JetBrains Research Podcast
In our second episode, we chat with Ibragim Badertdinov from Nebius about SWE-rebench, coding agents, and his path from dentistry and AI. Our first episode can be found here [https://youtu.be/W8wFQjKhb7s] Episode links: * SWE-rebench: https://swe-rebench.com/ [https://swe-rebench.com/] * ConTree: https://contree.dev/ [https://contree.dev/] * Ibragim: https://x.com/ibragim_bad [https://x.com/ibragim_bad] Also mentioned: * https://huggingface.co/datasets/nebius/SWE-bench-extra [https://huggingface.co/datasets/nebius/SWE-bench-extra] * https://scale.com/leaderboard/swe_bench_pro_public [https://scale.com/leaderboard/swe_bench_pro_public] * https://swe-bench-live.github.io/ [https://swe-bench-live.github.io/] Find out more about JetBrains Research: https://lp.jetbrains.com/research/software-engineering/ [https://lp.jetbrains.com/research/software-engineering/] Chapters: 00:00:00 Teaser & Introduction 00:02:41 Ibrahim's Non-Traditional Background: From Dentistry to Tech 00:10:45 The "School 42" Bootcamp Experience 00:13:00 Getting into Machine Learning & Kaggle 00:17:20 The First Internship & Learning from a Great Mentor 00:20:10 Contrasting the Worlds of Medicine and Tech 00:26:56 What Are Coding Agents? 00:33:40 Explaining SWE-bench: How Coding Tasks Are Evaluated 00:37:15 RLVR (Reinforcement Learning with Verifiable Rewards) 00:43:10 The Creation of SWE-Rebench 00:50:00 The Main Challenges with Task Quality and Scale 00:55:00 What's New in SWE-Rebench V2 (and Kotlin support) 00:58:42 Will AI Change Code Review? 01:02:00 contree.dev [http://contree.dev]: A New Checkpointing & Forking Tool for Agents 01:09:35 The Open-Source Community's Reaction to SWE-Rebench 01:15:20 SWE-Bench Pro vs. SWE-Rebench 01:18:20 Why Nebius Invests in Open-Source Research 01:21:00 Predictions for the Future of AI and Programming 01:26:00 Final Advice for Career Switchers
3 episodios
Comentarios
0Sé la primera persona en comentar
¡Regístrate ahora y únete a la comunidad de JetBrains Research Podcast!