Vibe NLP for Applied NLP
summary
Claims a concrete workflow: use LLMs to write reusable scripts and training data so runtime inference uses small private classifiers that are faster and cheaper; a new framework called Elf integrates with existing coding assistants, generates UIs from Python recipes, and orchestrates tasks on user‑controlled clusters (Kubernetes) so sensitive data never leaves the environment. It demonstrates auto‑annotating agents that create and validate datasets, LLMs used to build systems not replace them, with measurable wins in latency, cost, and reproducibility.
AI-written summary · about
chapters (10)
- 0:00Vibe NLP intro and challenges
- 3:29LLMs not enough for real NLP
- 6:39Systems with LLMs vs relying on them
- 9:50Agentic workflows in Software 2.0
- 12:22Elf: Explosion LLM Framework
- 15:55Scalable NLP workflows infra
- 18:42Human-agent data collaboration
- 21:29Developer APIs in agentic systems
- 23:59Future of supervised learning
- 25:26Key takeaways
related talks
38mNobody Pages the LLM: Engineering Rigour for Vibe Coding
26mAI-powered Productivity using Open-Source LLMs
47mOwn Your LLM: Running Generative and Agentic AI Locally
31mLeveraging LLMs to build supervised datasets suitable for smaller models
47mBuilding Python based AI Systems with LLMs
41mBuilding LLM Agents – Challenges and Learnings
43mAI Development Lifecycle: Learnings of What Changed with LLMs
36mAI Assistant Behind the Firewall: Keeping the LLM + Frontend = Velocity Formula Without Exposing Code