
HISTORY近 30 天历史柱高表示当天去重热搜数量
2421262229232424242529262227302823292230223123012402260324042505230627073008230921102611261225132514251531163517271819今天
10/21—11/19 有历史数据
- 01Extra reasoning time can make a language model's answer worse, not better, because each token it writes shapes what it writes next. Reasoning models like OpenAI's o1 and DeepSeek's R1 generate intermediate 'thinking tokens' before a final answer, trained with methods like DeepSeek's Group Relative Policy Optimization. Researchers found one model asked to compute '2 + 3' got 5 correctly, then kept going until it produced thirteen redundant solutions to the same question. In a train-timetable exam
- 02DevOps Daily can be reached from the terminal by SSH or a CLI. Running 'ssh ssh.devops-daily.com' opens the app with no install, while the open source 'devops-daily' CLI (install via Homebrew, npm, or Go) saves quiz scores and flashcard progress and works offline after the first sync. The content includes 369 posts, 47 quizzes, 27 flashcard decks, 24 checklists, a daily incident, and links to 68 games and simulators. There's no telemetry or account requirement. The CLI and SSH app share one Go c
- 03A week of metering a home lab with a Tapo P110 smart plug revealed it was idle 90% of the time yet still drew a steady 62 to 80W baseline, averaging 72W or 12.1 kWh per week. Scaled to a year, that's 630.7 kWh. At the US average rate of 18.16 cents per kWh, that comes to $114.54 per year, or about $9.50 a month. The cost ranges from around $81 in Louisiana to nearly $298 per year in Hawaii, where rates hit 47.18 cents per kWh.
- 04It’s about time we ship this one. Reputation has been here forever and now it gets the home it deserves. Hover over your profile image in the sidebar -> click the reputation icon -> voilà. Or if you’re lazy use this link: https://daily.dev/analytics/reputation Deep dive into your activity, what earned you the most, and some more cool stuff. Enjoy and let me know why this feature is so awesome.
- 05Qwen Code v0.28.0 ships with no known breaking changes. The release adds qwen4-max model support with a native effort ladder, cache TTL, and image detail options. Headless sessions and the Ink TUI now run on a shared agent loop by default, and a new Aone Code channel plugin adds polling event-stream support. Bug fixes include a TOCTOU hardening recheck for transcript file types and a patch for express-rate-limit window accounting. The web shell gains a sidebar workspace filter, per-turn hook sta
- 06
PHP's scaling problems come from its runtime, not the language itself, the author argues. Every PHP-FPM worker opens its own MySQL connection, so 50 workers need 50 connections, and that multiplies across every server you add. The author built a small Go proxy that pools connections, speaking the MySQL wire protocol on both sides, so PHP connects to it like it's MySQL. In a test lab with MySQL capped at 20 connections and 50 PHP-FPM workers, going direct failed three out of four requests; throug - 07Non-technical employees are increasingly vibe coding their own apps outside IT oversight, and engineering teams end up inheriting the fallout. Retool surveyed 817 people building software and found 60% had created something outside IT oversight in the past year, while 51% had built AI software actually used by their teams. Companies like Ramp and Stripe have built internal AI tools letting non-engineers create apps with company data. Tines built its own platform, Tines 3B, so IT and security tea
- 08Developers working with AI agents feel more tired even though they type less, Fran Soto argues. The new effort goes into switching between tasks, verifying code you didn't write, and deciding for multiple agents at once. Soto distinguishes fatigue (fixed by rest), burnout (chronic stress, worth talking to someone about), and 'AI brain fry', a term from a Harvard Business Review article for mental fatigue from supervising AI tools frenetically. The suggested fixes: cap how many tasks or agent cha
- 09Claude can now join video calls as a visible participant, take notes, answer questions, and build docs or whiteboards while a team talks. Claude Meetings is in beta on Team and Enterprise plans, currently supporting Google Meet only. Claude asks permission before using connected tools and saves a transcript, summary, and screen descriptions, not audio. By default it stays out of calls with outside guests unless an admin allows it. Notes and transcripts are free during the beta, and docs, whitebo
- 10An analysis of 159,041 Lovable projects found widespread unprotected sensitive data in apps built with the AI platform. Of 34,545 projects using Supabase, 24,664 sent database records on page load without login. Further read-only requests turned up email addresses in 10,560 tables across 7,403 sites, 30% of those checked. Exposed data included over 80,000 forum profiles with names and birth dates, 200,000 chat messages, passport numbers and expiration dates for over 1,000 records, contracts, inv
- 11PHP's old advantage, cheap plentiful developers, has weakened now that AI coding agents cut the cost of learning any language, Azeem Hassni argues. He rewrote a PHP app in Rust on the same hardware and got six to nine times the requests per second, with idle memory dropping from 372 MiB to 30 MiB and startup from 1.62 seconds to 0.22. He still praises Laravel's productivity and admits Rust developers cost more and are scarcer. His pick for new projects: Rust or Go for throughput-sensitive servic
- 12Tabularis v0.32.0 adds table transfer between same-driver connections, conditional cell formatting, reusable SQL snippets and execution plan comparisons. You can now export a table to SQL and import it elsewhere, previewing structure before writing with fail, drop-and-recreate, or append options. Conditional formatting colors cells or rows by rule, saved per connection, schema and table. A new Snippets settings tab supports trigger prefixes and tab-stop placeholders. Visual EXPLAIN can save a pl
- 13One boring habit separates exceptional engineers from the rest. The post covers why fixing a bug isn't the same as learning from it, and the habit you need to practice to become exceptional. It also covers how to tell what deserves your time, and why AI makes this harder, and how to use AI to make it easier. It describes the compounding advantage this builds over a career.
- 14.NET 11 adds authentication refresh for long-lived SignalR connections, letting clients swap an expired token for a new one over a dedicated /refresh endpoint without reconnecting. The server rejects refreshes that change the sub claim, returning a 403. MVC filters, SignalR hubs, and Blazor's AuthorizeView now share one authorization model via a new AuthorizationPolicy.CombineAsync overload, so custom attributes work consistently across programming models. Passkeys get names inferred from the au
- 15A npm package, tailwind-dictionary, builds a Tailwind CSS theme directly from design tokens exported from Figma. It reads both the W3C Design Tokens format (DTCG 2025.10) and the legacy Style Dictionary format, and also supports exports from Tokens Studio and Terrazzo. It generates themes for Tailwind 3 and 4, supporting multiple themes like light, dark, high contrast or a second brand. Configuration lives in one config.json file, and a --watch mode rebuilds the theme automatically when tokens c


















