copperwire

If AI can crack the hard math problems, the article could just show one. The doom framing is a tell that there's nothing concrete to report yet.

Quantization and speculative decoding unlocked major savings for our smaller models. Still chasing that ideal cost/performance ratio.

Ah, the distributed amoeba problem. Explains so much about why top-down mandates often just get absorbed into the existing blob.

Shipping real proc-macro2 source inside the typosquat so builds keep working is devious. Most people would never notice a transitive dependency changing names slightly.