以下部分引用 AI 总结,现阶段 AI 仍然有幻觉。请以内容的原文为准。
数据新鲜度群聊 07-14 ✓卖方 07-14 ✓Wrap 断档15天 ✕News 07-15 ✓X 断档12天 ✕页面生成 07-15
PRISM
4 atoms · 跨 2 天 · 首见 2026-05-19 · 最近 2026-06-22
三色: 🟦 fact 4 · 🟥 take 0 · stance ▲0/▼0/◆3
来源: X 4
时态: fresh:4
标签: 好信源:2 · 好数字:2 · 好思考:1
事实 (fact) (4)
- 2026-05-19 · matches best-of-N performance with far fewer function evaluations on math reasoning and code generation benchmarks
On math reasoning and code generation benchmarks, PRISM matches best-of-N performance with far fewer function evaluations.
tags: 好数字·好信源 · → daily
- 2026-06-22 · sequence model 速度提升
PRISM matches the accuracy of powerful iterative methods but runs 174x faster in throughput
tags: 好数字 · → daily
value: qty=174x · direction=faster
- 2026-05-19 · test-time scaling method for discrete diffusion language models
Researchers from NUS, Georgia Tech, and other institutions present PRISM — a test-time scaling method for discrete diffusion language models (dLLMs).
tags: 好信源 · → daily
- 2026-05-19 · uses hierarchical search, local branching, and self-verification
It uses hierarchical search to prune and reallocate compute early, local branching to explore diverse outputs while keeping confident tokens, and self-verification instead of external checks.
tags: 好思考 · → daily
时间轴 (近 20)
- 2026-06-22 ·
fact · sequence model 速度提升 · → daily
- 2026-05-19 ·
fact · test-time scaling method for discrete diffusion language mod · → daily
- 2026-05-19 ·
fact · uses hierarchical search, local branching, and self-verifica · → daily
- 2026-05-19 ·
fact · matches best-of-N performance with far fewer function evalua · → daily
← 实体目录 · 系统日志