W07 - Thoughts Prompted by DeepSeek

DeepSeek shows us what technology can be genuinely attractive: high openness and low cost fundamentally accelerate AI's historical progress. There is a Jevons paradox from the Industrial Revolution that says improving resource efficiency can actually increase overall consumption of that resource. DeepSeek seems to recreate that paradox: any new technology quickly captures and expands the market as costs fall sharply, and the net demand for compute and energy will inevitably grow. This helps explain NVIDIA’s market swing earlier this year.

DeepSeek's surge has produced an information explosion. The challenge is filtering noise: what information matters and what can be ignored. With the era of large language models upon us, everyone starts from a similar baseline—what will the next wave of talent competition look like?

High-quality sources are the most effective response to information overload. Even the best material degrades when passed along repeatedly. The cost to obtain most upstream information is not high; we’ve just become used to having it pushed to us.

Technology changes rapidly; only what withstands time becomes a classic. Reading Hackers & Painters last year—a collection written over twenty years ago—still gave me many Aha moments. The hacker’s approach to life, what makes good design and a good language, are perennial topics worth revisiting. I recently found an out-of-print book called On the Edge, which records fragments of Cutler’s development of Windows NT and illustrates his mercurial nature and his ability to both motivate and drive a team. Cutler is called “Silicon Valley’s greatest kernel programmer”; now in his seventies, he still writes code at Microsoft.

Someone on X raised a “70% problem”: non-programmers using AI can quickly accomplish about 70% of a coding task, but the final 30% yields diminishing returns and is hard to close—often you can only hope AI will fix issues created by AI. What does it take to finish the remaining 30%? Understanding the business and customers, architectural judgment, accumulated mental models, and a fundamental grasp of AI-driven coding.

In a one-on-one with a classmate, we explored what to learn when picking up a new language. It’s not about memorizing and mastering a second Red Book—many concepts and rules that arise under tight constraints are low-value and non-transferable. Don’t get lost in tools and rules; focus on the knowledge that cultivates genuine insight.