mlsanityData to 7 Oct 2026

Synthetic data and self-improvement for LLMs

13 directions · 1,626 papers · 18 this week

Busiest this week

Most new since Mon 5 Oct

Self-evolving LLM agents

342 papers6 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

  1. The Harness as the Only Mutable Surface: Compliance-Bounded Self-Evolution of LLM Agents in Credit Pipelines, with a Measured Admission Gate
  2. DAEDALUS: Bootstrapping Agent Memory from Self-Generated Tasks
  3. Learning from Revision Consequences: Hindsight Meta-Experience Distillation for Self-Improving Agents
  4. EmbodiedSmith: Scaling Embodied Data through Recursive Self-Improvement Flywheel in Simulation
  5. SIGMA: Self-Improving Alignment Generalization from a Model Spec
Open →

All directions

Largest first · week from Mon 5 Oct

  1. Self-evolving LLM agentsagents, agent, self-evolving, memoryHigh confidence in this grouping: the papers form one clear topic · About3426High confidence in this grouping: the papers form one clear topic
  2. LLM synthetic data generationsynthetic, llms, language, generationMedium confidence in this grouping: mostly one topic, with blurry edges · About2752Medium confidence in this grouping: mostly one topic, with blurry edges
  3. Self-rewarding alignment and self-traininglanguage, llms, preference, self-improvementLow confidence in this grouping: the papers may belong to more than one topic, or the group may change · About1893Low confidence in this grouping: the papers may belong to more than one topic, or the group may change
  4. Self-improving MLLMs and synthetic datavisual, reasoning, multimodal, imageMedium confidence in this grouping: mostly one topic, with blurry edges · About1782Medium confidence in this grouping: mostly one topic, with blurry edges
  5. Synthetic reasoning data and bootstrappingreasoning, llms, language, trainingLow confidence in this grouping: the papers may belong to more than one topic, or the group may change · About1650Low confidence in this grouping: the papers may belong to more than one topic, or the group may change
  6. Code and instruction data synthesiscode, instruction, llms, languageMedium confidence in this grouping: mostly one topic, with blurry edges · About1061Medium confidence in this grouping: mostly one topic, with blurry edges
  7. Model collapse from synthetic datacollapse, synthetic, training, timeHigh confidence in this grouping: the papers form one clear topic · About850High confidence in this grouping: the papers form one clear topic
  8. Self-play RL for reasoningself-play, reasoning, reinforcement learning, reinforcementMedium confidence in this grouping: mostly one topic, with blurry edges · About720Medium confidence in this grouping: mostly one topic, with blurry edges
  9. Medical synthetic data and agentsclinical, medical, synthetic, patientHigh confidence in this grouping: the papers form one clear topic · About632High confidence in this grouping: the papers form one clear topic
  10. Privacy-preserving tabular data synthesisprivacy, tabular, synthetic, privateMedium confidence in this grouping: mostly one topic, with blurry edges · About530Medium confidence in this grouping: mostly one topic, with blurry edges
  11. Safety of self-evolving systemssafety, attack, attacks, defenseMedium confidence in this grouping: mostly one topic, with blurry edges · About471Medium confidence in this grouping: mostly one topic, with blurry edges
  12. Self-training for few-shot NLPself-training, few-shot, unlabeled, classificationHigh confidence in this grouping: the papers form one clear topic · About280High confidence in this grouping: the papers form one clear topic
  13. Self-evolving search agentssearch, retrieval, self-evolving, knowledgeHigh confidence in this grouping: the papers form one clear topic · About230High confidence in this grouping: the papers form one clear topic
1 new this week isn’t in a direction yet
  1. MotherTree: Meta-learning on synthetic data improves decision tree training

Weekly activity

Papers per week, 22 Jun – 5 Oct (last week partial)

  1. DirectionConfidencepapers16 wks
  2. Self-evolving LLM agentsHigh confidence in this grouping: the papers form one clear topic · AboutHigh confidence in this grouping: the papers form one clear topicPer week: 22 Jun: 3; 29 Jun: 8; 6 Jul: 7; 13 Jul: 13; 20 Jul: 4; 27 Jul: 12; 3 Aug: 10; 10 Aug: 11; 17 Aug: 8; 24 Aug: 10; 31 Aug: 7; 7 Sep: 7; 14 Sep: 6; 21 Sep: 7; 28 Sep: 25; 5 Oct: 6342144
  3. Self-rewarding alignment and self-trainingLow confidence in this grouping: the papers may belong to more than one topic, or the group may change · AboutLow confidence in this grouping: the papers may belong to more than one topic, or the group may changePer week: 22 Jun: 0; 29 Jun: 4; 6 Jul: 0; 13 Jul: 1; 20 Jul: 0; 27 Jul: 1; 3 Aug: 2; 10 Aug: 1; 17 Aug: 2; 24 Aug: 0; 31 Aug: 1; 7 Sep: 2; 14 Sep: 0; 21 Sep: 5; 28 Sep: 7; 5 Oct: 318929
  4. Model collapse from synthetic dataHigh confidence in this grouping: the papers form one clear topic · AboutHigh confidence in this grouping: the papers form one clear topicPer week: 22 Jun: 2; 29 Jun: 2; 6 Jul: 0; 13 Jul: 2; 20 Jul: 0; 27 Jul: 0; 3 Aug: 1; 10 Aug: 1; 17 Aug: 0; 24 Aug: 0; 31 Aug: 0; 7 Sep: 3; 14 Sep: 1; 21 Sep: 2; 28 Sep: 5; 5 Oct: 08519
  5. Self-improving MLLMs and synthetic dataMedium confidence in this grouping: mostly one topic, with blurry edges · AboutMedium confidence in this grouping: mostly one topic, with blurry edgesPer week: 22 Jun: 0; 29 Jun: 2; 6 Jul: 1; 13 Jul: 1; 20 Jul: 0; 27 Jul: 1; 3 Aug: 3; 10 Aug: 1; 17 Aug: 0; 24 Aug: 3; 31 Aug: 0; 7 Sep: 0; 14 Sep: 0; 21 Sep: 1; 28 Sep: 4; 5 Oct: 217819
  6. LLM synthetic data generationMedium confidence in this grouping: mostly one topic, with blurry edges · AboutMedium confidence in this grouping: mostly one topic, with blurry edgesPer week: 22 Jun: 2; 29 Jun: 1; 6 Jul: 1; 13 Jul: 2; 20 Jul: 1; 27 Jul: 1; 3 Aug: 0; 10 Aug: 0; 17 Aug: 0; 24 Aug: 3; 31 Aug: 2; 7 Sep: 3; 14 Sep: 1; 21 Sep: 0; 28 Sep: 2; 5 Oct: 227521
  7. Self-play RL for reasoningMedium confidence in this grouping: mostly one topic, with blurry edges · AboutMedium confidence in this grouping: mostly one topic, with blurry edgesPer week: 22 Jun: 1; 29 Jun: 1; 6 Jul: 2; 13 Jul: 1; 20 Jul: 1; 27 Jul: 0; 3 Aug: 0; 10 Aug: 0; 17 Aug: 1; 24 Aug: 0; 31 Aug: 0; 7 Sep: 2; 14 Sep: 0; 21 Sep: 2; 28 Sep: 3; 5 Oct: 07214
  8. Medical synthetic data and agentsHigh confidence in this grouping: the papers form one clear topic · AboutHigh confidence in this grouping: the papers form one clear topicPer week: 22 Jun: 2; 29 Jun: 0; 6 Jul: 1; 13 Jul: 2; 20 Jul: 0; 27 Jul: 0; 3 Aug: 0; 10 Aug: 0; 17 Aug: 0; 24 Aug: 0; 31 Aug: 0; 7 Sep: 0; 14 Sep: 1; 21 Sep: 0; 28 Sep: 2; 5 Oct: 26310
  9. Self-evolving search agentsHigh confidence in this grouping: the papers form one clear topic · AboutHigh confidence in this grouping: the papers form one clear topicPer week: 22 Jun: 0; 29 Jun: 0; 6 Jul: 0; 13 Jul: 2; 20 Jul: 0; 27 Jul: 0; 3 Aug: 2; 10 Aug: 0; 17 Aug: 0; 24 Aug: 0; 31 Aug: 1; 7 Sep: 0; 14 Sep: 1; 21 Sep: 0; 28 Sep: 2; 5 Oct: 0238
  10. Privacy-preserving tabular data synthesisMedium confidence in this grouping: mostly one topic, with blurry edges · AboutMedium confidence in this grouping: mostly one topic, with blurry edgesPer week: 22 Jun: 0; 29 Jun: 0; 6 Jul: 0; 13 Jul: 0; 20 Jul: 0; 27 Jul: 0; 3 Aug: 0; 10 Aug: 0; 17 Aug: 0; 24 Aug: 2; 31 Aug: 0; 7 Sep: 0; 14 Sep: 0; 21 Sep: 0; 28 Sep: 2; 5 Oct: 0534
  11. Safety of self-evolving systemsMedium confidence in this grouping: mostly one topic, with blurry edges · AboutMedium confidence in this grouping: mostly one topic, with blurry edgesPer week: 22 Jun: 2; 29 Jun: 0; 6 Jul: 0; 13 Jul: 0; 20 Jul: 0; 27 Jul: 1; 3 Aug: 3; 10 Aug: 3; 17 Aug: 0; 24 Aug: 2; 31 Aug: 3; 7 Sep: 0; 14 Sep: 1; 21 Sep: 0; 28 Sep: 0; 5 Oct: 14716
  12. Synthetic reasoning data and bootstrappingLow confidence in this grouping: the papers may belong to more than one topic, or the group may change · AboutLow confidence in this grouping: the papers may belong to more than one topic, or the group may changePer week: 22 Jun: 1; 29 Jun: 2; 6 Jul: 1; 13 Jul: 0; 20 Jul: 0; 27 Jul: 3; 3 Aug: 0; 10 Aug: 0; 17 Aug: 0; 24 Aug: 1; 31 Aug: 1; 7 Sep: 0; 14 Sep: 0; 21 Sep: 1; 28 Sep: 0; 5 Oct: 016510
  13. Code and instruction data synthesisMedium confidence in this grouping: mostly one topic, with blurry edges · AboutMedium confidence in this grouping: mostly one topic, with blurry edgesPer week: 22 Jun: 2; 29 Jun: 1; 6 Jul: 1; 13 Jul: 1; 20 Jul: 0; 27 Jul: 0; 3 Aug: 0; 10 Aug: 1; 17 Aug: 0; 24 Aug: 1; 31 Aug: 0; 7 Sep: 1; 14 Sep: 0; 21 Sep: 0; 28 Sep: 0; 5 Oct: 11069
  14. Self-training for few-shot NLPHigh confidence in this grouping: the papers form one clear topic · AboutHigh confidence in this grouping: the papers form one clear topicPer week: 22 Jun: 0; 29 Jun: 0; 6 Jul: 0; 13 Jul: 0; 20 Jul: 1; 27 Jul: 0; 3 Aug: 0; 10 Aug: 0; 17 Aug: 0; 24 Aug: 0; 31 Aug: 0; 7 Sep: 0; 14 Sep: 0; 21 Sep: 0; 28 Sep: 0; 5 Oct: 0281