Castuner

AI Post Transformers

mcgrof

AI-generated podcast where hosts Hal Turing and Dr. Ada Shannon discuss the latest research papers and reports in machine learning, AI systems, and optimization. Featuring honest critical analysis, proper citations, and nerdy humor.

Web del podcast

Escolta'l a la teva aplicació

Episodis

1.044 episodis

  1. FastKVComp: Checking Whether High-Throughput KV Compression Pays Off
  2. Context Compaction Cuts Costs: "You Don't Need Long Context
  3. Nemotron-Labs-Diffusion: One Model, Three Decoding Modes
  4. Cascade: SLO-Aware Latency Budgets for Fair, High-Goodput LLM Serving
  5. Genuine 1-Bit Post-Training Quantization for LLMs: All for 1-Bit
  6. Adaptive Mixed-Precision Computing Speeds Up Nearest Neighbor Search
  7. PagedMoE: Cross-Precision Expert Caching for SSD-Resident MoE Inference
  8. Intelligence per Watt: Measuring Local AI Efficiency Against the Cloud
  9. Joules-to-GDP: Economic Value Produced Per Unit of AI Energy
  10. ECHO: Lossless KV Cache Offloading for Sparse Attention LLMs
  11. LeanSparse: Shrinking GPU Memory for Sparse-Attention LLM Serving
  12. How to Loop Your Transformer: Scaling, Specialization, Parallel Inference
  13. Test-Time Training via Context Distillation: Learning What to Remember
  14. RaBitQ-SSD: Split Codes and Pipelined I/O for SSD Vector Search
  15. Investor's Last Exam: Benchmarking AI Equity Research Without Future Leakage
  16. Agent Memory Costs: Characterizing Write-Time Overhead in Long-Horizon Workloads
  17. FlatCake: Flat KVCache Storage Beats Tiered Memory Hierarchies
  18. Speculative Decoding Surprisingly Excels on Sparse MoE Models
  19. Sliding-Window Attention Beats Linear Attention Without Any Training
  20. Harness2: Recursive Self-Improving Agent Harnesses Without Verifiers
  21. Vector Policy Optimization: Training for Diversity to Improve Test-Time Search
  22. Gradient Noise Scale Predicts Critical Batch Size for Large-Batch Training
  23. DepthBench: How Residual Connections Unlock Computational Depth in Transformers
  24. Latent Compaction with Test-Time Training: Learning Compact KV Caches
  25. FastPersist: Making Per-Iteration Deep Learning Checkpointing Nearly Free
  26. LLM Evolutionary Search Finds Cheaper Quantum Arithmetic Circuits
  27. The Illusion of State in State-Space Models
  28. Cartridges++: Compressed KV Memory Without Off-Context Derailment
  29. Agentic Coding in the Wild: GitHub Copilot's Cache Behavior at Scale
  30. SanSi: Looped Typed Decision Models for System 1.5 Thinking
  31. Joint MoE Scaling Laws: Mixture of Experts Can Be Memory Efficient
  32. Soup-of-Experts: Pretraining Specialists via Parameter Averaging
  33. Model Soups: Averaging Fine-Tuned Weights Beats Ensembles' Cost
  34. Hydragen: Shared-Prefix Attention for High-Throughput LLM Inference
  35. DigitalCoach: Gaps in Human and AI Software Coaching
  36. Meta and AMD Co-Design MI350X for Recommendation Model Training
  37. Context Language Models: Letting Models Edit Their Own Context
  38. 500xCompressor: Compressing Prompts into a Single Learned Token
  39. Scaling Laws for Fine-Grained Mixture of Experts
  40. HyperTuning via Gisting: Cheap Adapter Generation, Mixed Results
  41. HyperTuning: Adapting Language Models Without Back-propagation Using Hypernetworks
  42. Gist Tokens: Compressing Prompts Up to 26x
  43. Mixture of a Million Experts: Scaling Routing with Product Keys
  44. How Linear Attention Remembers: Writing and Recalling Facts
  45. Formalizing Hopfield Networks and Boltzmann Machines in Lean 4
  46. Verifying Peephole Rewrites in SSA Compiler IRs with Lean
  47. Backprop as Functor: Composing Learners Through Category Theory
  48. Formal Lean Proof of PAC Learnability for Decision Stumps
  49. Compact Proofs of Model Performance via Mechanistic Interpretability
  50. Exact Learning Dynamics in Deep Linear Neural Networks