Tech Videos — 2026-06-16#

Watch First#

You Might Not Need 50 Diffusion Steps — Ziv Ilan, Nvidia NVIDIA’s Ziv Ilan gives a highly pragmatic technical breakdown of how to make video generation latency acceptable for production using quantization, KV caching, and distribution-based distillation.

Highlights by Theme#

Developer Tools & Platforms#

Microsoft and AWS are aggressively unbundling AI developer tooling to support strict enterprise constraints. On the Visual Studio Code channel, Bring Your Own AI… No Sign‑In Required! demonstrates how you can now configure custom LLM endpoints—such as local models running via LM Studio—without needing a GitHub Copilot login. For AWS shops, Running Claude Code on Amazon Bedrock from AWS Developers shows how to route Claude Code through Bedrock so your codebase never leaves your AWS account, taking advantage of cross-region inference routing and their zero-operator-access Mantle endpoint. Meanwhile, Google Cloud Tech’s Building long-running AI agents with ADK offers a practical architectural pattern: using a state machine and database-backed memory schema to let agents sleep for days waiting on external webhooks without suffering context window bloat.

AI & Machine Learning#

For a serious mathematical deep-dive, Microsoft Research’s Rare event analysis via stochastic optimal control explains how to avoid the impossible compute costs of brute-force simulating rare physical events like protein folding by learning committer functions via a closed-loop stochastic optimal control framework. On the evaluation front, OpenAI’s Why Tejal Patwardhan stopped underestimating the models - Episode 21 reveals how quickly standard benchmarks like SWE-bench are saturating, forcing researchers to shift toward complex, multi-day operational evaluations. Adding to the focus on efficiency, the AI Engineer channel’s You Might Not Need 50 Diffusion Steps — Ziv Ilan, Nvidia details how to drop from 50 diffusion steps down to single digits by leveraging distribution-based distillation and dynamic quantization.

Hardware & Infrastructure#

Marques Brownlee’s The Most Interesting Displays In The World! provides a first look at “Project Aura,” an XR collaboration between Google and Qualcomm that splits compute between lightweight 91g tracking glasses and a tethered Snapdragon Reality Elite puck running Android XR. On a networking-focused note, Computerphile unpacks the realities of internet censorship in Is it Possible to Block Childrens’ Access to Social Media? - Computerphile, methodically walking through the technical hurdles and bypasses of enforcing device-level, ISP-level, or DNS-level traffic blocks without relying on digital ID verification.

Everything Else#

For a break from pure technology, Dwarkesh Patel’s interview with historian Ada Palmer in How Machiavelli’s Florence bargained with Cesare Borgia for survival – Ada Palmer delivers a masterclass on 16th-century statecraft. The conversation completely reframes Machiavelli, painting him not as a scheming villain, but as a selfless systems thinker trying to build durable institutional incentives to protect his republic.