https://macworks.dev/docs/month/tech-videos/Recent content in Tech@Youtube on MacWorksHugoenhttps://macworks.dev/docs/month/tech-videos/weekly-2026-W13/Mon, 01 Jan 0001 00:00:00 +0000https://macworks.dev/docs/month/tech-videos/weekly-2026-W13/<h1 id="tech-videos--week-of-2026-03-20-to-2026-03-26">Tech Videos — Week of 2026-03-20 to 2026-03-26<a class="anchor" href="#tech-videos--week-of-2026-03-20-to-2026-03-26">#</a></h1> <h2 id="watch-first">Watch First<a class="anchor" href="#watch-first">#</a></h2> <p><a href="#">加快語言模型生成速度 (2/2)：KV Cache</a> by Hung-yi Lee is the single highest-signal video this week for anyone dealing with AI infrastructure. It breaks down the brutal VRAM bottlenecks of LLM inference and the math behind architectural optimizations like Multi-Head Latent Attention far better than any high-level blog post.</p> <h2 id="week-in-review">Week in Review<a class="anchor" href="#week-in-review">#</a></h2> <p>The industry is aggressively moving away from monolithic, synchronous LLM chat wrappers toward decoupled, multi-agent swarms constrained by deterministic protocols like MCP and A2A. Simultaneously, the hyper-acceleration of the coding &ldquo;inner loop&rdquo; is exposing massive friction downstream, with machine-generated output completely overwhelming traditional CI/CD and human review pipelines. Infrastructure is shifting closer to the metal, with a clear focus on owning the compute layer and managing sub-millisecond latencies to make agentic workflows economically viable.</p>https://macworks.dev/docs/month/tech-videos/weekly-2026-W14/Mon, 01 Jan 0001 00:00:00 +0000https://macworks.dev/docs/month/tech-videos/weekly-2026-W14/<h1 id="tech-videos--week-of-2026-03-28-to-2026-04-03">Tech Videos — Week of 2026-03-28 to 2026-04-03<a class="anchor" href="#tech-videos--week-of-2026-03-28-to-2026-04-03">#</a></h1> <h2 id="watch-first">Watch First<a class="anchor" href="#watch-first">#</a></h2> <p>For the most impactful video, the Syntax channel&rsquo;s <a href="#">37,000 Lines of Slop</a> is the single best watch this week because it provides a brutal, necessary teardown of AI coding hype. It vividly demonstrates why blindly shipping massive LLM output without rigorous human review results in catastrophic production payloads, cutting through the marketing noise of effortless AI development.</p> <h2 id="week-in-review">Week in Review<a class="anchor" href="#week-in-review">#</a></h2> <p>The dominant theme this week is the awkward transition from isolated LLM chat interfaces to orchestrated, tool-using agents, exposing massive friction in both security and developer workflows. We are also seeing a definitive industry shift toward inference-bound hardware architectures, as scaling laws collide with concrete power, memory, and cooling bottlenecks.</p>