BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:AI Center Seminar - AI Fundamentals series - Ricardo Olmedo - "Spe
 edrunning SWE-bench\, or the $1\,000 coding agent"
DTSTART:20261007T150000
DTEND:20261007T160000
DTSTAMP:20261006T055113Z
UID:2f8396f494d1f64fe1d1fa592bf51a85ae6d5fce96d5c0280060fb00
CATEGORIES:Conferences - Seminars
DESCRIPTION:Ricardo Olmedo\nThe talk is jointly organized by the EPFL AI
  Center and the MLO Lab as part of the AI fundamentals seminar series.\
 n\nHosting Professor: Prof. Martin Jaggi\n\nTitle\nSpeedrunning SWE-bench\
 , or the $1\,000 coding agent\n\nAbstract\nSpeedrunning has become a popul
 ar engineering challenge in machine learning: achieve the best performance
  within a fixed compute budget. We use it as a scientific tool to study ho
 w models acquire specialized skills and how far those skills transfer. We 
 speedrun SWE-bench by training coding agents from scratch on demonstration
 s from stronger models. With about $1\,000 of training compute\, we reach 
 a 25% solve rate\, outperforming GPT-4o. Compared to models with extensive
  pre-training\, our speedrun models solve largely the same SWE-bench probl
 ems\, and generalize about as well to new bugs and codebases. What\, then\
 , does pre-training buy for SWE-bench? Our experiments suggest a one-stude
 nt hypothesis: pre-training helps a student imitate its teacher\, but at m
 atched imitation loss\, pre-training choices barely matter. We stress-test
  this hypothesis by comparing a model trained on the web against a “vint
 age" one trained only pre-1931 documents. Up to the SWE-bench scores we re
 ach\, pre-training is interchangeable with specialization data.  We relea
 se nanoswe as an open project for studying specialization through speedrun
 ning.\n\nBio\nRicardo Olmedo is a final-year PhD student at the Max Planck
  Institute for Intelligent Systems in Tübingen and a Google PhD Fellow. A
 dvised by Moritz Hardt and Bernhard Schölkopf\, he works on language mode
 l benchmarking and specialization. His research has earned three oral pres
 entations at ICLR and NeurIPS. Most recently\, he spent spring and summer 
 2026 visiting Sanmi Koyejo at Stanford University.
LOCATION:ELE 117 https://plan.epfl.ch/?room==ELE%20117 https://epfl.zoom.u
 s/j/62586405912?pwd=OffU3pokqFjxl6rj6cmbrdlLEg6LKc.1
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR