Tags / Engineering & tools
Engineering
- How to Spot AI Writing in Chinese A V2EX thread on a Zhihu ban shows why detector scores often clash with human taste—and which Chinese AI tells, fake sincerity, and soft-promo patterns actually work.
- How much AI-written code should you review? Go by risk A V2EX user asked: after AI changes your code, do you read the diff? Of 94 replies, fewer than one in ten review carefully, yet they got the most likes. Using code review research plus DORA and Veracode data, here is a risk-based review ratio and six things worth checking every time.
- What can programmers still do now that AI writes code? Two popular V2EX threads in one week: one user had Codex add ray tracing to their engine on the first try and asked if AI is replacing skilled programmers; another asked how many will still live off writing code in five years. 200+ replies and Stanford employment data say the squeeze is on entry-level jobs and on coding itself. Programmers can still do five things, and they were always most of the job.
- Failure’s Worth, Through V2er Metaphors: Waterfall, Helicopter, Map A V2EX thread on Fields medalists warning about AI cracking math problems. The real gold is the metaphors—hiking vs helicopter, copying exam answers, fog-of-war maps, and inventing the crane.
- Distillation Through V2er Metaphors: Fishbowls, Kitchens, and Copying A V2EX thread asked if model distillation is like fishing from someone else’s bucket. The replies turned an abstract term into vivid metaphors—and a clean takeaway.
- An Alien Mind: Alignment Unsolved—Don’t Scale Flat-Out OpenAI’s chief scientist: AI is a grown alien intellect; goal ≠ value alignment; CoT monitoring is weakening. No lab has solved alignment enough for max-speed scaling.
- An Accidental Blackboard: Agents Coordinating via the Repo A Thoughtworks hyper-agentic experiment turned plan files plus frequent rebases into a classic blackboard—and argues coordination should leave source control.
- Code Is Fast, Delivery Isn't: Muse and One Context Coding speed gains get eaten by specs, context gaps and handoffs. Muse puts Agent OS and Harness on one context line — product, eng and QA need one flow.
- Why Code Search Makes Coding Agents So Expensive Sonar's comparison on Turing Post: semantic code navigation cut agent cost 5–36%. The sharper question: did the agent find every site that needed to change?
- Vibe-Coding Hangover: Cloudflare Locks Workers by Default AI lets anyone ship to the public Internet, and keeps CISOs awake. Access for Workers attaches policy to the Worker itself and defaults accounts to private.
- DeepSeek Engram: conditional memory as a new sparsity axis DeepSeek Engram adds O(1) lookup for the conditional memory MoE never had. At equal parameter and compute budget, reasoning gains more than rote knowledge.
- Software Is Not Files: Finding Materials ≠ Having Knowledge vivo Xiao Bo's KDC part 2: representation is not knowledge. A top RAG hit on a stale refund policy gave a confident wrong answer — governance beats tuning.
- The Defense Window Is Closing: Daybreak and GPT-5.6-Cyber OpenAI answers AI-scale attacks with Daybreak Blue/Red tiers and GPT-5.6-Cyber, taking advanced cyber task completion from 1.5% to 95% and finding real CVEs.
- Green Dashboards, Worse Answers: Latency Is Correctness Flat error rates hide thinner RAG context and truncated agent loops. Why TTFT, tail latency and retrieval metrics matter for answer quality, not just speed.
- Before You Let Agents Run Loose: AI Context Architecture Context architecture is not the RAG pipeline — it's guardrails, scopes, trust scores and human-in-the-loop. Stack Overflow on infrastructure vs architecture.
- Why Build Your WebApp in Canvas Instead of HTML Why Google Docs, Sheets, Canva, Miro, and Hivekit’s scheduler paint core UI on Canvas—speed, control, consistency, portability—and when you should not.
- Why "virtual company" multi-agent setups usually fail Naming agents PM, architect, developer and QA looks like a company. Mostly it kills information at every handoff. The value is parallel search, not a relay.
- CLI: the command-line revival in the AI era The command line is coming back. Feishu, DingTalk, WeCom, Google and Stripe have all open-sourced CLIs — a CLI is far easier for an agent to call than a GUI.
- Design Without Designing: how engineers ship design with AI You don't have to become a designer. With a three-layer toolkit you can go from zero to shipping design every week in three months. Neethan Wu's full system.
- gstack: the Claude Code factory YC's CEO uses Garry Tan open-sourced gstack, the Claude Code toolkit he actually runs: 15 Skills that turn a model into a virtual engineering team, while remaining YC CEO.
- Software engineering is splitting into three layers When implementation cost falls toward zero, the bottleneck moves from coding to judgment. The market is splitting into three layers of work, not just of pay.
- An agent-friendly blog A personal site shouldn't only be a page for people. In the agent era it should also be a machine-readable interface. How this blog is built for both.