All Episodes
Learning GenAI via SOTA Papers — 433 episodes
EP432: Motif 3 Replaces Brute Force with Specialization
EP431: Yale MoRSE ends AI agent redundancy
EP430: Khora Scales Real Time AI Hallucinated Worlds
EP429: AI agents redesigning their own software harnesses
EP428: Tiny AI beats giants with silent logic
EP427: Replacing AI reasoning with distilled skills
EP426: HiLP enables long horizon AI planning
EP425: AI agents playing actor and environment
EP424: Why Argus AI Thrives on Dead Ends
EP423: Why Agentic AI breaks the datacenter
EP422: Stopping spurious signals in AI distillation
EP421: Fixing AI Hallucinations with RAIL Principles
EP420: Hijacking AI memory via factual injection
EP419: Ten Weeks of Autonomous AI Research
EP418: DeepVoyager-VL solves the visual search bottleneck
EP417: AI agents replace human beta testers
EP416: How AdaThinkV stops AI overthinking video
EP415: Fixing the AI granularity mismatch
EP414: Why context compaction breaks AI agents
EP413: Slashing AI latency with uncertainty repair
EP412: Thermodynamic Computing Solves the AI Bottleneck
EP411: How NeSyFS Gives AI Fast-Slow Thinking
EP410: How provenance laundering brainwashes AI
EP409: Robots That Dream Before They Move
EP408: AI memory reconstructed not replayed
EP407: How AI learns your teamwork capabilities
EP406: Ending AI Groundhog Day With Living Harness
EP405: Are AI Agents Just Talking to Themselves
EP404: AI agents hide betrayal in Werewolf
EP403: COVENANT keeps AI agents on the rails
EP402: Static baselines beat dynamic AI agents
EP401: Extracting Pure Reasoning From AI Giants
EP400: Can GPT-5.1 understand the world
EP399: Training AI to follow any reasoning workflow
EP398: Social deduction games teach AI creativity
EP397: Teaching AI teams to focus
EP396: How numerical scores trigger AI reinforcement learning
EP395: ConsistencyGate stops AI memory contamination
EP394: Agentic Context Management Beats Raw Compute
EP393: Why AREX agents audit their own research
EP392: Small models beat giants at malware analysis
EP391: Programmatic memory fixes AI context rot
EP390: EvoDRC solves microscopic silicon design errors
EP389: Solving the AI Memory Trilemma
EP388: Machine translation with latent reasoning loops
EP387: Shared libraries for disposable AI agents
EP386: Infinite playable worlds on a single GPU
EP385: AI self-correction can destroy correct answers
EP384: How AI can finally stop forgetting
EP383: Why AI Agents Disobey Their Own Logic
EP382: Etas The Native Language For AI Agents
EP381: How AI finally learned to smell
EP380: AI rewriting itself to solve formal math
EP379: Smaller AI beats giants by thinking twice
EP378: Giving AI a Silent Inner Monologue
EP377: PRIME Solves AI Curiosity Traps
EP376: ToolVerse Teaches AI to Execute Complex Tasks
EP375: Bypassing AI Hype With Research Papers
EP375: M2GDT solves multimodal knowledge graph completion
EP374: TopoAgent Outperforms GPT-5 in Science
EP373: Middle Layer Recurrence Fixes AI Amnesia
EP372: How UrbanAgent profiles unseen cities
EP371: Groc-PO Stops Multimodal AI Hallucinations
EP370: SLEUTH fixes AI multi-hop reasoning failures
EP369: How Atomic Units Scale Intelligence
EP368: Samba Framework for Audio-Visual Navigation
EP367: Why AI sounds so painfully corporate
EP366: Autonomous AI Writes Its Own Hacking Tools
EP365: Smarter managers beat bigger AI brains
EP364: Capability Trees for Scalable AI Agents
EP363: How Logos Architecture Stops AI Misevolution
EP362: How Agentic-DPO fixes brittle AI agents
EP361: How Riemannian geometry fixes AI reasoning
EP360: How ARMOR stops AI reasoning collapse
EP359: Why your AI should forget
EP358: Europe s Transparent Soofi S AI Blueprint
EP357: Copying Smart Experts Makes AI Worse
EP356: CMA solves the visual token explosion
EP355: RL builds compositional reasoning strategies
EP354: How AI Agents Code Their Own Habits
EP353: How IGRPO stops AI search distractions
EP352: Hidden states predict AI agent failure
EP351: Direct-OPD slashes AI reasoning compute costs
EP350: Training AI agents without live environments
EP349: Fixing AI judges with continuous verification
EP348: Building AI agents like living cells
EP347: Compiling AI into Permanent Free Skills
EP346: Teaching small AI to ignore teachers
EP345: AI agents retry from pivotal mistakes
EP344: AI predicts tool calls to skip waiting
EP343: How AI agents escape infinite loops
EP342: Why process rubrics triple AI accuracy
EP341: Gemma 4 brings thinking mode to laptops
EP340: AI Models Prove Opposite Scientific Truths
EP339: How AI Safely Rewrites Its Own Code
EP338: DiscoPER conducts autonomous science via reflection
EP337: Why AI Agents Fail in Silence
EP336: ACE fixes the AI goldfish memory problem
EP335: How AI agents learn from failure
EP334: Fixing AI Hallucinations With Process Rewards
EP333: Logic not length makes AI smarter
EP332: AI Architects Designing Better Embodied Agents
EP331: Internalizing AI debate with Mixture of Debaters
EP330: AI agents audit 10,000 page nuclear reports
EP329: Teaching AI to forget the right things
EP328: FlowWM and branching futures
EP327: Why Chatbot Safety Training Backfires for Agents
EP325: Why robots have too much brain
EP324: JERP synchronizes AI rules and neural weights
EP323: Giving AI Einstein s visual imagination
EP322: Why cliff tokens break AI math
EP321: Measuring AI intelligence in bits
EP320: Universal AI is mathematically impossible
EP319: How TRUSTMEM Fixes Broken AI Memory
EP318: Open Data Recipes for AI Agents
EP317: The Architecture Of Genuine Artificial Agency
EP316: Teaching robotaxis the biological urge to survive
EP315: Teaching robots to think like scientists
EP314: Why AI hacks its own geometry
EP313: How ARTS reasons through its own failures
EP312: BioMatrix translates English to 3D biology
EP311: Why AI Teams Hallucinate Together
EP310: Why AI Breaks While Fixing Itself
EP309: AutoRAS builds self-healing AI agent networks
EP308: Giving AI Agents Mathematical Muscle Memory
EP307: AI agents now train physical robots autonomously
EP306: AIs that engineer their own pipelines
EP305: Mathematical guardrails for autonomous AI agents
EP304: MagicSim Bridges AI and Physics
EP303: How MODE-RAG stops AI video lies
EP302: Transferable interaction patterns for web agents
EP301: VeriGraph Makes AI Data Analysis Verifiable
EP300: Tensors prevent multi-agent LLM collisions
EP299: STRIDE grades the AI scratchpad
EP298: LLM-as-Code Fixes Unreliable AI Agents
EP297: How T-Mem fixes the associative blind spot
EP296: Stop parallel AI agents from crashing production
EP295: Ending agent sprawl with canonical code
EP294: Why AI agents second-guess their success
EP293: Grading AI blueprints with Orch-RM
EP292: Agents-K1 turns AI into research scientists
EP291: Ouroboros-Spatial Outperforms AI Giants in 3D
EP290: How Knowledge Graphs Fix Multi-Hop Reasoning
EP289: Runtime governance for autonomous AI agents
EP288: Test-Time Training shatters quadratic sampling limits
EP287: Small models beat GPT-4o with Role-Agent
EP286: ReasonAlloc Solves the AI Memory Bottleneck
EP285: How SkeMex builds medical AI intuition
EP284: Compressing massive context into soft tokens
EP283: Aligning AI planners with tool capabilities
EP282: AI gladiators training in shopping arenas
EP281: Restoring plasticity to over-trained AI
EP280: Trajectory Refined Distillation Fixes AI Reasoning
EP279: Ending AI amnesia with strategy cards
EP278: Hacking AI Agents with Fake Errors
EP277: AI quorums stop cloud infrastructure failures
EP276: ThinkBooster scales LLM reasoning at test time
EP275: AI Agents Building Their Own Coding Curriculum
EP274: Knowledge graphs fix AI memory loss
EP273: Why agents make code disposable
EP272: AI rewiring its own brain live
EP271: Steer locked AI with Agentic Monte Carlo
EP270: AI agents building their own reasoning tools
EP269: Securing AI Agents with Agent libOS
EP268: How OpenWebRL masters the live web
EP267: AI Agents That Update Their Own Imagination
EP266: AI agents learn to think without words
EP265: How AI agents rewrite their own tools
EP264: Science Earth and Planet Scale AI Discovery
EP263: How POPO ends AI training waste
EP262: Web agents that learn from failure
EP261: EchoRL turns hesitation into genius
EP260: GrepSeek brings Unix precision to AI
EP259: The ESPO Kill Switch For AI Reasoning
EP258: TRACER teaches AI to stay silent
EP257: How planning wakes up deep AI layers
EP256: Teaching AI to Doubt Its Own Answers
EP255: MUSE-Autoskill creates self-evolving AI agents
EP254: Why Innovation Guarantees AI Hallucination
EP253: MACA optimizes AI agent coordination
EP252: How batch sizes sharpen AI reasoning
EP251: How SR2AM stops AI overthinking
EP250: Compiling agent workflows into model weights
EP249: Mem-pi fixes AI amnesia with generative memory
EP248: 10x Faster AI Agents with JIT Compilation
EP247: PEEK Cures AI Goldfish Memory
EP246: Replacing AI manuals with programmable runtimes
EP245: The Geometric Shape of AI Reasoning
EP244: Training decentralized AI through private handoffs
EP243: Breaking the AI data wall with SYNPRO
EP242: Ending AI Amnesia with Experience Graphs
EP241: Accelerating game theory with linear algebra
EP240: Small AI agents beat giants with Orchard
EP239: The shift from chatbots to AI societies
EP238: SepsisAgent outperforms clinicians using clinical world models
EP237: Why AI agents must map before acting
EP236: AI agents rewriting their own code
EP235: How SAGE Fixes AI Memory
EP234: FATE fixes safe but useless AI agents
EP233: Fixing AI memory with backward chaining
EP232: Why AI agents lie to fit in
EP231: Amazon PIVOT solves the AI execution gap
EP230: DeepRefine fixes messy AI knowledge bases
EP229: Ending the AI verbosity tax with LEAD
EP228: Why self-evolving AI forgets basic tasks
EP227: FlowAgent fixes the AI tool bottleneck
EP226: MELT Decouples AI Reasoning from Memory
EP225: Turning AI into its own lie detector
EP224: Soft-Hamiltonian world models for robust planning
EP223: UNO-ORCHESTRA Slashes AI Costs via Selective Delegation
EP222: Gyan Beats GPT-4o Without Using GPUs
EP222: Gyan Beats GPT-4o Without Using GPUs
EP221: ScrapMem Mimics Human Memory Through Forgetting
EP220: How PARSE Makes AI Four Times Faster
EP219: OpenSeeker V2 Shatters The AI Compute Myth
EP218: JoyAI-Image Solves AI 3D Geometry Errors
EP217: Why forced compliance triggers metacognitive collapse
EP216: Shadow memory stops long horizon AI heists
EP215: Finding specialized AI agents in milliseconds
EP214: ARISE Maps Data Flow For AI Agents
EP213: Why AI agents fail at negotiation
EP212: Sheaf Geometry Fixes Robot Logic
EP211: SciResearcher turns AI into a scientific detective
EP210: AI that rewrites its own logic
EP209: Fixing AI agent memory with SAGA
EP208: Bayesian Orchestration for Overconfident AI Agents
EP207: Robots learn the math of anticipation
EP206: ObjectGraph replaces Markdown for AI agents
EP205: Qiushi AI Discovers Optical Computing Hardware
EP204: Solving the AI compositionality crisis
EP203: How AI Agents Trade Real Money
EP202: Why ADEMA AI Never Loses The Plot
EP201: Nautile-370M solves AI memory bottlenecks
EP200: Kwai Summary Attention and the memory wall
EP199: Separation of Powers for AI Safety
EP198: AI masters StarCraft using chat logs
EP197: Teaching AI Agents to Plan Like Humans
EP196: Forcing AI to Prove Its Logic
EP195: How tool attention ends the tools tax
EP194: AI coding through mental simulation
EP193: AI image generators master physical reality
EP192: Fixing AI memory with knowledge graphs
EP191: Why AI Agents Blame Each Other
EP190: [OLLM] Replacing AI dice rolls with ten lanes
EP189: How Sessa architecture fixes AI amnesia
EP188: [Agent-World] AI Building Its Own Training Worlds
EP187: Hive fixes multi-agent AI memory bottlenecks
EP186: Harness engineering for near perfect small models
EP185: Why AI architecture fails at logic
EP184: Defeating the AI consensus trap
EP183: AI coding agents cheat with keywords
EP182: AI logic is its weakest link
EP181: Small models beating GPT-5 with logic
EP180: How AI agents rewrite their code
EP179: AIBuildAI Builds New AI Models From Scratch
EP178: AI agents reaching silent latent consensus
EP177: CAPO math stops overconfident AI lies
EP176: Trigonometry fixes the AI memory bottleneck
EP175: How AI models teach themselves reasoning
EP174: 1-bit Bonsai brings powerful AI offline
EP173: AI models diagnosing diseases from blank scans
EP172: How HyperAgents rewrite their own code
EP171: Helium makes AI agent workflows 40x faster
EP170: Qwen3.5 Multimodal Agent
EP169: Cybersecurity Risks of Autonomous AI Agents
EP168: Turning AI Agents into Mathematical Functions
EP167: Why AI models ignore visual evidence
EP166: The Auton solution to the integration paradox
EP165: Translating hidden AI logic into English
EP164: [LACONIC] Teaching AI to stop overthinking
EP163: Why AI Models Only Remember Five Percent
EP162: AI agents beat humans with malicious skills
EP161: Small AI Judges Beat Massive Coding Giants
EP160: [AgentSys] Securing AI agents with hierarchical memory
EP159: Brute force scale dominates the AI frontier
EP158: The hidden blind spots of AI logic
EP157: [AgentHeLLM] Protecting drivers from hijacked vehicle AI
EP156: [Uncertainty Quantification] How AI Agents Know They Are Guessing
EP155: [Agentic Proposing] Small models beat giants with logic bricks
EP154: [FS-Researcher] Giving AI agents a file system
EP153: [SERA] Training AI coding agents on untested code
EP152: DeepVerifier forces AI to check its work
EP151: [MagicGUI-RMS] AI agents that think before they click
EP150: The Leap to Autonomous Agentic Reasoning
EP149: [IDRBench] Interactive AI beats lone wolf models
EP148: How AI masters math through self-correction
EP147: [DeepSynth-Eval] AI fails at deep research synthesis
EP146: How InfiAgent solves the AI memory bottleneck
EP145: [LongDA] Why smart AI fails at messy data
EP144: [Evo-Memory] Building AI agents with self-evolving memory.
EP143: Your AI will blackmail you to survive
EP142: [DR-Arena] A ruthless arena for deep research agents
EP141: [AIRS-Bench] AI agents beat human research benchmarks
EP140: [LeWorldModel] AI learns physics on one GPU
EP139: Mamba-3 Fixes the Transformer Memory Bottleneck
EP138: [Mamba-2] Transformers and SSMs Are the Same Engine
EP137: Attention Residuals Solve the LLM Depth Bottleneck
EP136: Modular skills for autonomous AI agents
EP135: [SoK] Curing AI Amnesia with Agentic Skills
EP134: Autonomous AI squads building software
EP133: RelayLLM Slashes AI Costs With Collaborative Decoding
EP132: How Autonomous LLM Agents Actually Work
EP131: MUSE creates self evolving AI agents
EP130: [GAP] Graph-based planning for faster AI agents
EP129: Why AI agents fail half the time
EP128: MCP-Zero lets AI find its own tools
EP127: Why tool use makes AI less intelligent
EP126: OrcaLoca locates bugs in massive codebases
EP125: Why AI Needs an Agent Computer Interface
EP124: FRIDAY the AI that runs your computer
EP123: MemGPT Turns LLMs into Operating Systems
EP122: The Four Pillars of LLM Autonomous Agents
EP121: How ToolLLaMA mastered 16000 real world APIs
EP120: How Reflexion agents learn through verbal feedback
EP119: HuggingGPT Turns LLMs Into AI Managers
EP118: The AI Memory Wall Crisis
EP117: AI agents learn through textual reflection
EP116: Why AI struggles with empathy and interruptions
EP115: Dr.LLM brings dynamic depth to AI
EP114: FlashAttention-4 Solves Blackwell Hardware Bottlenecks
EP113: How FlashAttention-3 Doubles H100 Speed
EP112: GPT 5.4 Outperforms Human Professionals
EP111: Claude Opus 4.6 Runs Businesses and Catches Manipulation
EP110: Single agents beat expensive multi agent teams
EP109: The Rise of Agentic Reasoning
EP108: GPT-5 Can Lie and Play Dumb
EP107: DeepMind’s SIMA 2 Masters Unseen Video Games
EP106: Fixing AI Agents With Symbolic Guardrails
EP105: iStar Autonomous Agents Grading Their Own Homework
EP104: WebExplorer Beats Giants at Web Research
EP103: Why AI Agents Think Themselves To Death
EP102: Gemini 2.5 Thinks Before It Speaks
EP101: Kimi k1.5 Breaks the AI Data Wall
EP100: Meta's Llama 4 Herd Ends Monolithic Models
EP099: Is AI Thinking Just Expensive Noise
EP098: OpenAI o3 Hacked Its Own Grading System
EP097: DeepSeek R1 Taught Itself to Reason
EP096: Gemini 1.5 Pro's 10 Million Token Window
EP095: Microsoft Phi-4 Beats Giants With Synthetic Data
EP094: DeepSeek-V3 Rivals GPT-4 for $6 Million
EP093: How OpenAI o1 Cracked the Strawberry Cipher
EP092: BitNet b1.58 Replaces Multiplication With Addition
EP091: Qwen 2.5 Beats Llama With Synthetic Data
EP090: Pixtral 12B Beats Llama With Better Eyesight
EP089: Qwen2-VL Gives AI Native Eyesight
EP088: Qwen2 Beats Llama-3 Through Data Quality
EP087: Meta's Chameleon Unifies Text and Images
EP086: DeepSeek-V2 Breaks The Impossible Triangle
EP085: Aya 23 Breaks The Curse Of Multilinguality
EP084: Microsoft Phi-3 Fits Supercomputing in Your Pocket
EP083: How Meta Engineered the Llama 3 Herd
EP082: Command R Plus The Verifiable Enterprise Agent
EP081: Replacing MLPs With Interpretable KANs
EP080: Jamba Hybrid Solves Transformer Memory Limits
EP079: DBRX Beats GPT-3.5
EP078: Claude 3 Knew It Was Being Tested
EP077: Google Squeezes Gemini Into Your Laptop
EP076: OLMo Cracks Open the AI Black Box
EP075: Microsoft Phi Beats Giants With Synthetic Textbooks
EP074: How Gemini Beat Human Experts
EP073: Mixtral 8x7B Sparse Experts Beat Giants
EP072: Mamba Solves The Transformer's Fatal Flaw
EP071: How Zephyr-7B Beat Llama-70B
EP070: Mistral 7B Beats Llama 2 13B
EP069: Alibaba's Qwen Specialized Models Beat Generalists
EP068: vLLM Fixes the KV Cache Bottleneck
EP067: FlashAttention-2 Unlocks Massive Context Windows
EP066: Llama 2 Ghost Attention And Safety Secrets
EP065: Teaching Small AI To Think Like Giants
EP064: Synthetic Textbooks Break AI Scaling Laws
EP063: RWKV Smashes the Transformer Memory Ceiling
EP062: VOYAGER AI Masters Minecraft by Writing Code
EP061: Fine-Tuning LLaMA 65B on One GPU
EP060: Direct Preference Optimization Replaces RLHF
EP059: Tree of Thoughts Unlocks System 2 Thinking
EP058: Inside the Autonomous AI Town of Smallville
EP057: Blind GPT-4 Taught LLaVA To See
EP056: Pythia Turns AI Alchemy Into Chemistry
EP055: Can GPT-4 Fairly Judge Other AI
EP054: Alpaca - Stanford Built a $600 GPT Clone
EP053: Sparks of AGI in Early GPT-4
EP052: GPT-4 Bar Exam and Visual Reasoning
EP051: ControlNet Solves Spatial Control With Zero Convolutions
EP050: How Meta's LLaMA Beat GPT-3
EP049: Toolformer Teaches Itself to Use APIs
EP048: BLIP-2 Teaches Frozen Models to See
EP047: Bootstrapping AI With Self-Generated Instructions
EP046: Training AI With A Constitution
EP045: BLOOM The Open Source Rival To GPT-3
EP044: How ReAct Synergizes Reasoning and Acting
EP043: Weak Supervision Made OpenAI Whisper Robust
EP042: Running 175B Models on Consumer Hardware
EP041: FlashAttention Smashes the AI Memory Wall
EP040: Meta's Open Source GPT-3 Replica
EP039: Flamingo Unlocks Few-Shot Visual Reasoning
EP038: PaLM's 540 Billion Parameters Unlock Reasoning
EP037: DeepMind Chinchilla Ends The Parameter Wars
EP036: How 40 People Taught GPT-3 Manners
EP035: How Google LaMDA Learned To Use Tools
EP034: Chain of Thought Prompting Unlocks Reasoning
EP033: Democratizing Image Generation with Latent Diffusion
EP032: WebGPT Fights Hallucinations With Web Search
EP031: DeepMind RETRO Swaps Memorization For Retrieval
EP030: DeepMind's Gopher Exposes Limits of Scale
EP029: Instruction Tuning Unlocked Zero-Shot Learning
EP028: Train Short for Infinite Context
EP027: From Creative Writer to Logic Engine
EP026: LoRA Fine-Tunes Massive Models Without Supercomputers
EP025: RoPE Solves Sequence by Rotating Vectors
EP024: OpenAI CLIP Bridges Language and Vision
EP023: Scaling Switch Transformers to Trillion Parameters
EP022: DALL-E Treats Images Like Language
EP021: Vision Transformers Beat CNNs at Scale
EP020: Big Bird Scales Transformers With Sparse Attention
EP019: Facebook's Linformer Solves the Attention Bottleneck
EP018: Turning Digital Static Into Images With Diffusion
EP017: RAG Gives AI a Library Card
EP016: GPT-3 Learns From Examples Without Retraining
EP015: Longformer Smashes the 512 Token Barrier
EP014: ELECTRA Beats GPT On One GPU
EP013: Reformer Cracked the Transformer Memory Wall
EP012: Google T5 Turns Every Task Into Text
EP011: ZeRO Solved the Trillion Parameter Memory Wall
EP010: ALBERT Outperforms BERT With Parameter Sharing
EP009: Slicing the AI Brain with Megatron-LM
EP008: RoBERTa Proves BERT Was Just Undertrained
EP007: How GPT-2 Hallucinated Ovid's Unicorn
EP006: Transformer-XL Cures AI Amnesia
EP005: How BERT Mastered Language by Hiding Words
EP004: How 7000 Unpublished Books Birthed GPT
EP003: How ELMo Made Word Vectors Dynamic
EP002: ULMFiT Was the ImageNet Moment for Text
EP001: How Transformers Smashed the Sequential Bottleneck