Sources
- AI Engineer
- All-In Podcast
- Andrej Karpathy
- Anthropic
- Apple
- Apple Developer
- AWS Events
- ByteByteGo
- Computerphile
- Cursor
- Dwarkesh Patel
- EO
- Fireship
- GitHub
- Google Cloud Tech
- Google DeepMind
- Google for Developers
- Hung-yi Lee
- Lenny's Podcast
- Lex Clips
- Lex Fridman
- Life at Google
- Marques Brownlee
- Microsoft
- No Priors: AI, Machine Learning, Tech, & Startups
- Numberphile
- NVIDIA
- OpenAI
- Perplexity
- Quanta Magazine
- Slack
- The Pragmatic Engineer
- Visual Studio Code
Tech Videos — 2026-06-12#
Watch First#
5 Papers That Show Where AI Research Is Heading Right Now A dense, highly technical breakdown of cutting-edge AI research that skips the marketing fluff to cover the limits of current LLM self-play, formal code verification via Lean, and scaling laws in computational biology.
Highlights by Theme#
Developer Tools & Platforms#
The AI bubble is bursting (Syntax) covers a critical shift in AI development economics: GitHub Copilot is moving from usage-based to token-based pricing, making long-running agentic workflows massively more expensive for developers. Meanwhile, GitHub’s The Download highlights the release of Claude Fable 5, a “Mythos-class” model designed for autonomous coding that costs a steep $10/$50 per million tokens and requires 30-day data retention for safety classifiers. On the debugging front, OpenAI released a notable update in Debug web apps with browser use in Codex, integrating the Chrome DevTools Protocol to finally allow Codex to profile runtime performance and inspect network traffic directly.
AI & Machine Learning#
The standout deep-dive is 5 Papers That Show Where AI Research Is Heading Right Now (Y Combinator), which explains why vanilla reinforcement learning and self-play for LLMs eventually plateau without techniques like Self-Guided Self-Play to ground synthetic data. The presentation also credibly argues that the “bitter lesson” of data scaling is replacing handcrafted features in protein biology with ESM-Cambrian, and demonstrates Stream RAG for reducing voice AI latency. For local deployment, Google’s AIventure: Vibe Coding Journey shows how to run the new Gemma 4 open-weights model locally in the browser using Transformers.js for zero-cost inference.
Hardware & Infrastructure#
For local edge compute, Microsoft teased the Surface RTX Spark Dev Box in The Download, a developer PC packing an Nvidia Blackwell RTX GPU and Grace CPU capable of 1 petaflop of local AI compute for local model fine-tuning. At the macro scale, the All-In’s Best Ideas Pitch Competition includes a highly credible infrastructure thesis on Talon Energy, detailing the severe 106GW power constraints hitting the PJM grid as hyperscalers exhaust existing baseload power, forcing them to sign massive, long-term power purchase agreements.
Everything Else#
For engineering leadership, A leader’s guide to advanced team structures in an agentic world (AWS Events) offers a pragmatic look at how AI code generation is hollowing out junior execution roles while vastly multiplying the output of expert generalists. It warns organizations to deliberately maintain an “hourglass” team structure rather than a top-heavy diamond, ensuring that the pipeline of future senior engineers with grounded judgment isn’t destroyed by over-relying on AI execution.