Fig. 06 — One ink: Plum
Technical Interests
| Shape | ID | Title | Blurb | Year | PC1 | PC2 |
|---|---|---|---|---|---|---|
| T-01 | Time series | Learning from sequences that unfold in time: market returns, disease counts, ICU stays and vibration signals. | — | 0.39 | 0.82 | |
| T-02 | Formal verification | Proving that code meets its specification, and using LLMs to write the specifications in the first place. | — | 0.21 | −0.74 | |
| T-03 | Neural operators | Networks that learn maps between functions rather than between fixed-size vectors. | — | 0.72 | 0.52 | |
| T-04 | Physics-ML | Machine learning for physical systems, where the signal has structure a generic model tends to miss. | — | 0.64 | 0.71 | |
| T-05 | Representation learning | What a model's internal space captures, and whether that space is useful for anything downstream. | — | 0.64 | 0.62 | |
| T-06 | Mechanistic interpretability | Reading a model's behavior off its activations, and steering it by editing them. | — | 0.44 | 0.05 | |
| T-07 | Multi-agent safety | How agents that work together can pull each other off course, and how to catch it when they do. | — | 0.56 | −0.71 | |
| T-08 | Blockchain | Systems where the rules live on a shared ledger: smart contracts, on-chain commitments and incentives that settle without a middleman. | — | 0.00 | −0.33 | |
| T-09 | Automated market makers | Exchanges that price assets with a formula over a liquidity pool instead of matching buyers and sellers in an order book. | — | 0.18 | −0.44 | |
| T-10 | Agent evals | Measuring what AI agents actually do: benchmarks, harnesses and scoring that hold up when you look closely. | — | 0.64 | −0.55 |
The technical areas I keep coming back to, from formal verification and neural operators to blockchains and agent evals. Each page links to the work where the interest shows up. Grey marks are the other five layers, kept in place so positions stay honest.