Episodes (Page 5)
✨
Sourcegraph's Amp Code achieves 15 shipping cycles per day with no code reviews through effective dogfooding and best-in-class coding agent architecture, rejecting subagents and prompt optimizers a...
Quinn Slack
Thorsten Ball
✨
Context engineering moves beyond simple RAG by addressing context rot, retrieval quality, chunking strategies, and memory structuring
Lance Martin
✨
Gorkem and Batuhan from Fal.ai (raised $125M Series C, crossed $100M ARR) discuss technical history of generative media and model impact on inference
✨
Ari Morcos (Datology) argues data curation is most impactful and underinvested area in AI, with effective filtering, rebalancing, sequencing, and synthetic generation outweighing architecture focus
Ari Morcos
✨
Modern vector databases differ fundamentally from information retrieval systems; Chroma prioritizes context quality and preventing 'context rot' as systems grow rather than just adding more embeddings
Jeff Huber
✨
OpenAI continues scaling RL for reasoning, moving beyond offline learning with online interactions and sample-efficient human curation; wall-clock time limitations in training remain an engineering...
✨
Nathan Lambert discusses evolution from RLHF to RLVR (Reinforcement Learning with Verifiable Rewards) in Tulu 3 paper for tasks with clear success criteria
✨
ChatGPT processes 2.5B prompts daily and will match Google's search volume by end of 2026; AI agents require structured, queryable data unlike human browsing patterns, creating new AI SEO/GEO industry
✨
Cline pioneered plan+act paradigm for coding agents as alternative to fast-apply models; operates as VS Code extension rather than fork to provide transparency and avoid vendor lock-in
Saoud Rizwan
✨
Speak represents third-generation language learning software (after Rosetta Stone Gen 1 and Duolingo Gen 2) by leveraging speech and language models to provide adaptive, fluent instruction previous...
Andrew Hsu
✨
Video diffusion models evolved from simple frame extension to sophisticated systems; Google's Veo 3 dominates 2024 by adding native audio generation, eliminating need for separate lipsynching and S...
✨
Jack Morris focuses on information-theoretic understanding of LLMs including embeddings and latent space representations, offering underrated research with accessible explanations for mass audience
✨
Noam Brown's work on solving Poker and Diplomacy demonstrates test-time compute scaling with multi-agent reasoning; System 1/2 analogy oversimplifies
Noam Brown
✨
Modular breaks CUDA monopoly through MAX framework and Mojo language, matching NVIDIA performance with AMD hardware through specialized low-level GPU programming
Chris Lattner
✨
Circuit Tracing reveals computational graphs in language models, providing mechanistic interpretability breakthrough through attribution graphs and open-source tooling released alongside Anthropic ...
Emmanuel Amiesen
✨
Solomon Hykes (Docker creator) leads Dagger, addressing agent chaos through environment isolation, standardization, and modular design enabling reproducible, trustworthy agent execution
Solomon Hykes
✨
Google's Gemini 2025 advances include real-time voice AI through Gemini Live API and native audio capabilities; millisecond-latency real-time workflows unlock new voice agent use cases
✨
CloudChef's Zippy provides industrial-scale kitchen robotics using one-shot demonstration learning, achieving Michelin-star food quality at $12/hour labor cost with simple business model
✨
Factory.ai raised $15M Series A from Sequoia to build autonomous software engineering 'droids' handling code generation and production incident response; launched in GA after demonstrating product ...
✨
Will Brown discusses multi-turn RL for multi-hour agents and inference-time reasoning advances in Claude 4 Opus and Gemini's Deep Think
Will Brown