--- title: Best sean goedecke posts — July 2026 | daily.dev description: The most upvoted sean goedecke posts from July 2026, curated by the daily.dev community. canonical: https://daily.dev/sources/seangoedecke/best-of/2026/07 twitter:card: summary_large_image twitter:site: @dailydotdev og:url: https://daily.dev/sources/seangoedecke/best-of/2026/07 og:type: website og:site_name: daily.dev og:title: Best sean goedecke posts — July 2026 | daily.dev og:description: The most upvoted sean goedecke posts from July 2026, curated by the daily.dev community. og:image: https://media.daily.dev/image/upload/s--VAY5ToZt--/f_auto/v1724209435/public/daily.dev%20-%20open%20graph --- # Best of sean goedecke — July 2026 1. 1 [](https://daily.dev/posts/in-defense-of-not-understanding-your-codebase-kkk0ijcml "In defense of not understanding your codebase") Article sean goedecke · 9w In defense of not understanding your codebase A defense of working with partial codebase understanding in large software systems, arguing against Peter Naur's 'Programming as Theory Building' thesis. The author contends that Naur's recommendation to scrap and rewrite programs when team knowledge is lost is impractical at scale — large systems can't be rebuilt from scratch, and abandoned codebases are regularly revived. In modern large codebases, everyone operates with an incomplete mental model, and effective engineers must make confident decisions despite that uncertainty. The post also frames 'maintaining a theory of the codebase' as just one engineering value among many, to be traded off against speed, legal requirements, and organizational realities — not treated as an absolute prerequisite. 34 4 2. 2 [](https://daily.dev/posts/llms-reward-expertise-ssrxi8jij "LLMs reward expertise") Article sean goedecke · 7w LLMs reward expertise Domain expertise is the most important factor in getting value from LLMs, not generic prompting tricks. Using Terence Tao's ChatGPT conversation about the Jacobian Conjecture as an example, the author shows that experts extract far more from models because they can steer conversations, identify wrong answers, and suggest better approaches. The same principle applies to software development: deep familiarity with a codebase lets developers push LLMs much harder than novices can. This suggests human expertise remains valuable even as models improve, because the bottleneck is often the human's ability to communicate precisely what solution they want. 19 2 3. 3 [](https://daily.dev/posts/overtraining-as-the-path-to-human-like-ai-exvvztsf4 "Overtraining as the path to human-like AI") Article sean goedecke · 8w Overtraining as the path to human-like AI Gwern's 13,000-word essay 'Human-like Neural Nets by Catapulting' proposes that LLMs fail to achieve human-like generalization because they haven't 'grokked' their training data. Grokking — a phenomenon where continued overtraining past apparent convergence causes a sudden leap in generalization — requires training a very large, over-parameterized model on a relatively small dataset. This is the opposite of what frontier AI labs currently do (training smaller models on massive datasets). Gwern argues that training a \~100-trillion-parameter model on constrained data, at a cost of $3–10B, could force the model to discover deeper generalizations rather than relying on memorization. The author summarizes and evaluates this argument, noting the political and engineering obstacles, and expresses cautious optimism that a major lab should attempt it. 14 5 [See all sean goedecke archives](/sources/seangoedecke/best-of) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@graph":[{"@type":"CollectionPage","@id":"https://daily.dev/sources/seangoedecke/best-of/2026/07#page","url":"https://daily.dev/sources/seangoedecke/best-of/2026/07","name":"Best sean goedecke Posts — July 2026","description":"The most upvoted sean goedecke posts from July 2026, curated by the daily.dev community.","isPartOf":{"@type":"WebSite","url":"https://daily.dev"}},{"@type":"ItemList","@id":"https://daily.dev/sources/seangoedecke/best-of/2026/07#items","numberOfItems":3,"itemListElement":[{"@type":"ListItem","position":1,"url":"https://daily.dev/posts/in-defense-of-not-understanding-your-codebase-kkk0ijcml","name":"In defense of not understanding your codebase"},{"@type":"ListItem","position":2,"url":"https://daily.dev/posts/llms-reward-expertise-ssrxi8jij","name":"LLMs reward expertise"},{"@type":"ListItem","position":3,"url":"https://daily.dev/posts/overtraining-as-the-path-to-human-like-ai-exvvztsf4","name":"Overtraining as the path to human-like AI"}]},{"@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Sources","item":"https://daily.dev/sources"},{"@type":"ListItem","position":3,"name":"sean goedecke","item":"https://daily.dev/sources/seangoedecke"},{"@type":"ListItem","position":4,"name":"Best of","item":"https://daily.dev/sources/seangoedecke/best-of"},{"@type":"ListItem","position":5,"name":"July 2026"}]}]} ```