Undergrad-led team lands 4 NeurIPS papers, including hybrid neural world models with 26x-72x speedups
paraschopra · x · 2026-09-25
Paras Chopra announced that LossFunc got 4 papers into the NeurIPS main conference, with all first authors being undergraduates:
- Hybrid Neural World Models: a single horizon-conditioned network trained against reference solvers predicts any future state in one forward pass, implicitly encoding discontinuity locations as an error map concentrated on shocks, fronts and contacts. It matches or beats deep ensembles and other label-free baselines without a calibration set, delivers 26x-72x CPU speedups on PDE environments, and can gate reference-solver fallback to roughly halve residual error.
- Frontier Coding Agents Use Metaprogramming to Adapt to Unfamiliar Programming Languages
- EsoLang-Bench: evaluating genuine LLM reasoning via esoteric programming languages
- Verified-Source Authority Is Not Generic Sycophancy: cue-family decomposition of LLM compliance
Related event: LossFunc Lands 4 NeurIPS Papers, All First-Authored by Undergrads(2 posts)→
More from Research
- GRAM accepted to NeurIPS 2026: recursive reasoning becomes stochastic latent trajectories — SungjinAhn_ · 2026-09-25
- Photonic matrix core on thin-film lithium niobate runs in-situ backpropagation at 8-bit precision — jwt0625 · 2026-09-25
- Agentic robotics keeps bolting GPT-6 onto harnesses — but where are systems that learn from physical failure? — DJiafei · 2026-09-25
- Dev Proposes "Evidentiality Framework": Provenance Markers for Every LLM Claim — jzesbaugh · 2026-09-25
- New Paper Quantifies When Rubric-Based RL Genuinely Improves Models vs. Hacks the Verifier — heghbalz · 2026-09-25
- DeepSeek's DSec sandbox platform serves 3M sandboxes daily for agent RL training; Liang Wenfeng co-authors — teortaxesTex · 2026-09-25