Kittipat Kampa

0 followers · 1 following · 254 views

on the atlas · 4

highlights · 8

Future-proof architecture. As models get better at coding, your agent gets better. Every improvement in code understanding translates directly. You're leveraging the training distribution instead of fighting against it.
How to build agents with filesystems and bash - Vercel
handles edge cases we never anticipated because it's working with the raw information, not parameters we defined.
How to build agents with filesystems and bash - Vercel
The intuition is that the agent treats the transcript like a codebase. It searches for patterns, reads sections, and builds context just like it would debug code. No custom retrieval logic.
How to build agents with filesystems and bash - Vercel
Do not train a child to learn by force or harshness; but direct them to it by what amuses their minds, so that you may be better able to discover with accuracy the peculiar bent of the genius of each.
Quote of the day by Plato: ‘Do not train a child to learn by force or…’ - The Economic Times
You can tighten / relax in categories that matter to users, semantically, up a layer to still stay within the realm of “kid play”. If the taxonomy makes sense to users, if it reflects search behaviors (big ifs), they’ll understand why you include / exclude certain results.
Semantic Search Without Embeddings
We need is a hierarchical tokenizer. A function that produces the full path of every parent directory
Semantic Search Without Embeddings
who was a grocer and he read the progressive grocer magazine, and he read articles on how to stock a meat department... What he’s really done is he’s created this immense vertical filing cab
Curius / Onboarding
The subject that interests him, he’s read newspapers, biographies, trade press.
Curius / Onboarding