Disruptive ConceptsDisruptive ConceptsDISRUPTIVECONCEPTS

Interactive demo

Weekly curation

We do not pretend every paper is revolutionary. Here is a transparent slice of this week: top selections versus near-misses and why they did not make the shortlist.

Selected (sample)

Full week of 20 →

Near misses

  • 49

    ActSafeGuard: Differentiable and Training-Aligned Constraint Enforcement for Flow-Matching Policies

    Below top cut for this week’s automated tranche

  • 48

    Artificial Id: Drive and Persistent Alignment in Agentic AI

    Below top cut for this week’s automated tranche

  • 47

    SEED-UMI: Sharing the Exoskeleton between human and robot for onE-to-one Dexterous demonstration

    Below top cut for this week’s automated tranche

  • 45

    GPU-CFR: 80x Faster Counterfactual Regret Minimization by Compiling the Game to Static Dataflow and CUDA Graph Replay

    Below top cut for this week’s automated tranche

  • 45

    Data Scarcity and Model Sparsity: Mixtures-of-Experts Overfit More to Repeated Data

    Below top cut for this week’s automated tranche

  • 45

    ORCH: Organizational Principles Enable Collective Intelligence in Embodied AI

    Below top cut for this week’s automated tranche

  • 43

    EFI Pairs Without One-Way Puzzles: Oracle Separations from Communication Complexity

    Below top cut for this week’s automated tranche

  • 42

    Distance generalization in transformers: why bother with positional encoding?

    Below top cut for this week’s automated tranche

  1. 1Discover — arXiv queries per hot topic
  2. 2Score — five-factor disruptiveness model
  3. 3Select — top ~20 + near-miss transparency
  4. 4Explain — fixed article skeleton for humans
  5. 5Serve — free HTML + x402 JSON for agents