AI Research
9 essays on this topic.
Why Do Neural Networks Generalize? Overfitting, Double Descent and Grokking
Big networks memorise their data and still work on new data. Overfitting, double descent and grokking in plain English, with a numpy demo you can run.
How to Read an ML Paper in One Hour
A three-pass method for reading machine learning papers in an hour: what to read, what to skip, and how to check the experiments before you trust a claim.
What the Research Says: Mapping RIAI Against the Field
RIAI Lab Notes #6: I read four research fields against my own results. What others built, where RIAI stands, and the experiments I'll run next.
Write It, Then Prove It: AI That Checks Its Own Work With Math
AI paired with proof checkers like Lean: Olympiad medals, Fermat's Last Theorem, verified code, and the catch: a proof only proves what the spec says.
Watching JEPA Write Code Live: What It Got Right, and the Check It Fooled
RIAI Lab Notes #5: a live test of the JEPA code writer. Fast, readable, no LLM, and correct for its examples. But the examples didn't say what was meant.
Next: Teaching an Agent to Imagine Code With JEPA
RIAI Lab Notes #4: why exact-value prediction hit a wall, how a JEPA plans code in vector space, and first results: 10 of 12 when grounded.
Writing Its Own Code: From a Teacher Model to Code Built by Search
RIAI Lab Notes #3: moving the 'write the code' step from Claude Haiku into the agent, with program search, execution traces and honest gaps.
A Terminal That Learns What Commands Do
RIAI Lab Notes #2: an agent that learns what sort, grep and its own Python tools do by experimenting, then plans in its imagination.
RIAI: Toward a Terminal Agent That Writes Its Own Skills
Lab notes from RIAI: my goal is an agent that does everything from the terminal and writes its own Python skills to keep growing. Where it stands today.
