Notes
All notes
Every note in the river, newest first, grouped by publishing day.
MONDAY 17 AUGUST 2026
- 17 AUG 2026 · 09:20 · EVALUATIONHow to compare two language models without fooling yourself
- 17 AUG 2026 · 08:05 · CONTEXTToken counts are not word counts
FRIDAY 14 AUGUST 2026
- 14 AUG 2026 · 16:40 · BENCHMARKSWhat a published benchmark score leaves out
- 14 AUG 2026 · 11:15 · BENCHMARKSA tie is a result, not a failed test
- 14 AUG 2026 · 09:00 · EVALUATIONBuilding an evaluation set from fifty examples you care about
TUESDAY 11 AUGUST 2026
- 11 AUG 2026 · 15:30 · CONTEXTContext windows: capacity is not attention
- 11 AUG 2026 · 10:10 · PROMPTINGSampling temperature changes what you are measuring
THURSDAY 6 AUGUST 2026
- 6 AUG 2026 · 14:25 · COST & LATENCYLatency, throughput and cost move in different directions
- 6 AUG 2026 · 11:40 · COST & LATENCYCached prompts distort a latency comparison
- 6 AUG 2026 · 09:45 · DEPLOYMENTWhat changes when the weights are on your own machine