Sara Chrouf / Idea Maps
Idea Maps 図
A map is not an essay; it arranges dozens of materials into a graded route. Knowledge is a network, so it deserves cartography.
Map 01 · Understanding language models
Not an essay; a learning path of eight stations, each linking a moving term, a paper, or an essay. Start at the top and follow the current.
01 Data
A model is a compression of its corpus; an honest corpus is mopped before it is admired.
02 Tokenization
The first border: how text becomes integers, and who pays the tariff.
03 Training and injection
You need not retrain the world to add a language; Kuwain injected Arabic into a small model and kept its memory.
04 Diacritics
The vowels the page hides are grammar the model can learn to restore, for free.
05 Translation and dialects
Between Arabic and English, then inside Arabic itself; the hardest border can be domestic.
06 Evaluation
A capability equals its exam; inherit no exam without reading its provenance.
07 Inference and its economics
Serving is where fertility becomes a bill; the token is the meter.
08 Agents
Long traces, long memory, and a dialect gap between user and system; the current frontier of my desk.
Authorship, simulation, originality, ownership, taste.
Perception, language, models, and the philosophical objections.
Attention, memory, note-making, and compound knowledge.