Concept detail
Context compaction
0 tagged videos · 2 connection videos · degree 0
Publication timeline
- 2026-092 videos
Source videos
Observed neighbors
Coherence over long contexts
Compaction is the mechanism that trades long-context capability against instruction retention, making it the operational knob for the long-context coherence sub-capability.
1 source videos
CMU AI Agents 2026: 1. What are Agents and How Do They Work?KV prefix caching
Both attack the same prefill cost term at different layers — application-level state compression versus engine-level prefix reuse.
1 source videos
CMU AI Agents 2026: 3. Long Context Modeling for AgentsRelated tagged insights
No core insights are tagged with this concept.