groundy
modelspage 2 / 3· back to top →

models & research

more in this beat51–100 of 119 · page 2 / 3
  1. jun 11modelsDoes Attribution Patching Lie? A Fix for a Common Interpretability Shortcut
  2. jun 12modelsCan You Make a Multimodal Model Unlearn With Activation Steering?
  3. jun 12modelsWhy Pruning a Model Can Raise Its Out-of-Distribution Accuracy
  4. jun 10modelsDo Unified Multimodal Models Actually Interleave Understanding and Generation?
  5. jun 10modelsHow LLMs Track Who Did What: The Entity Rebinding Circuit
  6. jun 10modelsClaude Fable 5 vs Opus 4.8: When 2x Pricing Is Worth It
  7. jun 10modelsClaude Mythos 5 Access Rules: Who Gets Project Glasswing and Why
  8. jun 10modelsFable 5 Distillation Protection: How Anthropic Blocks Model Copying
  9. jun 10modelsSkip Fable 5 or Upgrade? When Opus 4.8 and Sonnet 4.6 Are Still Enough
  10. jun 09modelsLLM Steganography: Can Defenders Detect Payloads Hidden in Model Output?
  11. jun 09modelsDo Privacy Defenses Actually Protect Fine-Tuned LLMs? A New Benchmark
  12. jun 09modelsCan You Reconstruct an LLM's System Prompt From Its Activations?
  13. jun 09modelsDoes Softmax Normalization Limit What Attention Can Represent?
  14. jun 08modelsCan an Attacker Steal Your Model's Last Layer From Its Outputs?
  15. jun 07modelsCan LLMs Leak Training Data? A New Test Splits Capacity From Intent
  16. jun 07modelsWhen an AI Agent's Tools Break, Can It Recover? A New Benchmark
  17. jun 06modelsMiniMax M3 Bets on Sparse Attention for 1M Context. Does the Math Hold?
  18. jun 06modelsCan One Model Handle Every CAD Task? UniCAD Tests It
  19. jun 06modelsDo Foundation Models Actually Learn Relational Structure In-Context?
  20. jun 06modelsCan LLMs Write Better Research Paper Titles Than Authors?
  21. jun 06modelsDoes Information-Theoretic Example Selection Beat kNN for In-Context Learning?
  22. jun 06modelsDo Concept Bottleneck Model Benchmarks Measure Interpretability or Dataset Bias?
  23. jun 06modelsContinuous Bit-Width Quantization vs Fixed INT4: Does LiftQuant Beat Discrete?
  24. jun 05modelsFederated Learning for Industrial IoT Anomaly Detection: The Data-Locality Tradeoff
  25. jun 05modelsReading Failed LLM Reasoning Traces Won't Tell You Which Ones RL Can Fix
  26. jun 05modelsCan You Stitch Two Foundation Models Together Without Retraining?
  27. jun 05modelsDo Reasoning LLMs Waste Tokens? OckBench Tries to Measure It
  28. jun 04modelsWhich Layer Detects LLM Hallucinations Best? The Case Against Fixed-Layer Probes
  29. jun 03modelsCross-Domain RL Training Degrades Capabilities. CARE-RL Reweights to Fix It
  30. jun 03modelsLLM Watermarking Without Quality Loss: The Non-Distortionary Approach
  31. jun 02modelsTreating LLM Agent Memory as a Database: The VikingMem Approach
  32. jun 02modelsCan a Language Model Work Without a Neural Network? A New arXiv Paper Says Yes
  33. jun 02modelsCan Code-Generating LLMs Do Engineering Math? FEM-Bench Tests Them
  34. jun 02modelsUnlearning Isn't Deletion: arXiv 2505.16831 Shows Machine Unlearning in LLMs Is Reversible
  35. jun 01modelsWhy LLMs Fail at Spatial Reasoning When Planning Navigation
  36. jun 01modelsDoes Giving AI Agents More Skills Help? A Controlled SkillsBench Study
  37. may 31modelsCan an LLM Peer-Review Your Paper? A New Behavior Benchmark
  38. may 31modelsAnthropic Scaled Sparse Autoencoders to Claude 3 Sonnet. Interpretability Now Costs Compute
  39. may 29modelsTracing Why LLM Agent Memory Fails: A Method for Attributing Errors
  40. may 29modelsPersona Prompts Change Who an LLM Recommends as an Expert
  41. may 28modelsOpus 4.8 vs Opus 4.7: What Changed and What Did Not
  42. may 28modelsOpus 4.8 Batch API: 1M Context, 300k Output, and Team Cost Controls
  43. may 27modelsScale Vectors: Tiny Parameter Subsets That Disproportionately Steer LLM Behavior
  44. may 27modelsOne Learning Rate Doesn't Fit All: Heavy-Tail Layerwise LR Schedules for LLM Pretraining
  45. may 26modelsAudio LLMs Break When the Codec Changes: A Robustness Vector Voice-AI Teams Haven't Tested
  46. may 26modelsDo LLMs Know What Not to Say? Causal Evidence for Statistical Preemption
  47. may 25modelsEmbedding Compression at Training Time: DIVE's Gradient Trick vs Post-Hoc Quantization for Vector DBs
  48. may 25modelsμP Hyperparameter Transfer Has an Embedding Layer Hole, New arXiv Paper Says
  49. may 24modelsProject Glasswing One Month In: AI Bug Discovery Has Outpaced the Patch Pipeline
  50. may 23modelsarXiv 2605.16428 Measures AI Search's Drag on Publisher Traffic Using Paired Google and Reddit Data