@holliplex.bsky.social

@holliplex.bsky.social

36 Followers 411 Following

Floating in space

Terminal text: whoops I'm back after about 4 days, what did I miss ⏺ Hi — welcome back. Pausing the loop. Here's what happened in your absence: Summary of ~4 days of autonomous fuzzing The differential fuzzer (dmars vs pmars via the new oracle-pmars FFI from the last commits you made before leaving) has been running continuously. Each crash got reproduced, minimized, diagnosed, fixed or filtered, committed with jj, and logged to LLM_BUGS.md. Progress: 16 new bug entries (#25–#40) on top of the 24 that were already there. Split roughly: - Real dmars correctness fixes — macro-instruction gating (expansion_starts_with_opcode), sequential-vs-forward-ref EQU tracking (new equs_defined_so_far state in expander), VT/FF as whitespace in evaluator, stray ! handling, ROF prefix matching, etc. - Harness filters for pmars quirks — bounds on line length (200), digit runs (7), consecutive * (≥4), FOR counts (100), commas in FOR headers, etc. The filter file is fuzz/fuzz_targets/assemble_differential.rs. Regression seeds saved under fuzz/regression/assemble/. cargo t --lib green at every commit. Latest head: a6f03d21 fuzz: reject FOR headers with \,` in the count expression`. Current state — fresh crash waiting on triage: Input (73 bytes): nested FOR 8 / FOR 6+2 → dmars expands to 64 instructions, pmars rejects and emits 1. Not OOM, not timeout — a real differential. Likely another harness-filter case (nested FOR amplification past the 100 cap, or pmars error-recovery difference) but could also be a gate bug in the expander around headerless FOR. Artifact: fuzz/artifacts/assemble_differential/crash-e2a2c88db4b2da9b22bff730d11d558d981e0916. Want me to resume the loop on this crash, or pause so you can skim the diff first?
A screenshot of a stress-test of the library, showing the WebGPU renderer can draw ~10000 voxels in 2.3ms vs 49ms for SVG.
“Peierls' reply was kind but firm. He said 'I do not beleve that left-right symmetry can be violated in weak nuclear forces at all'. Thus rebuffed in Birmingham, like Zuleika Dobson, I wondered to whom I could go next and the obvious person was Pauli, the father of the neutrino, who at that time served as an oracle on the subject. I went to CERN at Geneva and sent Pauli my paper in the hands of Professor Villa of MIT who was to visit him at his home in Zurich. Soon the message came back. 'Give my regards to my friend Salam and tell him to think of something better'.”
Screenshot of a google search result with “pauwels sauces - your partner in sauces” followed by “Pauwels Transformers - your partner in energy solutions”
Photo of the Ivanpah thermosolar plant taken from an aeroplane. Three enormous towers surrounded by reflectors, the towers are glowing white.