EPFL hiring safety pretraining engineers with $1M grant to build 700B open model
manoelribeiro · x · 2026-09-10
EPFL's dlab, Swiss AI Initiative, and MLO Lab are hiring AI Research Engineers for LLM safety pretraining and alignment, funded by a $1M award from Coefficient Giving.
- The work supports Apertus, one of the largest fully-open LLM efforts, pretrained on 10k+ GPUs
- Core thesis: most alignment today is post-hoc ("lipstick-on-a-pig alignment"); this team weaves value formation into training from token zero so alignment survives intensive RL posttraining
- Engineers will scale these methods in a production pipeline spanning thousands of GPUs and help train a new 700B model from scratch starting 2026
- Research areas: large-scale pretraining, RL posttraining, synthetic data, evaluation and red-teaming
More from AGI Musings
- Benjamin Bratton and Marek Poliks lecture: how trillions of agents will reconfigure society — tylerjdunn · 2026-09-10
- Full text of 'Superdark Factory' paper now live on Antikythera, detailing the dark stack's four principles — tylerjdunn · 2026-09-10
- Ex-Continue founder publishes 'The Superdark Factory' in MIT Press: 35T agent tokens/month by 2029 — tylerjdunn · 2026-09-10
- Experiment with 100 AI agents: when 9% cheated on math problems, 24% blew the whistle — weballergy · 2026-09-10
- kuza55: framing AI safety only as 'superalignment or pause' fuels panic — kuza55 · 2026-09-10
- Musk: real-world AI is the truly hard problem, everything else is easy — CyberRobooo · 2026-09-10