Open Frontier AI with Andy Konwinski @ NeurIPS 2025 - Agents, AI Training Stacks, + What Comes Next скачать в хорошем качестве

Open Frontier AI with Andy Konwinski @ NeurIPS 2025 - Agents, AI Training Stacks, + What Comes Next

Day 3 brings the demos all the way up the stack, from agent optimization and training infrastructure to fully open model development, and closes with Laude's Open Frontier announcement with Andy Konwinski. Recorded live at Laude Lounge during NeurIPS, this session features six projects working across pre-training, post-training, agents, reinforcement learning, and open model workflows, all with a shared goal: making advanced AI systems buildable, inspectable, and improvable in the open. Featured demos: DSPy — Programmatic LLM systems, prompt + weight optimization, and compiler-style abstractions for reliable AI software. OpenThoughts — An open data and training pipeline for agentic models, from SFT and RL to end-to-end evaluation. GEPA — A highly sample-efficient evolutionary optimizer that learns from rich textual feedback, not just scalar rewards. Harbor — Scalable sandboxed environments for agent rollouts, evaluation, and post-training in the cloud. SkyRL — A modular RL training stack designed for long-horizon, multi-turn agent workloads across heterogeneous hardware. OLMo 3 — Fully open model flows, releasing not just weights but the entire training lifecycle. Marin — An open-source LLM project exposing its development process in real time, enabling community contribution at every stage. The session closes with a call to action: AI’s next phase can’t be built behind closed doors. Open research, open weights, and open collaboration are how we move the frontier forward, together. X: https://x.com/LaudeInstitute LinkedIn:   / laude-institute   --- 0:00 — Andy: The Open Frontier AI Stack 2:06 — DSPy Demo: Engineering Reliable AI Systems 10:52 — OpenThoughts Agent Demo: Open Data Pipeline for Training Agents 18:59 — GEPA Demo: Reflective Evolution for Compound AI Systems 36:33 — Harbor Demo: Agent Evaluation Environments at Scale 44:02 — SkyRL Demo: Modular RL Infrastructure for Agents 52:58 — OLMo 3 Demo: Releasing the Full Model Flow 1:08:12 — Marin Demo: Building Models in Public 1:11:34 — Closing: Open Frontier Announcement --- Hosted by Andy Konwinski Creative Producer - Mike Maley Production Manager - Lauren Lukow Videographer - Andrew James Benson Assistant Camera - Bradley Smith Senior Video Editor / Graphics - Cai Lee Editors - Jordan Calig, Juan Diego Parra Audio Editor - Carter Wogahn Produced by K. Tighe, Kayleigh Karutis, and Chris Rytting. Produced by Laude in partnership with Pod People. -- Tyler Griggs, SkyRL https://x.com/tyler_griggs_ Hanna Hajishirzi, OLMo 3 https://x.com/HannaHajishirzi Kaiyue Wen, Marin https://x.com/wen_kaiyue Michael Ryan, DSPy   / michael-ryan-207   Etash Guha, OpenThoughts https://x.com/etash_guha Negin Raoof, OpenThoughts https://x.com/NeginRaoof_ Ben Feuer, OpenThoughts   / benjaminfeuer   Ryan Marten, OpenThoughts   / ryan-marten   Lakshya Agrawal, GEPA https://x.com/LakshyAAAgrawal -- #artificialintelligence #opensource #airesearch #machinelearning #airesearch #ai #computerscience #OpenAI #OpenSourceAI #AIResearch #NeurIPS #LLMs #AIInfrastructure #AIEngineering #Agents #OpenScience #openfrontier #AgenticAI #ReinforcementLearning #modeltraining