Category: research
-
MiniMax Sparse Attention: Teaching Long-Context Models to Use an Index
MiniMax Sparse Attention turns long context into searchable memory: a learned index selects relevant key-value blocks, then exact softmax attention reads only those blocks.
-
Self-Harness: Harnesses That Improve Themselves
Self-Harness is best read as evidence-driven harness versioning: mine failed traces, propose bounded scaffold edits, and promote only changes that survive regression tests.
-
Pi Agent: containerization and compaction
Coding agents need two boundaries at the same time: an execution boundary that controls what they can do, and a context boundary that controls what they can remember across long work.
-
Yuandong Tian talks: search quality is action-space quality
TLDR: More rollouts are not enough. Search becomes powerful when the action space, representation, evaluator, and memory make good trajectories easier to find.
-
Concordia: LLM agents as social simulation actors
Concordia is useful because it treats LLM agents as situated social actors with memory, roles, norms, partial observations, and a world state mediated by a Game Master.
-
Autocurricula and Multi-Agent Innovation: 社会互动如何生成新问题
TLDR: Multi-agent intelligence should study how cooperation, competition, specialization, and shared discoveries create abilities that isolated agents would miss.
-
Social Dilemmas: 三个经典社会困境
TLDR: Social dilemmas show why individually rational actions can damage group outcomes, and why cooperation depends on payoffs, repetition, reputation, and norms.
-
A Social Path to Human-Like AI: 社会互动如何生成新数据
TLDR: Human-like AI may require populations of agents learning through social interaction, where cooperation and competition generate skills beyond single-agent training.