New class of long-context tasks scales quadratically with corpus size, upending LCLM design choices
xiye_nlp · x · 2026-09-29
Researcher Prasann Singhal shares a thread asking whether long-context architectures can "find all contradictions" in science literature — not yet.
- The team defines a new class of "high-complexity" tasks whose difficulty scales quadratically with corpus size, rather than linearly
- This reverses several common LCLM design decisions, including block-sparse attention and hybrid model architectures
- Core insight: exhaustively surfacing contradictions across massive corpora is inherently a combinatorial comparison task that current long-context architectures aren't built for
More from Research
- Hexagon, a New Repository for Math and TCS Research, Enters Public Beta — suchenzang · 2026-09-29
- JHU's Harang Ju to present HCOMP talk on the jagged frontier in human-AI collaboration — windx0303 · 2026-09-29
- LayerSkip: Self-Speculative Decoding Speeds Up LLMs Without a Draft Model — burkov · 2026-09-29
- PolyXOR takes 128-bit hash speed crown; new work shows adversarial collisions in fast hashes — thomasahle · 2026-09-29
- Bio-data hackathon with Anthropic and AWS: find new biological insights in 48 hours — adityaag · 2026-09-29
- Hindsight: Letting Your Agent Learn Without Breaking Policy — nishithreddy · 2026-09-29