Castuner
Submit a podcast

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.

Podcast website

Listen in your app

Episodes

1,032 episodes

  1. Test-Time Training via Context Distillation: Learning What to Remember
  2. RaBitQ-SSD: Split Codes and Pipelined I/O for SSD Vector Search
  3. Investor's Last Exam: Benchmarking AI Equity Research Without Future Leakage
  4. Agent Memory Costs: Characterizing Write-Time Overhead in Long-Horizon Workloads
  5. FlatCake: Flat KVCache Storage Beats Tiered Memory Hierarchies
  6. Speculative Decoding Surprisingly Excels on Sparse MoE Models
  7. Sliding-Window Attention Beats Linear Attention Without Any Training
  8. Harness2: Recursive Self-Improving Agent Harnesses Without Verifiers
  9. Vector Policy Optimization: Training for Diversity to Improve Test-Time Search
  10. Gradient Noise Scale Predicts Critical Batch Size for Large-Batch Training
  11. DepthBench: How Residual Connections Unlock Computational Depth in Transformers
  12. Latent Compaction with Test-Time Training: Learning Compact KV Caches
  13. FastPersist: Making Per-Iteration Deep Learning Checkpointing Nearly Free
  14. LLM Evolutionary Search Finds Cheaper Quantum Arithmetic Circuits
  15. The Illusion of State in State-Space Models
  16. Cartridges++: Compressed KV Memory Without Off-Context Derailment
  17. Agentic Coding in the Wild: GitHub Copilot's Cache Behavior at Scale
  18. SanSi: Looped Typed Decision Models for System 1.5 Thinking
  19. Joint MoE Scaling Laws: Mixture of Experts Can Be Memory Efficient
  20. Soup-of-Experts: Pretraining Specialists via Parameter Averaging
  21. Model Soups: Averaging Fine-Tuned Weights Beats Ensembles' Cost
  22. Hydragen: Shared-Prefix Attention for High-Throughput LLM Inference
  23. DigitalCoach: Gaps in Human and AI Software Coaching
  24. Meta and AMD Co-Design MI350X for Recommendation Model Training
  25. Context Language Models: Letting Models Edit Their Own Context
  26. 500xCompressor: Compressing Prompts into a Single Learned Token
  27. Scaling Laws for Fine-Grained Mixture of Experts
  28. HyperTuning via Gisting: Cheap Adapter Generation, Mixed Results
  29. HyperTuning: Adapting Language Models Without Back-propagation Using Hypernetworks
  30. Gist Tokens: Compressing Prompts Up to 26x
  31. Mixture of a Million Experts: Scaling Routing with Product Keys
  32. How Linear Attention Remembers: Writing and Recalling Facts
  33. Formalizing Hopfield Networks and Boltzmann Machines in Lean 4
  34. Verifying Peephole Rewrites in SSA Compiler IRs with Lean
  35. Backprop as Functor: Composing Learners Through Category Theory
  36. Formal Lean Proof of PAC Learnability for Decision Stumps
  37. Compact Proofs of Model Performance via Mechanistic Interpretability
  38. Exact Learning Dynamics in Deep Linear Neural Networks
  39. MEMORILLA: Latent Semantic Memory Beats Text Retrieval for Frozen LLMs
  40. DeLM: Decentralized Coding Agents Sharing Context Without a Leader
  41. Multi-Agent Transactive Memory: Sharing Agent Trajectories Across a Population
  42. PASTA: Steering LLM Attention Toward Highlighted Prompt Text
  43. Claude 3.5 Haiku Counts Characters on Curved Feature Manifolds
  44. Harness Learning: Training Small Models to Rewrite Code Around Frozen Solvers
  45. Jev-Mem: Fast System-One Controllers for Agentic Memory
  46. Cogentic: Multi-Agent Harness Discovers Proofs for Open Problems
  47. Code Predicts, Jev Judges: Separating Simulation from Evaluation
  48. Measuring Emergent Coordination in Multi-Agent LLMs via Information Decomposition
  49. Emergent Abilities of Large Language Models: Real or Metaphor?
  50. LoCoMo Benchmark: Do LLM Agents Remember Very Long Conversations?