Construction Technology Podcast
This episode covers Nomic's pivot to verticalized AI for construction, and the broader question of how to build useful AI products in AEC. The conversation works through two paths: training custom models, or harnessing foundational ones. We get into Rich Sutton's Bitter Lesson, why the harness around the model does most of the heavy lifting in construction use cases, how to benchmark the full harness, the change management work that determines whether a rollout sticks, and why outcome-based billing becomes more plausible as these systems embed deeper into preconstruction workflows. Takeaways * Domain-specific harnesses drive most of the performance gains in verticalized AI * Benchmarks should measure the full harness, including retrieval, prompting, and tool use * Change management is typically the binding constraint on AI adoption in construction * Outcome-based pricing becomes viable once AI owns the deliverable end to end * The biggest win for estimators and PMs is reduced mental load on repetitive review work Chapters * 00:00 Introduction and pivot to AEC * 04:58 Debate on LLMs in construction * 10:50 Building custom models * 15:57 Harnessing foundational models * 21:01 Rich Sutton's Bitter Lesson * 26:01 Scaling compute and harnesses * 31:56 Why the harness matters * 41:50 Efficiency and time savings * 56:28 Impact on professional workflows
12 episodios
Comentarios
0Sé la primera persona en comentar
¡Regístrate ahora y forma parte de la comunidad de Construction Technology Podcast!