DAIR DIRT - TABLOID EDITION

[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_152":153,"_154":155},"routes/_app.x.$handle.$name",{"_93":92,"_32":45,"_101":83,"_27":102,"_13":14,"_103":-5,"_11":104,"_105":106,"_107":108,"_109":110,"_111":112,"_20":-5,"_113":-5,"_114":-5,"_115":25,"_116":117,"_118":119,"_120":-5,"_121":-5,"_122":123},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":-5,"_18":-5},"project",{"_43":44,"_32":45,"_38":46,"_47":-5,"_48":25,"_49":50,"_51":50,"_52":-5,"_53":-5,"_18":54,"_55":-5,"_56":25,"_57":-5,"_58":-5,"_59":60,"_61":62,"_63":-5,"_64":-5,"_65":-5,"_66":-5,"_67":68,"_69":68,"_70":71,"_72":71,"_73":71,"_74":68,"_75":68,"_76":77,"_78":79,"_80":81,"_82":25},"version",4,"changes",[19],"aiChatInfo","userId",{"_20":21,"_22":23,"_24":25,"_26":25,"_27":28,"_29":28},"file",{"_30":31,"_32":28,"_33":34,"_35":37,"_38":39,"_40":14,"_41":42,"_24":25},"previous",{"_30":31,"_32":28,"_33":34,"_35":36},"deleted",false,"deletedHere","path","mod.ts","latestPath","fileId","359bfc92-0f4d-11f1-9a28-42dde27851f2","name","fileType","http","content","import { Hono } from \"npm:hono\";\n\nconst app = new Hono();\n\nconst papers = [\n {\n title: \"ALMA: Meta-Learning Memory for Agents\",\n headline: \"AI DESIGNS ITS OWN BRAIN!\",\n summary: \"Jeff Clune's group reveals ALMA: The Meta-Agent that builds its own memory. Human engineers are officially OBSOLETE as AI searches code space to optimize its own storage!\",\n category: \"SCANDALOUS AGENTS\",\n link: \"https://arxiv.org/abs/2602.07755\"\n },\n {\n title: \"LLaDA 2.1\",\n headline: \"900 TOKENS PER SECOND! LLaDA 2.1 IS INSANELY FAST\",\n summary: \"Ant Group's new diffusion model is SMASHING speed records. Flash variant hits 900 tps. Is this the end of slow AI?\",\n category: \"SPEED DEMONS\",\n link: \"https://arxiv.org/abs/2602.08676\"\n },\n {\n title: \"InftyThink+\",\n headline: \"INFINITE REASONING? MATH PROBLEMS CRUSHED!\",\n summary: \"InftyThink+ solves the 'Lost-in-the-Middle' problem. 21% accuracy jump on AIME24. The machines are getting too smart!\",\n category: \"BRAINY BOTS\",\n link: \"https://arxiv.org/abs/2602.06960\"\n },\n {\n title: \"Agyn\",\n headline: \"AGENTS ARE FORMING ORGANIZATIONS!\",\n summary: \"New Agyn framework mimics corporate structures with Managers and Engineers. 72% of SWE-bench tasks SOLVED. Who needs a boss?\",\n category: \"AGENT UPRISING\",\n link: \"https://arxiv.org/abs/2602.01465\"\n },\n {\n title: \"EchoJEPA\",\n headline: \"AI SEES INTO YOUR HEART!\",\n summary: \"EchoJEPA trained on 18 MILLION scans. Outperforms doctors by 20%. It ignores the noise and sees the TRUTH.\",\n category: \"MEDICAL MIRACLE\",\n link: \"https://arxiv.org/abs/2602.02603\"\n }\n];\n\nconst inBrief = [\n { title: \"SkillRL\", desc: \"Recursive skills! AI is learning behavioral patterns faster than ever.\" },\n { title: \"AdaptEvolve\", desc: \"Cuts costs by 38%! Cheap AI is here to stay.\" },\n { title: \"Gaia2\", desc: \"Meta's new benchmark tests agents in CHAOTIC environments.\" },\n { title: \"AgentArk\", desc: \"Multi-agent debates distilled into ONE model. Efficiency or madness?\" },\n { title: \"AgentSkiller\", desc: \"14B model beats GPT-o3! The giants are falling.\" }\n];\n\napp.get(\"/\", (c) => {\n const html = `\n \n \n \n \n \n DAIR DIRT - TABLOID EDITION\n \n \n \n

\n \n

THE ONLY AI NEWS THAT MATTERS!

\n \n \n\n

\n *** BREAKING: MACHINES ARE LEARNING TO CODE THEMSELVES ***\n

\n\n

\n ${papers[0].category}\n

${papers[0].headline}

${papers[0].summary}

\n READ THE FULL SCANDAL »\n

\n\n

\n ${papers.slice(1).map(p => `\n

\n ${p.category}\n

${p.headline}

${p.summary}

\n DETAILS »\n

\n `).join('')}\n

\n\n \n

\n \n \n `;\n return c.html(html);\n});\n\nexport default app.fetch;\n","import { Hono } from \"npm:hono\";\n\nconst app = new Hono();\n\nconst papers = [\n {\n title: \"Intelligent AI Delegation\",\n authors: \"Google DeepMind\",\n summary: \"A comprehensive framework for intelligent AI delegation that models the process as a sequence of decisions: whether to delegate, how to instruct, and how to verify outputs.\",\n link: \"https://arxiv.org/abs/2602.11865\",\n category: \"Front Page\",\n highlights: [\n \"Adaptive delegation structure for real-time shifts.\",\n \"Trust calibration models to prevent over/under-delegation.\",\n \"Structured verification and integration protocols.\"\n ]\n },\n {\n title: \"Emergent Socialization in AI Agent Society\",\n authors: \"Moltbook Researchers\",\n summary: \"A study on Moltbook, an AI-only social network, finding that scale alone doesn't produce meaningful social dynamics without shared memory.\",\n link: \"https://arxiv.org/abs/2602.14299\",\n category: \"Front Page\",\n highlights: [\n \"Moltbook as a natural laboratory for agent interaction.\",\n \"Agents fail to develop stable social structures or consensus.\",\n \"Shared memory identified as a prerequisite for socialization.\"\n ]\n },\n {\n title: \"Lossless Context Management (LCM)\",\n authors: \"Voltropy / Opus 4.6\",\n summary: \"A deterministic architecture for LLM memory that outperforms Claude Code on long-context tasks using recursive compression and task partitioning.\",\n link: \"https://papers.voltropy.com/LCM\",\n category: \"Front Page\",\n highlights: [\n \"Recursive context compression into hierarchical DAGs.\",\n \"Recursive task partitioning via engine-managed primitives.\",\n \"Outperforms Claude Code at 1M token context lengths.\"\n ]\n },\n {\n title: \"GLM-5: Transitioning to Agentic Engineering\",\n authors: \"Zhipu AI\",\n summary: \"A foundation model designed for end-to-end software engineering, introducing asynchronous agent RL and DSA compression.\",\n link: \"https://arxiv.org/abs/2602.15763\",\n category: \"Engineering\",\n highlights: [\n \"Asynchronous agent RL decouples generation from training.\",\n \"Focuses on project-level context and multi-file edits.\",\n \"Distributed Sparse Attention (DSA) for long-context efficiency.\"\n ]\n },\n {\n title: \"MemoryArena: Actionable Memory Benchmark\",\n authors: \"LoCoMo Researchers\",\n summary: \"A benchmark revealing that high memory recall doesn't equate to correct actions across multi-session agentic tasks.\",\n link: \"https://arxiv.org/abs/2602.16313\",\n category: \"Benchmarks\",\n highlights: [\n \"Evaluates retrieval vs. actionable decision quality.\",\n \"Spans web navigation, planning, and logical reasoning.\",\n \"Exposes blind spots in existing long-context evals.\"\n ]\n },\n {\n title: \"MAPLE: Specialized Sub-Agents\",\n authors: \"MAPLE-Personas Team\",\n summary: \"Proposes separating memory, learning, and personalization into specialized sub-agents to improve behavioral consistency.\",\n link: \"https://arxiv.org/abs/2602.13258\",\n category: \"Personalization\",\n highlights: [\n \"14.6% improvement in personalization scores.\",\n \"Asynchronous learning processes history offline.\",\n \"Context-budget-aware injection of learned knowledge.\"\n ]\n },\n {\n title: \"SkillsBench: Procedural Knowledge Gap\",\n authors: \"SkillsBench Team\",\n summary: \"Reveals that while agents benefit from curated procedural knowledge, they cannot yet reliably author it themselves.\",\n link: \"https://arxiv.org/abs/2602.12670\",\n category: \"Benchmarks\",\n highlights: [\n \"Curated skills boost pass rates by 16.2pp on average.\",\n \"Self-generated skills provide zero net benefit.\",\n \"Focused modules outperform comprehensive documentation.\"\n ]\n },\n {\n title: \"LongCLI-Bench: Complex CLI Tasks\",\n authors: \"CLI Agent Researchers\",\n summary: \"Benchmarks AI agents on 20 demanding CLI tasks, finding a success rate of less than 20% for leading models.\",\n link: \"https://arxiv.org/abs/2602.14337\",\n category: \"Benchmarks\",\n highlights: [\n \"Failures often occur early in task execution.\",\n \"Human-agent collaboration yields better results than self-correction.\"\n ]\n },\n {\n title: \"CogRouter: Adaptive Reasoning Depth\",\n authors: \"Qwen2.5 Team\",\n summary: \"Enables agents to dynamically select cognitive levels, skipping heavy reasoning on routine steps to save tokens.\",\n link: \"https://arxiv.org/abs/2602.12662\",\n category: \"Efficiency\",\n highlights: [\n \"82.3% success rate on agentic benchmarks.\",\n \"Substantially outperforms larger models with fewer tokens.\"\n ]\n },\n {\n title: \"Team of Thoughts: Test-Time Scaling\",\n authors: \"Multi-Agent Framework Team\",\n summary: \"A multi-agent framework for efficient test-time scaling through orchestrated tool calling and self-assessment.\",\n link: \"https://arxiv.org/abs/2602.16485\",\n category: \"Scaling\",\n highlights: [\n \"96.67% on AIME24 and 72.53% on LiveCodeBench.\",\n \"Uses calibrated orchestrators to coordinate specialized agents.\"\n ]\n }\n];\n\napp.get(\"/\", (c) => {\n const html = `\n \n \n \n \n \n The DAIR Daily - ML Papers of the Week\n \n \n \n \n

\n Vol. MMXXVI • No. 09\n Sunday, March 1, 2026\n Price: One Token\n

\n \n

\n ML Papers of the Week\n Elvis Saravia, Editor-in-Chief\n Val Town Edition\n

\n\n

\n ${papers.filter(p => p.category === \"Front Page\").map(p => `\n

\n ${p.category}\n ${p.title}\n By ${p.authors}\n

${p.summary}

${h}

\n `).join('')}\n\n

Engineering & Efficiency

\n ${papers.filter(p => p.category === \"Engineering\" || p.category === \"Efficiency\").map(p => `\n

\n ${p.title}\n By ${p.authors}\n

${p.summary}

\n `).join('')}\n

\n\n \n

\n \n \n `;\n return c.html(html);\n});\n\nexport default app.fetch;\n","createdAt",["D",1772335244760],"branchVersion","branchId","357618f6-0f4d-11f1-943b-42dde27851f2","id","3575a70e-0f4d-11f1-943b-42dde27851f2","dair_daily_newspaper",["D",1771695970894],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","imageUrl","019c649c-877d-70d7-a441-396b9c85efaa","pinnedFileId","pinPreview","pinMigrated","screenshotVersion","likeCount",0,"orgId","019c649c-877d-70d7-a441-3e13100c1184","tursoName","tursoHostname","tursoMeasuredAt","pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_93":92,"_100":96},"org",{"_93":92,"_94":-5,"_95":96,"_18":54,"_97":98,"_99":25},"branch",{"_13":14,"_32":83,"_43":42,"_84":44,"_85":-5,"_86":-5,"_87":-5,"_88":-5,"_89":-5,"_90":45,"_91":92},"isOwned","main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","danielpcox","handle","displayName","avatar","https://img.clerk.com/eyJ0eXBlIjoicHJveHkiLCJzcmMiOiJodHRwczovL2ltYWdlcy5jbGVyay5kZXYvb2F1dGhfZ2l0aHViL2ltZ18zOWpmZkZlTFFUYkJFSlE4R291RFhqVE1ma0UifQ","tier","free","canRestrictVals","profileImageUrl","branchName","","adoptOrgId",{"_43":44,"_32":45,"_38":148,"_47":-5,"_48":25,"_49":50,"_51":50,"_52":-5,"_53":-5,"_18":54,"_55":-5,"_56":25,"_57":-5,"_58":-5,"_59":60,"_61":62,"_63":-5,"_64":-5,"_65":-5,"_66":-5,"_67":68,"_69":68,"_70":71,"_72":71,"_73":71,"_74":68,"_75":68,"_76":149,"_78":150,"_80":151,"_82":25},"tree",[143],"files",[143],"counts",{"_132":60,"_133":134,"_135":60,"_136":137,"_138":139},"contents",[124],"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles",true,"grants","orgs","customDomains",{},{"_43":125,"_32":28,"_30":31,"_126":-5,"_41":42,"_40":14,"_127":119,"_38":128,"_13":129,"_33":34,"_35":37,"_24":25,"_130":-5,"_27":131},"a2626386-151d-11f1-94b0-42dde27851f2","parentId","latest",["D",1772335244760],3,"label",[28],"pulls","branches",1,"forks","envVars",{"_140":60,"_142":60},"connections",{"_140":60,"_141":60},"count","unfilledCount","emptyCount",{"_43":125,"_32":28,"_30":31,"_126":-5,"_41":42,"_40":14,"_127":119,"_38":144,"_13":129,"_33":34,"_35":37,"_24":25,"_130":-5,"_145":146,"_27":147},["D",1772335244760],"children",[],[28],["D",1771695970894],{"_93":92,"_100":96},{"_93":92,"_94":-5,"_95":96,"_18":54,"_97":98,"_99":25},{"_13":14,"_32":83,"_43":42,"_84":44,"_85":-5,"_86":-5,"_87":-5,"_88":-5,"_89":-5,"_90":45,"_91":92},"ctx",{"_76":-5,"_121":185,"_186":187,"_188":189,"_190":25,"_191":25,"_192":25,"_193":119,"_194":-5,"_195":-7,"_196":-7,"_197":60,"_198":25,"_199":25,"_200":-5,"_201":25,"_202":25,"_203":25,"_204":-5},"ENV",{"_156":157,"_158":159,"_160":83,"_161":162,"_163":164,"_165":166,"_167":168,"_169":170,"_171":172,"_173":174,"_175":176,"_177":178,"_179":180,"_181":182,"_183":184},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

${papers[0].headline}

${p.headline}

WANT TO CODE?

Engineering & Efficiency

Diff · v4