与「Discovery」相关的搜索结果

🚨 Anthropic 一位资深工程师刚刚放出了一份 11页PDF ——《Loop Engineering》（循环工程），专门讲 agentic 系统！ 核心转变来了： 你不用再自己去提示 agent，而是构建一个能自动提示它的系统。 循环结构： Schedule → Discover → Build → Verify → Repeat 每个循环跑一轮，包含这5个关键动作： Discovery（发现）：agent 自己去寻找工作——失败的 CI、开放的 issue、最近的提交——而不是等着别人扔任务列表。 Handoff（交接）：每个任务分配一个独立的 git worktree，并行 agents 也不会互相冲突。 Verification（验证）：第二个 agent 被要求「假设代码是坏的」来审查第一个 agent 的成果。这就是那个「能说不」的东西。 Persistence（持久化）：结果直接写入磁盘，而不是留在随时可能被清空的上下文窗口里。 Scheduling（调度）：自动化定时唤醒循环。这才是让它真正成为「循环」的原因！ PDF链接评论区，想要的同学留言👇👇👇

The Steam Summer Sale is here! ☀️ For the next two weeks, discover thousands of discounts, take a gander through your Discovery Queue to meet new games and stickers, and check out new Points Shop items to match the summer vibes. Now until July 9!

Teams can now launch token auctions directly from the Uniswap Web App Onchain price discovery, transparent distribution, and instant liquidity, all without writing a single line of code Powered by Continuous Clearing Auctions

Today, China’s reported pause on new cross-border TRS allocations. This could have a larger impact on global semiconductor equities than many realize. Chinese HFs have evolved into some of the most sophisticated researchers of the semiconductor supply chain globally. They are often early in identifying underfollowed opportunities long before they become consensus trades. In recent years, they were among the early buyers of photonics names such as LITE and storage-related names such as SNDK before broader institutional participation arrived. Their reach extends well beyond the US into Korea, Taiwan and Japan, particularly across MLCC, power semis, analog, specialty materials, equipment and component suppliers. If cross-border TRS growth is restricted, many small/mid-cap semiconductor names could lose an important source of marginal demand and price discovery. Negative for breadth across KR/TW/JP semis. Potentially positive for HK-listed semiconductor as capital is redirected toward markets with fewer outbound constraints.

我用 Apodex 做了一次深度研究测试。 Apodex 的定位是 Self-Evolving Heavy-Duty Solver，也就是“自进化重型求解器”。它面向的不是简单问答，更专注那些重要、复杂、没有现成答案的问题：需要拆解、搜索、比较证据，再在下结论前核查关键主张。 这次我选的问题是： AI Agent 公司如何选择产品方向：开发者工具、企业工作流、研究助手，哪个更值得做？ 这个问题比单纯问“某个技术最近有什么进展”更难，因为它没有标准答案。要同时看市场需求、付费意愿、竞争格局、技术门槛、销售周期、融资叙事、短期落地难度和长期空间。 我用中档 Deep Reasoning 跑了一次，也尝试了 Deep Discovery。后面这个模式更能体现 Apodex 的核心能力：它会把问题拆成多条研究线，分别查开发者工具、企业工作流、研究助手，再补充 VC 视角、企业采用率、市场规模、客户流失风险和具体创业机会。 比较有意思的是，它没有在第一轮搜索后马上给结论。它先做总览，再发现证据不够，于是继续补查 TAM、创业方向排名、Menlo Ventures、SaaStr、BCG、企业 AI 报告等来源。这个过程能看到它在不断确认：哪些判断有数据支撑，哪些只是看起来合理。 最后它给出的排序是： 1. 垂直企业工作流 Agent 2. 垂直研究助手 3. 开发者工具 它认为，2026 年对大多数 AI Agent 创业公司来说，最值得做的是“垂直企业工作流 Agent”。理由是这类产品更容易找到明确买方，也更容易证明价值：比如保险理赔、医疗账单、物流异常处理、合规监测、采购和库存管理。这些场景本来就有人力和外包成本，Agent 如果能节省时间、降低错误率或提升收入，客户更容易付费。 开发者工具当然是 AI 最成熟的应用之一，但竞争也最强。Codex、Cursor、Claude Code、Devin 这些玩家已经占住用户心智。新公司如果还只是做通用 coding assistant，很难讲出差异。除非团队本身有很强的开发者工具背景，并且能切入更细的方向，比如合规代码、安全审查、CI/CD 自动化、企业代码治理。 研究助手的机会也存在，但前提是必须垂直化。通用 research assistant 很容易被大模型和浏览器插件覆盖。更有价值的是法律、金融、药研、监管、投研这类高价值场景，因为它们需要引用来源、审计记录和人工确认。换句话说，好的研究助手最后往往会变成“研究型企业工作流 Agent”。 这次测试让我更清楚地感受到 Apodex 和普通聊天机器人的区别：它的重点是先验证、后下结论。对这种变量多、信息散、需要做取舍的问题，过程透明和证据核查比答案本身更重要。 所以我觉得 Apodex 更适合拿来处理这类问题： · 一个创业方向值不值得做？ · 某个行业现在是否适合进入？ · 技术趋势背后有没有真实商业机会？ · 一个投资判断有哪些反方证据？ · 复杂议题里，哪些结论可以相信？ 这类问题很难靠一次搜索或一次对话解决，需要一个系统把资料找齐、拆开比较、反复验证。Apodex 想做的就是这件事。 体验入口： 开发者可以在 Hugging Face 下载模型： 感兴趣也可以加入 Discord。

After Batgirl (2022) was canceled, director Bilall Fallah says co-director Adil El Arbi told him to "shoot everything on your phone" before the footage vanished. By the time he checked the server, he had already been blocked by the studio, "everything was gone" — including the scenes featuring Michael Keaton's Batman. Adding to the frustration, Brendan Fraser who was cast as the villain Firefly, later said the film's Gotham, created using Glasgow, was one of the best versions of the city he'd ever seen. It was Warner Bros. Discovery that made the unprecedented decision to shelve the completed $90 million Batgirl movie as a tax write-off in August 2022. Both directors expressed extreme frustration that they couldn't even keep the rare scenes they had shot featuring Michael Keaton as Batman, which were permanently lost to them.

I cannot even imagine the consequences of the discovery of immortality. If and when biomedical engineering reaches that point, who gets to live? Who gets to reproduce? How expensive is it? You would be buying time in a severely limited space. It would be the end of us. Unending wars will be fought. The balance of power will massively shift.

Been iterating on @tomosman's loop. This one's winning: /goal produce a verified, code-derived behavioral spec for this web platform, captured in one canonical spreadsheet that carries every feature from spec -> tested -> fixed -> verified. Why: we need a single source of truth that maps every feature to its expected behavior *as the code implements it*, so that gaps and bugs surface and the platform can be driven to a known-good state. The spreadsheet is the source of truth. Work on the current repo. Do Phase 0 and Phase 1 under this goal; when the spec is complete, switch into the /loop below to drive testing and remediation. Keep moving through phases without stopping, except at a real checkpoint (defined below). Phase 0 - Plan (first): Detect the stack, the feature surface (routes, pages, components, API endpoints, background jobs, auth, settings…), and the test infra that already exists (unit/integration/e2e, browser automation, seeds/fixtures, a runnable dev server). Propose (a) how you'll inventory features, (b) the spreadsheet schema, and (c) how you'll test in the loop given what's available. Proceed once the plan holds. Phase 1 - Catalog & spec: Read the code and, for every feature, write a user story + the expected behavior as implemented, citing the file/function. Where the code is ambiguous, or behavior is undefined, log an open question - don't guess. Record every feature as a row in the canonical spreadsheet (create with the xlsx skill). Exit: every discoverable feature has a row. One row, concretely: | Area | User story | Expected behavior (from code) | Status | Defects | Type | Notes / source | |---|---|---|---|---|---|---| | Auth | As a returning user I want to log in with email+password so I can reach my dashboard | `POST /api/login` validates via bcrypt, sets httpOnly session cookie, 302 -> `/dashboard`; bad creds -> 401 + inline error | Spec'd | - | - | `api/auth/login.ts`, `LoginForm.tsx` | Canonical artifact: exactly one .xlsx, updated in place across every phase and loop iteration - never fork into per-phase or per-iteration files. Status flows Spec'd -> Tested-Pass / Tested-Fail -> Fixed -> Verified. The main thread is the single writer. Agentic execution: - Delegate breadth to subagents: fan feature discovery and per-area testing across subagents so the main thread stays focused. - Verify by running, not claiming - report real command/test output; state skips and unknowns plainly. - Checkpoint (pause, ask, end the turn) only for a destructive/irreversible action, a fix needing a genuine product decision, or input only I can give. Otherwise, keep going. - Self-check at each phase/loop boundary via a fresh-context subagent: re-verify the spreadsheet against the code (Phase 1) and against actual results (each loop pass). /loop Quality cycle - once the spec is complete, iterate test -> fix -> re-test until clean. Each iteration, in order: 1. Test: exercise every user story not yet Verified against the running app, preferring the strongest method available (browser/e2e automation > existing suites > documented static check only where execution truly isn't possible). Record actual pass/fail in the same spreadsheet; log every defect with its type (functional/logistical or UX). No app-behavior changes in this step. 2. Fix: think hard about root cause, then fix every functional/logistical and UX defect logged this iteration - cause, not symptom. Scope: only logged defects; no new features, no unrelated refactors. Update each row's status. 3. Re-test: re-run every story touched by a fix using the same method; set Verified, or back to Tested-Fail with notes if the fix didn't hold. Exit when all user stories are Verified and no open functional/UX defects remain. Safety cap: if a story is still failing after 3 full iterations, stop, leave it Tested-Fail with root-cause notes, and report it rather than looping further.

倪大说的对，欺诈是极难成立的。 但我们的主角MSTR 在 2000 年已经吃过一轮集体诉讼，被迫和解支付了超六亿美元的和解金。 信心是金融市场的血液，集体诉讼就是放血的刀子。 不需要割到动脉，一个伤口，一次证据开示，32 个币，就够了。 所以不需要等到判决，只需要等到不确定性（达成和解）。 确实在美国，欺诈很难成立。 《私人证券诉讼改革法案》（PSLRA，Private Securities Litigation Reform Act）门槛高，Tellabs 案（Tellabs v. Makor Issues, 551 U.S. 308）标准严，主观故意（Scienter）要"强烈推断"，每一项都是墙。 但事实上，美国集体诉讼从来不是打到判决的游戏。 是打到证据开示（Discovery）的游戏。 原告只需要熬过驳回动议（Motion to Dismiss），不用全部诉求（Claim），一条就够了。 一条活下来，案件进入证据开示。 证据开示是什么？不是法庭辩论，是取证。 法院命令 Strategy 交出：Saylor 的内部邮件。 董事会对 STRC 目标受众的讨论。 STRK/STRC 内部定价模型，Saylor 和承销商关于"这个产品到底卖给谁"的沟通。 这些材料一旦进入公开记录，不需要判决。 光是尴尬就足够杀人了。 这就是为什么 90%以上的证券集体诉讼以和解收场。 不是被告认输，是被告算了一笔账： 和解赔钱是一次性的，是可控的，封口的。 证据开示，不可控的，每一个内部邮件都可能变成下一个新闻头条。 这不是法律计算，这是风险计算。 实打实的案例。 瑞幸咖啡（Luckin Coffee），2020年，3亿美元销售造假。原告起诉，瑞幸否认，法律上各种辩护。 然后案件要进入证据开示了。 随后爆出 1.75 亿美元天价和解。 没等到判决。是证据开示的威胁，就让他们不得不掏出天价和解费。 再看我们的主角自己——MSTR 在 2000 年已经吃过一轮。 一份虚假收入报告，让市值一天蒸发 110 亿。 随后引发集体诉讼，索赔金额超 6 亿美元，最后和解。 再看美国 2025 年全国数据： 证券集体诉讼中位数和解金 1730 万美元，创三十年新高。全年总和解金超 30 亿美元。 2000 年的 MSTR 是一家软件公司。 2026 年的 Strategy 是全球最大的 BTC 持有者。850,000 个币。 这次的标的比 2000 年大三个数量级。 这些闻到血腥味的律所不会被几亿美金打发。 他们绝对会要历史级别的和解金。 而Saylor 的嘴！他的播客！他的"退休收入机器"！他的"月度收入机器"（monthly income machine）！ 给原告律师提供了这辈子最丰盛的靶子。 最后在看这个脆弱的市场。 2026年5月底，Strategy 卖了 32 个 BTC。 就 32 个，0.0038% 的持仓，价值 250 万美元。 BTC 应声暴跌，砸向 60,000 美元。 一个管理 850,000 BTC 的公司，卖掉了万分之一的持仓。 市场反应是什么？崩了。 不是技术面崩。不是宏观崩。是信心崩了。 明白了吗？ 市场不是被那 32 个 BTC 砸下去的。是被"Strategy 在卖币"这五个字砸下去的。 Saylor 花了四年建立的人设——永不售出，永不，永远——32 个币就碎了。 现在你告诉我： 如果 250 万美元的公司开支就让他卖币，那要支付不低于六亿美金的法律和解金，他还会不会卖？ 信心是金融市场的血液。 集体诉讼是放血的刀子。 不需要割到动脉。一个伤口。一次证据开示。32 个币。就够了。 我们不需要等到判决，你只需要等到不确定性的那一刻。 那就是现在即将要发生的事。 #STRC# #MicroStrategy# #MSTR# #Saylor# #BTC# #Bitcoin# #ClassAction# #集体诉讼# #证券欺诈# #和解# #Discovery# #证据开示# #LuckinCoffee# #瑞幸# #币圈# #Crypto#