Must-read papers of the week: recursive self-improvement, world models, KV cache compression

TheTuringPost · x · 2026-09-22

The Turing Post rounds up the week's notable papers:

• Recursive self-improvement: Dream-RSI (self-improvement via evolving worlds), ModularRSI (modular generalizable harness self-improvement), SoL-Pi (recursively scaling auto-research loops), and ScienceBuddy (interactive scientific agents)
• Models & architectures: JEPA-Anything, Modality-Autoregressive World-Action Models, DeepSeek-V4.1-Flash (pushing KV cache compression limits), Video DeltaNet (video-native hybrid attention for livestream generation), plus the AliceAI-Foundation-80B-A3B-Base model release
• Agents & efficiency: In-context robot learning with VLM agents, experiential confidence estimation from reasoning to agents, and When2Think (difficulty-aware length control for hybrid reasoning)
• Theory: world modeling in transformers

Original post →

More from Research

• NVIDIA Open-Sources VoiceChat, a Full-Duplex Speech-to-Speech Model with Tool Calling — chaumian · 2026-09-22
• SteerDuplex: full-duplex speech model gains 44.5pp in steerability, new SteerBench released — ScaleAI · 2026-09-22
• Sarah Hooker suspects fake AI paper submissions cluster at a few universities — sarahookr · 2026-09-22
• Rollout scheduling: the underappreciated infra trick boosting inference efficiency — stochasticchasm · 2026-09-22
• Benchling benchmarks Claude and ChatGPT on wet-lab protocols: helpful, not solved — nlarusstone · 2026-09-22
• LLMs as probability-guided search in token space: data coverage gaps are the real weakness — AlexTensor · 2026-09-22

Research channel →