dictionary-of-ai-coding

Clarified the explanation of output tokens and their billing. Output generation is not inherently several times more computationally expensive per token. Its higher serving cost primarily comes from the sequential nature of autoregressive decoding and the resulting lower hardware utilization and greater state-management burden. Technics like continuous batching amortizes much of the weight-access cost across users and substantially improving decode efficiency but it cannot parallelize successive tokens of the same sequence and does not eliminate per-sequence KV-cache traffic so it's memory heavey in nature.

has an unquoted containing (right after "in-memory"). YAML reads that as a nested mapping, so the frontmatter fails to parse. Repro Open in VS Code and toggle the markdown preview. Instead of the entry you get: Failed to parse frontmatter** Nested mappings are not allowed in compact mappings at line 1, column 14 Or from the shell: Why it went unnoticed is unaffected. slices between the markers and never parses YAML, so the generated README is correct either way. The defect only surfaces in tools that actually read the frontmatter, which is also why the generated looks fine. The fix Wrap the value in double quotes, the same way seven other entries already do. No content change, and the text stays at 133 characters, under the 140 limit in . This was the only one of the 69 entries affected. I checked the other 68 parse cleanly. 🤖 Generated with Claude Code