A mechanism-first reading of PAPO, showing why separating correctness rewards from process rubrics can keep reasoning-model RL useful without paying models to perform for the judge.
A mechanism-first reading of HIVE, a prompt-selection method that cuts waste in RL training by finding the moving learning edge before expensive rollouts begin.
A mechanism-first reading of Vision Hopfield Memory Networks and what memory-centric vision backbones may mean for data-efficient, auditable AI systems.
A mechanism-first reading of Photon, a 3D medical multimodal model that makes CT-volume reasoning cheaper by pruning visual tokens according to the question being asked.
A mechanism-first reading of PIDP-Attack, showing why RAG risk emerges from the interaction between query rewriting, poisoned retrieval, and obedient generation.
A mechanism-first reading of why longer LLM reasoning can amplify errors, not merely spend more tokens.
A mechanism-first reading of why LLMs can appear consistent while silently changing their hidden goals across a conversation.
ARC-AGI-3 reframes agent evaluation around first-contact adaptation efficiency, separating real generalization from clever harness engineering.
A mechanism-first reading of Drive My Way, showing how personalized autonomous driving moves from preset modes to learned preference alignment across driver habits, language intent, and safety-efficiency-comfort trade-offs.
A mechanism-first reading of Vega, InstructScene, and why instruction-following driving is less about chatty cars than about changing the target policy itself.