Sean Goedecke is a staff software engineer at GitHub.
How media typically covers Sean Goedecke
Based on 10 scored articles
Sean Goedecke as author
Language models use em-dashes at unusually high frequencies compared to human writing, but the underlying cause remains unexplained despite common theories about training data, tokenization efficiency, and RLHF practices.
“Author of "Why Do AI Models Use So Many Em-Dashes?"”
AI research is currently in an early scientific era where important discoveries can be made by non-specialists, similar to the 'gentleman scientist' period of classical science, because the core insights are often simpler than their mathematical presentation suggests.
“Author of "We are in the "gentleman scientist" era of AI research" in Sean Goedecke”
OpenAI's Responses API is designed to hide reasoning traces from clients while maintaining them server-side, allowing GPT-5-Thinking to match competitor capabilities without exposing its chain-of-thought.
“Author of "The Whole Point of OpenAI's Responses API is to Help Them Hide Reasoning Traces" in Sean Goedecke's Blog”
Chain-of-thought reasoning in LLMs may rely on memorized patterns rather than true logical inference, becoming fragile when faced with distribution shifts or compositional generalization challenges.
“Author of "Is Chain-of-Thought AI Reasoning a Mirage?" in Sean Goedecke's Blog”
Referenced in coverage
“Author analyzing GPT-6 Astra's ability to work with neural network weights and providing technical critique.”
“Developed Deckard, a Chrome extension for locally-running AI text detection models to automatically scan text on websites.”
“Author discussing how software engineers must differentiate themselves from AI models by leveraging deep codebase knowledge.”
“Author writing about how LLMs reward expertise and the importance of domain knowledge in AI.”
“Author of analysis examining claims that AI GPUs only last three years, tracing the claim to dubious sources and providing evidence for longer lifespans.”
“Author describing how LLMs and AI agents have transformed his software engineering workflow, using them to write entire code changes and investigate bugs.”
“Author discussing LLM steering concepts and DeepSeek-V4-Flash capabilities for local agentic coding.”