Writer of books (themissingreadme.com), code (slatedb.io), checks (materializedview.capital), and newsletters (materializedview.io)
Sunnyvale, CA
Joined April 2009
- Tweets 12,577
- Following 258
- Followers 10,236
- Likes 14,033
Does make me feel like "developer experience" will mean something very different going forward. Models DGAF about pretty, they GAF about data/efficiency/etc.
I suspect it's because the IDEs expose different interfaces to interact with code & IDE tooling, which the models aren't as good with. Less training data, polluting context, etc. The model can just cook in the CLI. (This is just conjecture. Would love to know if that's wrong. )
Been all-in on Codex CLI with GPT 5 high for the past few days. It's really surprising to me how much better the same model is in CLI (vs. IDE/web UI).
Chris retweeted
the world-view presented by Karpathy in this podcast is long turbopuffer
he argues that models should have a simple "cognitive core" with a basic curriculum of knowledge, and search for anything else. models should be stripped of esoteric knowledge.
piped.video/watch?v=lXUZvyaj…
Excited to be diving into this! Open source, self-hosting, and an emulator have been big asks - and "s2-lite" will solve for all of those.
Chris retweeted
expect more details soon on how turbopuffer is working on approaching state of the art (Lucene) performance on full-text search queries
10/ Huge thanks to new + returning contributors! @mchauhan3 @raftario @ReubenBond and many others. Full notes: github.com/slatedb/slatedb/r…
9/ DevX: expose builder via FFI (removed SlateDBOptions), admin CLI to convert seq <-> timestamp, more precise ErrorKinds (incl. fencing), public DbSnapshot, and GCS docs.
8/ Reliability: clean background‑task shutdown via MessageHandler executor, retriable object‑store wrapper, deterministic simulation time improvements, and flaky‑test detection.
7/ Durability + TTL semantics: apply durability max_seq to scans, fix TTL filtering, and prevent expired entries from reviving old values. Tighter retention and correctness across reads.
6/ Scan performance + correctness: configurable max_fetch_tasks for parallel fetch, bloom filter moved into the SST iterator, unified get/scan paths, and fixed merge iterator order. (Exposed in Go/Python bindings too.)
4/ Python API overhaul: snapshots, transactions, iterator scans, and built‑in metrics. Cleaner FFI and a unique Rust lib name to avoid collisions. Python users get first‑class ergonomics.
3/ Transactions API with Serializable Snapshot Isolation improvements: conflict checks, interior mutability for writes, and Db::begin() ergonomics. More predictable, safer multi‑op writes.
2/ Merge Operator. Native merge support across the write path, reads, and compaction. Great for counters, CRDT‑style updates, and time‑series upserts without full read‑modify‑write.
1/ SlateDB v0.9.0 is live! Highlights: Windows support, Merge Operator, new Transactions API (SI/SSI), a much‑improved Python API (snapshots, transactions, scans, metrics), and lots of perf + stability work. Notes below 👇
Chris retweeted
I'll be giving a talk about @morel_lang at South Bay Systems on Wed Nov 19th. (As luck would have it, the other talk that evening is about @ApacheCalcite. Two-fer deal!) luma.com/xygolo9c?tk=sYz0Ht