CMU's new Looni Lab targets long-horizon, out-of-distribution, non-verifiable intelligence
niloofar_mire · x · 2026-11-12
Niloofar Mireshghallah launches Looni Lab at CMU (Long-horizon, Out-Of-distribution, Non-verifiable Intelligence) with three pillars:
- Privacy & social intelligence: outcome- and simulation-based privacy, mental health + AI, physical-world privacy, with collaborators at CMU, KAIST, USC, Stanford.
- AI for science & chemistry: long-horizon experience learning, memorization vs exploration, dual use, 3D structural reasoning, partnering with Brookhaven and others.
- AI security & verification: verifying workloads via hardware side channels, internal commitment points, distillation/lineage provenance, poisoned reasoning traces.
The lab publishes open problem lists and its founding members.
More from Research
- François Fleuret asks: are formal proof checkers actually unused in RL training? — francoisfleuret · 2026-11-12
- AGI debate: pretraining won because the economy is PDFs, but LLMs still can't learn — akbirthko · 2026-11-12
- AudioCodex: fully causal streaming autoencoder compresses stereo audio to ~10Hz latents for live generation — serrjoa · 2026-11-12
- In-context learning is gated lookup of prompt-stored info, not rule fitting, mechanism study finds — burkov · 2026-11-12
- CogBattery benchmarks 54 open-weight LLMs against humans on 13 cognitive paradigms — Dengzhe Hou · 2026-11-12
- PISA: Prompt-Adaptive Cost-Aware LLM Routing Tops Four Routing Benchmarks — Saul Urso · 2026-11-12