trends.barnas.net

Rising now

The live ranking for 16 Nov 2026. It changes every day; each Monday the top ten are frozen from it into the ledger.

25 ideas · new = entered since the day before

  1. weight-space programmingin the ten

    Using an LLM to directly edit neural network weights to diagnose and fix model behaviour without gradient training.

    22 mentions · 21 groupsMentions per week: 0, 0, 0, 0, 0, 1, 13, 8

  2. zero-knowledge proofs of machine learning inferencein the ten

    Cryptographic proofs that a model inference was computed correctly without revealing the model or inputs.

    6 mentions · 6 groupsMentions per week: 0, 0, 0, 0, 0, 1, 1, 4

  3. recursive self-improvementin the ten

    Agent autonomously improving its own capabilities through iterative learning cycles that revise its own knowledge or models.

    65 mentions · 64 groupsMentions per week: 0, 4, 26, 24, 20, 16, 22, 7

  4. NVFP4 quantizationin the ten

    A 4-bit floating-point number format with block scaling for quantizing neural network weights and activations on NVIDIA hardware.

    8 mentions · 8 groupsMentions per week: 0, 1, 5, 2, 1, 1, 4, 2

  5. Agent Client Protocolin the ten

    A protocol standardizing communication between code editors and AI coding agents.

    11 mentions · 11 groupsMentions per week: 0, 0, 3, 1, 1, 1, 7, 2

  6. pixel-space diffusionin the ten

    Diffusion image generation operating directly on pixels rather than in a latent autoencoder space.

    13 mentions · 13 groupsMentions per week: 0, 1, 3, 2, 3, 1, 6, 3

  7. diffusion language modelin the ten

    Language model that generates text by iteratively denoising masked or noisy token sequences.

    72 mentions · 69 groupsMentions per week: 0, 8, 24, 43, 29, 15, 17, 11

  8. driving world model

    A generative model that predicts future autonomous driving sensor observations given control signals.

    7 mentions · 7 groupsMentions per week: 0, 0, 0, 1, 1, 0, 3, 3

  9. EEG visual decodingin the ten

    Reconstructing or classifying visual content from electroencephalography brain signals.

    7 mentions · 7 groupsMentions per week: 0, 0, 2, 0, 1, 0, 4, 2

  10. LLM agent for roboticsnew

    Placing a large language model planner in the robot control loop to compose perception and motion tools.

    6 mentions · 5 groupsMentions per week: 0, 0, 0, 1, 1, 0, 2, 3

  11. agent harness evolution

    Iteratively optimizing the surrounding scaffolding of an LLM agent using execution feedback.

    57 mentions · 57 groupsMentions per week: 0, 6, 11, 17, 21, 18, 14, 4

  12. on-policy distillationin the ten

    Distilling a student model from teacher signals computed on the student's own generated trajectories.

    78 mentions · 77 groupsMentions per week: 0, 10, 36, 47, 23, 25, 17, 13

  13. spec-driven development

    Development practice where machine-readable specifications and acceptance criteria drive implementation and verification.

    10 mentions · 10 groupsMentions per week: 0, 0, 3, 4, 2, 3, 4, 1

  14. multi-agent coding

    Running several AI coding agents in parallel to carry out software development tasks.

    27 mentions · 27 groupsMentions per week: 0, 4, 6, 6, 7, 9, 5, 6

  15. expert pruningnew

    Removing less useful experts from a mixture-of-experts model after training to reduce its size and cost.

    9 mentions · 8 groupsMentions per week: 0, 0, 1, 2, 1, 2, 4, 2

  16. world action model

    Models that jointly forecast future scenes and ego trajectories to support planning in autonomous driving.

    52 mentions · 52 groupsMentions per week: 0, 5, 30, 18, 15, 16, 11, 10

  17. specialized decision modelin the ten

    A model that selects among given options instead of generating free-form text, trained for decision making.

    8 mentions · 8 groupsMentions per week: 0, 0, 2, 1, 2, 1, 4, 1

  18. self-evolving agents

    LLM-based agents that autonomously modify their own runtime instructions or tools to adapt to changing conditions.

    16 mentions · 16 groupsMentions per week: 0, 1, 11, 7, 4, 2, 5, 5

  19. feed-forward 3D reconstruction

    Predicting 3D scene geometry such as point maps directly in one forward pass without optimization.

    21 mentions · 21 groupsMentions per week: 0, 1, 6, 2, 8, 1, 8, 4

  20. looped transformer

    A transformer that reuses shared blocks over repeated recurrent loops to increase depth without adding parameters.

    37 mentions · 34 groupsMentions per week: 0, 1, 11, 13, 11, 9, 11, 6

  21. GPU kernel optimization

    Automatically writing and tuning GPU kernels for speed and correctness.

    14 mentions · 13 groupsMentions per week: 0, 1, 2, 1, 2, 5, 2, 5

  22. differential privacynew

    Formal privacy guarantee limiting how much any single record can influence a released output.

    40 mentions · 38 groupsMentions per week: 0, 4, 10, 13, 13, 9, 12, 6

  23. conditional value-at-risknew

    A risk measure equal to the expected loss in the worst tail of a distribution, used to penalise rare severe outcomes.

    12 mentions · 12 groupsMentions per week: 0, 2, 4, 8, 0, 3, 7, 2

  24. parameter-efficient fine-tuningnew

    Adapting large pretrained models by training only a small number of added or modified parameters.

    31 mentions · 31 groupsMentions per week: 0, 4, 13, 10, 4, 7, 11, 9

  25. vLLMnew

    High-throughput serving engine for large language models using paged attention and continuous batching.

    18 mentions · 18 groupsMentions per week: 0, 1, 5, 3, 3, 7, 7, 1