{"slug": "asd-ste100-simplified-technical-english-for-llms", "title": "ASD-STE100 Simplified Technical English for LLMs", "summary": "A new open-source toolchain applies ASD-STE100 Simplified Technical English to large language model outputs, achieving a 74% reduction in writing violations on Claude Sonnet and a 50% reduction on GPT-5.5, according to tests by the developer. The package includes a distilled agent skill, a deterministic linter, and reproduction scripts, all requiring only Python 3.", "body_md": "Everything the episode promises \"in the description\": the distilled STE skill, the linter, and the full first-party test data. Self-contained - no external dependencies beyond Python 3.\n\n| File | What it is |\n|---|---|\n`ste-writing-skill.md` |\nThe distilled ASD-STE100 agent skill, two modes: strict (procedures, error messages) and flavored (prose, no dictionary lockdown) |\n`ste-lint.py` |\nThe heuristic anti-slop linter - the machine-checkable subset of STE. Deterministic; the score delta between two texts is the signal |\n`experiment-results.md` |\nThe cross-model test: 6 writing tasks x 4 conditions on Claude and gpt-5.5 |\n`experiment-results-openai.md` |\nPer-category detail for the gpt-5.5 run |\n`before-after-samples.md` |\nThe real baseline-vs-STE outputs quoted in the episode |\n`run-openai.py` |\nReproduction script for the OpenAI side of the experiment |\n\n```\npython3 ste-lint.py your-draft.md\n```\n\nScore is violations per 100 words - lower is cleaner. Lint a draft, apply the skill, then lint it again - the delta between the two scores is the signal.\n\n| Condition | Claude sonnet | gpt-5.5 |\n|---|---|---|\n| baseline | 4.36 | 3.54 |\n| banned-words list | 4.21 (-3%) | 2.14 (-40%) |\n| Orwell's 6 rules | 2.48 (-43%) | 1.69 (-52%) |\n| STE skill | 1.12 (-74%) | 1.76 (-50%) |\n\nGive the model a writing system and slop drops by half or more, on every model tested. STE was best or tied-best. A banned-words list is the least reliable fix.\n\nNot a certified STE checker. The judgment rules of ASD-STE100 need a human; this covers the mechanical subset - which is where the slop lives.\n\nSpec: ASD-STE100 Issue 9, free at asd-ste100.org", "url": "https://wpnews.pro/news/asd-ste100-simplified-technical-english-for-llms", "canonical_source": "https://github.com/woosal1337/blog/tree/main/videos/ep01-the-cure-for-ai-slop", "published_at": "2026-07-27 06:52:01+00:00", "updated_at": "2026-07-27 07:22:50.415779+00:00", "lang": "en", "topics": ["large-language-models", "ai-tools", "natural-language-processing"], "entities": ["ASD-STE100", "Claude Sonnet", "GPT-5.5", "Python 3"], "alternates": {"html": "https://wpnews.pro/news/asd-ste100-simplified-technical-english-for-llms", "markdown": "https://wpnews.pro/news/asd-ste100-simplified-technical-english-for-llms.md", "text": "https://wpnews.pro/news/asd-ste100-simplified-technical-english-for-llms.txt", "jsonld": "https://wpnews.pro/news/asd-ste100-simplified-technical-english-for-llms.jsonld"}}