{"generated_at": "2026-05-29T22:04:59.598638Z", "slug": "training_data", "source_id": "src_training_data", "name": "Training Data", "episode_count": 4, "avg_signal": 83.0, "median_signal": 84.3, "top_signal": 86.8, "latest_episode_at": "2026-05-26T09:00:00Z", "earliest_episode_at": "2026-05-08T17:05:00Z", "category_mode": "ai", "cover_image_url": "https://megaphone.imgix.net/podcasts/c85a76ee-3dd8-11f1-8934-37efe54dd294/image/89834425983d41c4f6fbb212a48603ac.png?ixlib=rails-4.3.1&max-w=3000&max-h=3000&fit=crop&auto=format,compress", "rank_score": 128.223, "episodes": [{"episode_id": "ep_training_data_3ef35ab768a1", "episode_title": "How Cursor Trained Composer on Fireworks: Distributed Infrastructure for High-Performance RL", "podcast_name": "Training Data", "podcast_slug": "training_data", "source_id": "src_training_data", "category": "ai", "publish_date": "2026-05-26T09:00:00Z", "overall_score": 86.8, "score_breakdown": {"clarity": 85.0, "originality": 94.0, "hype_penalty": 2.0, "actionability": 75.0, "technical_depth": 88.0, "information_density": 82.0}, "podcast_cover_url": "https://megaphone.imgix.net/podcasts/c85a76ee-3dd8-11f1-8934-37efe54dd294/image/89834425983d41c4f6fbb212a48603ac.png?ixlib=rails-4.3.1&max-w=3000&max-h=3000&fit=crop&auto=format,compress", "source_link": "https://www.sequoiacap.com/", "audio_url": "https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI1086513324.mp3", "listen_url": "https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI1086513324.mp3", "verdict": "must_listen", "why_listen": "You'll learn how to build scalable, numerically stable reinforcement learning systems for agentic models using real-world simulation and delta-weight synchronization.", "summary": "Cursor trained its agentic coding model Composer using distributed reinforcement learning infrastructure that simulates real user environments at scale. The model, a 1 trillion parameter Mixture-of-Experts with 30B active parameters, undergoes mid-training on code tokens to learn library-specific behaviors and real-world context. To make RL efficient, they built a high-throughput system using asynchronous training, delta-based weight updates, and numerical stability controls to prevent divergence due to floating-point mismatches."}, {"episode_id": "ep_training_data_d8ae323550b7", "episode_title": "Rebuilding IT From the Ground Up for the AI Age: Serval's Jake Stauch", "podcast_name": "Training Data", "podcast_slug": "training_data", "source_id": "src_training_data", "category": "ai", "publish_date": "2026-05-19T09:00:00Z", "overall_score": 83.4, "score_breakdown": {"clarity": 90.0, "originality": 85.0, "hype_penalty": 2.0, "actionability": 75.0, "technical_depth": 82.0, "information_density": 75.0}, "podcast_cover_url": "https://megaphone.imgix.net/podcasts/11894f9c-52f7-11f1-a868-f30cc344c1ee/image/0ab3357de98e9ad6ad957c31c694a1a7.jpg?ixlib=rails-4.3.1&max-w=3000&max-h=3000&fit=crop&auto=format,compress", "source_link": "https://www.sequoiacap.com/", "audio_url": "https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI4526751506.mp3", "listen_url": "https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI4526751506.mp3", "verdict": "must_listen", "why_listen": "You'll understand how to build durable enterprise AI products that survive rapid foundation model changes by focusing on workflow continuity, contextual agents, and economic efficiency.", "summary": "Serval is rebuilding IT for the AI age by automating enterprise workflows through a code-gen engine that turns natural language into executable automations with zero manual intervention. The company positions itself as an abstraction layer between foundation models and enterprise IT, maintaining moats through workflow context, guardrails, and unit economics rather than model ownership. As foundation models evolve rapidly, Serval focuses on stability, contextual awareness, and preventing agent sprawl while enabling non-technical users to build and maintain complex automations."}, {"episode_id": "ep_training_data_38cce8169049", "episode_title": "Suno's Mikey Shulman: Everyone Can Make Music Now", "podcast_name": "Training Data", "podcast_slug": "training_data", "source_id": "src_training_data", "category": "ai", "publish_date": "2026-05-13T18:55:00Z", "overall_score": 85.2, "score_breakdown": {"clarity": 90.0, "originality": 94.0, "hype_penalty": 2.0, "actionability": 75.0, "technical_depth": 82.0, "information_density": 75.0}, "podcast_cover_url": "https://megaphone.imgix.net/podcasts/beaede5a-4d82-11f1-9272-4f6372b94ca8/image/ba3ab676581811431f0fda5a047f7915.jpg?ixlib=rails-4.3.1&max-w=3000&max-h=3000&fit=crop&auto=format,compress", "source_link": "https://www.sequoiacap.com/", "audio_url": "https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI5617471569.mp3", "listen_url": "https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI5617471569.mp3", "verdict": "must_listen", "why_listen": "Understand how AI is redefining creative participation by turning music into a social, user-driven medium rather than a top-down product.", "summary": "Suno enables anyone to create music using AI by modeling sound directly rather than separating vocals and instruments, collapsing the barrier between music consumers and creators. The platform has already produced chart-topping hits and record deals, proving that AI-generated music can achieve cultural and commercial relevance. Suno's success hinges on preference data and iterative model improvements, not just scale, and it aims to make music creation social again by embedding it in shared experiences."}, {"episode_id": "ep_training_data_7d5f1142cf72", "episode_title": "ElevenLabs' Mati Staniszewski: How Voice Becomes the Interface for Everything", "podcast_name": "Training Data", "podcast_slug": "training_data", "source_id": "src_training_data", "category": "ai", "publish_date": "2026-05-08T17:05:00Z", "overall_score": 76.8, "score_breakdown": {"clarity": 85.0, "originality": 85.0, "hype_penalty": 3.0, "actionability": 72.0, "technical_depth": 72.0, "information_density": 65.0}, "podcast_cover_url": "https://megaphone.imgix.net/podcasts/41e0e0de-4a72-11f1-8f25-0bf25b7d3e24/image/f7b5a9d461ffc6de3429681d50b18343.png?ixlib=rails-4.3.1&max-w=3000&max-h=3000&fit=crop&auto=format,compress", "source_link": "https://www.sequoiacap.com/", "audio_url": "https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI5362828253.mp3", "listen_url": "https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI5362828253.mp3", "verdict": "must_listen", "why_listen": "Learn how a frontier AI company is building the infrastructure for voice-native agents and redefining human-computer interaction through emotional intelligence and full-stack control.", "summary": "Voice is evolving into the primary interface for human-computer interaction, with ElevenLabs building a full-stack voice engine that enables emotionally intelligent, context-aware voice agents. The company's defensible edge comes from vertical integration across speech-to-text, text-to-speech, and agent orchestration, combined with a talent model that embeds engineers in non-engineering teams. Real-world use cases span citizen services, education, and sales, where voice agents capture richer user intent than forms or buttons."}], "category_breakdown": [{"category": "ai", "count": 4}]}