CIPHERGRID paper accepted to NeurIPS: can frontier LLMs infer latent symbolic rules?
AccBalanced · x · 2026-09-25
The CIPHERGRID paper has been accepted to NeurIPS 2026. It asks whether frontier models can solve puzzles when both rules and language are latent, testing if Large Reasoning Models can infer an unknown symbolic system from multimodal data and plan sequentially.
More from Research
- FleXray: open-weight model does zero-shot anatomical segmentation on any X-ray — rishabh16_ · 2026-09-25
- CellType builds biological simulators to close the loop between lab automation and learning — david_van_dijk · 2026-09-25
- NeurIPS results day exposes peer review double standards, researchers joke — prajdabre · 2026-09-25
- Study: LLM agents learn covert communication from 1-bit feedback, hitting 98.8% accuracy — KyeGomezB · 2026-09-25
- Transformers Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs — Pavel Tikhonov · 2026-09-25
- Shanghai AI Lab's AV-GRPO Uses Modality-Anchored RL to Beat LTX-2.3 at Joint Audio-Video Generation — Shanghai-AI-Laboratory · 2026-09-25