{"id":51076993,"url":"https://github.com/zozo123/how-stackoverflow-taught-ai","last_synced_at":"2026-06-23T15:02:06.129Z","repository":{"id":363098335,"uuid":"1261956512","full_name":"zozo123/how-stackoverflow-taught-ai","owner":"zozo123","description":"Interactive deep-dive: how Stack Overflow became the foundational training data for AI coding agents — and trained its own replacement. Vanilla JS + Plotly + KaTeX.","archived":false,"fork":false,"pushed_at":"2026-06-07T12:19:22.000Z","size":862,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-07T13:19:37.762Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"HTML","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/zozo123.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-07T11:37:07.000Z","updated_at":"2026-06-07T12:19:25.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/zozo123/how-stackoverflow-taught-ai","commit_stats":null,"previous_names":["zozo123/how-stackoverflow-taught-ai"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/zozo123/how-stackoverflow-taught-ai","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/zozo123%2Fhow-stackoverflow-taught-ai","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/zozo123%2Fhow-stackoverflow-taught-ai/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/zozo123%2Fhow-stackoverflow-taught-ai/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/zozo123%2Fhow-stackoverflow-taught-ai/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/zozo123","download_url":"https://codeload.github.com/zozo123/how-stackoverflow-taught-ai/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/zozo123%2Fhow-stackoverflow-taught-ai/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34694786,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-23T02:00:07.161Z","response_time":65,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2026-06-23T15:02:04.030Z","updated_at":"2026-06-23T15:02:06.123Z","avatar_url":"https://github.com/zozo123.png","language":"HTML","funding_links":[],"categories":[],"sub_categories":[],"readme":"# The Data That Taught the Machines\n\n**An interactive deep-dive into how Stack Overflow became the foundational training data for AI coding agents — and then trained its own replacement.**\n\n🔗 **Live:** https://zozo123.github.io/how-stackoverflow-taught-ai/\n\nBuilt in the spirit of a *learn-by-dragging* sandbox (inspired by the [market-making-sandbox](https://zozo123.github.io/market-making-sandbox/)): vanilla JS, [Plotly](https://plotly.com/javascript/) \u0026 [KaTeX](https://katex.org/). No backend, no telemetry, single file.\n\n## What's inside\n\nA single-page interactive essay with hands-on widgets:\n\n1. **The perfect classroom** — click through the anatomy of a Stack Overflow post and see how each part (question, code, explanation, votes) maps onto a phase of LLM training.\n2. **The reward signal** — slide upvotes / toggle \"accepted\" and watch the StackLLaMA reward function the model is trained to maximize.\n3. **The corpus** — the documented contribution of Stack Exchange to The Pile, LLaMA, InCoder, RedPajama, Dolma and StarCoder2, with the labs' own justifications.\n4. **The collapse** — the famous question-volume chart (−76% since ChatGPT), from Stack Overflow's own Data Explorer, with toggles.\n5. **The ouroboros** — a live model-collapse simulator (Shumailov *Nature* 2024) with a \"replace vs. accumulate\" switch (Gerstgrasser 2024).\n6. **Mirrors \u0026 memory** — models learn culture and code; the memorization \u0026 CC BY-SA attribution problem.\n7. **The historical arc** — 2008 → today in seven beats.\n8. **What's next** — the incentive problem nobody has solved.\n9. **Go deeper** — a curated learning path (Karpathy, fast.ai, Hugging Face, the foundational papers).\n10. **Receipts** — every headline number sourced; contested and illustrative claims explicitly flagged.\n\nSee [`POST.md`](./POST.md) for the written article version, and [`SOCIAL.md`](./SOCIAL.md) for launch copy (X thread, LinkedIn, HN).\n\n## Social cards\n\n- `og-image.png` (1200×630) — wired into Open Graph + Twitter meta; auto-previews when the link is shared.\n- `og-square.png` (1200×1200) — square variant for Instagram / Mastodon / manual posts.\n\nBoth are generated from `og.html` / `og-square.html` via headless Chrome — edit the HTML and re-render.\n\n## Run locally\n\nIt's a single static file — just open it:\n\n```bash\nopen index.html      # macOS\n# or serve it:\npython3 -m http.server 8080   # then visit http://localhost:8080\n```\n\n## Sourcing \u0026 honesty\n\nEvery statistic links to a primary source on the page. Figures that are contested (e.g. traffic-decline methodology, the exact peak) are tagged, and one illustrative narrative element (the \"StackGPT\" thought experiment) is clearly labeled as *not a verified event*. The point stands without overclaiming.\n\n---\n\n*Generated with [Claude Code](https://claude.com/claude-code).*\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fzozo123%2Fhow-stackoverflow-taught-ai","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fzozo123%2Fhow-stackoverflow-taught-ai","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fzozo123%2Fhow-stackoverflow-taught-ai/lists"}