NYU CDS research: AdaJEPA world model keeps learning post-training, plus Anthropic introspection rebuttal
mengyer · x · 2026-11-04
NYU CDS's November research feature highlights:
- AdaJEPA: an adaptive latent world model — Ying Wang, with Oumayma Bounou, Yann LeCun, and Mengye Ren, built a world model that adapts during planning when deployment conditions mismatch training (changed lighting, unexpected object weights), addressing robots failing after millions of practice steps in one environment.
- Can VLMs track stroke rehab from video? — Victor Li and Carlos Fernandez-Granda's team tested whether vision-language models can extract clinical information (hand movements, object interactions) directly from rehabilitation videos; even sophisticated models struggle.
- Challenging Anthropic's introspection evidence — Shashwat Singh, Tal Linzen, and Shauli Ravfogel contest Anthropic's 2025 claim that Claude's ability to detect injected concepts constitutes "emergent introspective awareness."
More from Research
- New Thread Tackles Long Video Generation Drift: Let Every Frame See the Whole Clip — mangahomanga · 2026-11-04
- TLAPS-Bench: 265 real-world TLA+ proof targets; top model GPT-6.1 Sol solves only 44.3% — tianyin_xu · 2026-11-04
- StreamDuo pairs a commentator with a wiki research agent to commentate unseen games — Zixian Gao · 2026-11-04
- Redwood researcher: AI research automation and runaway recursive self-improvement can come apart — deanwball · 2026-11-04
- UMAP 0.6 adds Adam and momentum layout optimizers, becoming defaults in 0.7 — leland_mcinnes · 2026-11-04
- UMAP 0.6 preview release candidate lands on PyPI with major changes — leland_mcinnes · 2026-11-04