Lately I’ve been reading a lot of books and interviews and felt the urge to share a few ideas. So I’m launching a new column—“Gecko’s Thoughts”—a small homage to the legendary blog Programming Thoughts (编程随想). I’ll post thoughts here from time to time. Most entries will be short and probably incomplete; if a topic seems especially interesting, I may expand it into a full article later.
Podcasts: the “Zhihu” of a New Era
Interview‑style podcasts have quietly become what early Zhihu once was for the tech‑savvy crowd. For those of us in our twenties—the so‑called “lost generation”—they’re one of the few reliable ways to break out of an information rut. If you care about frontier tech and business insights, following a handful of good podcasts is invaluable.
AI Is Killing Once‑Glorious Paradigms
If you run into a coding problem, your first instinct might be to open Stack Overflow (hopefully not CSDN 😅). Yet in May 2025 the site saw only 20 000 new questions—down from over 300 000 per month in 2020. AI tools have cannibalized the need for “how‑to” questions. Likewise, homework‑answer sites like Chegg or CourseHero are faring even worse.
AI is wiping out these once‑booming platforms; the next generation of giant AI agents may render them extinct.
Startups: A Tale of Two Ecosystems
Everyone knows it’s easier to start a company in the U.S. than in China. Without connections or a strong pedigree, even a novel business model struggles to get funding in China. Investors love to ask:
“What if a tech giant copies your product?”
That question alone crushes many indie successes before they can bloom. Instead of acquisitions or licensing, we see cloning, undercutting, and eventual suffocation of small teams. Yes, China files an impressive number of patents every year—but how many are actually granted? How many begin with WO‑ rather than CN‑? We still have a long road ahead. Economic downturns can, paradoxically, be the breeding ground for true breakthroughs.
How Much Glue Work Is Enough?
“Glue Work” is anything that makes intra‑ and inter‑team collaboration smoother—onboarding, docs, internal courses. In both tech and manufacturing firms, these tasks are often uncredited in KPIs, meaning low ROI and poor maintenance. Finding the right balance between glue work and “real” technical work is a worthwhile study.
Lessons from the Perceptron’s Rise and Fall
AI history is a saga of Symbolists vs. Connectionists. Before AlexNet, Transformers, and GPT, there was Rosenblatt’s perceptron, two AI winters, and everything in between.
A recent three‑part essay series (in Chinese) traces that journey:
Part 1-Rosenblatt and the birth of the perceptron
Part 2-How Minsky’s critiques plunged AI into its first winter
Part 3-From that winter to GPT‑3 and today
The author even built a simple perceptron simulator-worth a spin.
A Little SSL→TLS Trivia
Ever wondered why SSL was renamed TLS? This blog post explains the inside politics. No doubt that corporate turf wars (think Netscape vs. IE) can be as bewildering as they are petty.