feat(daily): 2026-04-16 digest (#9)

https://claude.ai/code/session_01CETxAEEzLk6jFPHoTWEKSe

Co-authored-by: Claude <noreply@anthropic.com>

Author: yayajjiang

src/lib/daily.ts

@@ -17,6 +17,25 @@ export interface DailyPaper {
 
 export const dailyPapers: DailyPaper[] = [
   {
+    date: "2026-04-16",
+    title: "Introspective Diffusion Language Models",
+    titleZh: "内省扩散语言模型（I-DLM）",
+    authors: "Yifan Yu et al.",
+    arxivId: "2604.11035",
+    tags: ["Diffusion LM", "Theory"],
+    why: "Identifies introspective inconsistency as the root quality gap in diffusion LMs; ISD verifies past tokens while advancing new ones in one forward pass.",
+    whyZh: "揭示内省不一致性为扩散LM质量差距根源，ISD算法单次前向传播同步验证历史token并生成新token。",
+    pick: true,
+  },
+  {
+    date: "2026-04-16",
+    title: "Accelerating Speculative Decoding with Block Diffusion Draft Trees",
+    titleZh: "利用块扩散草稿树加速推测解码",
+    authors: "Liran Ringel and Yaniv Romano",
+    arxivId: "2604.12989",
+    tags: ["Diffusion LM", "Efficient Inference"],
+    why: "DDTree builds a best-first draft tree from block diffusion per-position distributions — SOTA speculative decoding verified in one target model forward pass.",
+    whyZh: "DDTree从块扩散逐位置分布构建最优优先草稿树，单次目标模型前向传播完成验证，达到推测解码SOTA。",
     date: "2026-04-15",
     title: "Introspective Diffusion Language Models",
     titleZh: "内省扩散语言模型",
@@ -69,6 +88,14 @@ export const dailyPapers: DailyPaper[] = [
     whyZh: "发现模板坍塌这一被熵指标忽视的推理失效模式，提出信噪比滤波策略有效缓解。",
   },
   {
+    date: "2026-04-14",
+    title: "DFlash: Block Diffusion for Flash Speculative Decoding",
+    titleZh: "DFlash：面向快速推测解码的块扩散方法",
+    authors: "Jian Chen et al.",
+    arxivId: "2602.06036",
+    tags: ["Diffusion LM", "Efficient Inference"],
+    why: "Block diffusion drafter achieves 6x lossless speedup over base LLM — 2.5x faster than EAGLE-3 with no quality loss.",
+    whyZh: "用块扩散模型作为推测解码草稿器，实现6倍无损加速，比EAGLE-3快2.5倍。",
     date: "2026-04-12",
     title: "SUPERNOVA: Eliciting General Reasoning in LLMs with Reinforcement Learning on Natural Instructions",
     titleZh: "SUPERNOVA：利用自然指令强化学习激发LLM通用推理能力",