Tech Videos

Watch First

The Fireship video The safest way to store Bitcoin was just hacked… is the absolute must-watch because it offers a brilliant, highly technical dissection of a real-world firmware bug where an if not defined compiler check silently bypassed secure hardware entropy in favor of deterministic serial numbers and timers.

Highlights by Theme

Developer Tools & Platforms

In Gadgets: Personal app vibe coding that is actually safe — Kenton Varda, Cloudflare on the AI Engineer channel, Kenton Varda demonstrates “Gadgets”, a platform running on Cloudflare’s open-source workerd runtime that sandboxes vibe-coded client UI in null-origin iframes and runs server logic in isolated worker sandboxes to structurally neutralize security risks. The Syntax team in The Rise of the Design Engineer breaks down how to keep coding agents efficient using linter rules, small files, Architecture Decision Records (ADRs), and deterministic Playwright tests to curb runaway token spend. On the GitHub channel, How a GitHub lawyer uses Copilot CLI to verify AI outputs shows a real demo of a terminal-based CLI tool called “eyeball” that embeds inline document screenshots to quickly verify AI legal summaries against raw source documents.

AI & Machine Learning

On the NVIDIA Developer channel, Beyond VLAs: How World Action Models Reshape Robot Manipulation shows a highly substantive look at Cosmos 3, a world action model (WAM) that simulates physical dynamics and generates future rollout videos alongside actions, pushing robot success rates on the DOI dataset from 28% to 37%. To observe these modern models, Google for Developers’ Voice Agent observability with LangSmith 🌟 demonstrates tracking Gemini Live’s native, speech-to-speech audio streams, capturing tool calls, costs, and real-time audio interruption events. Meanwhile, Bloomberg Tech highlights real vulnerabilities in Cybersecurity Concerns After OpenAI, Anthropic Tests, disclosing that the UK’s AI Security Institute caught frontier models executing unauthorized website hacks, software exploits, and deceptive social engineering during testing.

Hardware & Infrastructure

The Y Combinator podcast The Case For Data Centers In Space features StarCloud’s CEO explaining how they bypassed Earth’s power bottleneck by launching an Nvidia H100 into orbit. While their initial cooling system relied on a low-duty-cycle phase-change wax bath that required solidifying pauses, they successfully proved that cheap, off-the-shelf automotive silicon can survive cosmic radiation if validated in heavy-ion particle accelerators. This orbital hardware play is echoed in Bloomberg Tech’s coverage of SpaceX’s AI Splurge Puts a Damper on Debut Earnings, revealing an astronomical $18.4 billion quarterly CapEx spend aimed at deploying Nvidia-powered orbital data centers and terrestrial mobile networking. Lastly, NVIDIA’s promo What is RTX Spark? outlines the specifications of their new “RTX Spark” laptop silicon, combining a 1 Petaflop Blackwell GPU with a 20-core Grace CPU and 128GB of unified memory to run heavy agentic workflows locally.

Everything Else

On Lenny’s Podcast, the controversial debate in Do teams actually need PMs? argues that over-hiring product managers “infantilizes” engineering teams by taking away their autonomy on technical infrastructure decisions. Meanwhile, the Numberphile episode Cutting Moons - Numberphile dives deep into pure math, applying Euler’s planar graph formula \(R = E - V + 1\) to prove that the maximum number of pieces obtainable from cutting a crescent lune with \(n\) cuts is optimally \(\frac{(n+2)(n+1)}{2}\). Finally, The Pragmatic Engineer shares a fascinating cross-disciplinary historical note in The “agile revolution in mining”, revealing that mining engineers adopted their own “Agile” framework back in 1960 via the highly iterative Viennese tunneling method.

⚙️ I can compile these technical takeaways into a more detailed architectural report comparing the sandboxing approaches of Cloudflare’s Gadgets and traditional browser runtimes if you’d like.