transformer, GPT-3, chain-of-thought, RAG, Chinchilla and the Bitter Lesson, read in the original, with the caveats lost in the retelling.
6 posts
September 2026
System 1 AI is here: Jev decides in milliseconds
How TypeSafe's Jev decides in milliseconds, where it matches general models and where it trails them, and how to route on its confidence score.
3 posts

Posts

The Bitter Lesson, explained in 4 minutes
Rich Sutton's 1,100-word case that methods which scale with computation eventually beat hand-built human knowledge. What it claims, and what it leaves out.
23 Sept 20265 min read4 min listen
Routing AI decisions on a confidence score you can trust
A calibrated model that reports 0.8 should be right 8 times in 10. How that turns an AI step into a threshold your code can actually branch on.
22 Sept 20266 min read5 min listen
Chinchilla: why parameter counts stopped being the headline
A 70B model beat a 280B one by reading 4 times more text. What compute-optimal training changed, and the transferable idea sitting under the ratio.
21 Sept 20268 min read6 min listen
Jev vs general-purpose LLMs: what 75x cheaper actually buys
Jev matches a general model that costs 75 times as much, and trails the strongest by 6 points. How to decide whether that trade fits your workflow.
20 Sept 20266 min read5 min listen
The RAG paper: what retrieval actually buys you
Human judges called the retrieval-backed model more factual 42.7% of the time against 7.1% for the plain one. The original is more careful than its descendants.
19 Sept 20267 min read7 min listen
What is Jev? The fast model that answers in JSON
Jev is a System 1 model built for classification. What it does, where it is far faster, and the 4 jobs it is wrong for.
18 Sept 20267 min read7 min listen
Chain-of-thought prompting, and when it backfires
Asking for step-by-step reasoning only helps above a model-size threshold. Below it, results often get worse. Why that matters more on cheap models now.
17 Sept 202610 min read8 min listen
The GPT-3 paper: where prompting came from
Language Models are Few-Shot Learners is why you put examples in your prompts. What in-context learning claimed, and the caveats it admitted.
16 Sept 20269 min read7 min listen
Attention Is All You Need: what the paper actually says
The 2017 paper behind every model you use. What the Transformer replaced, why attention is quadratic, and which of its choices you are still paying for.
15 Sept 202611 min read10 min listen

Playlists

6 landmark AI papers you must read

System 1 AI is here: Jev decides in milliseconds

Posts