逐日AI

面试题库

共 328 题,当前筛选 1 题。

标签
还有 235 个标签
#idempotency5#streaming5#structured-output5#chunking4#deployment4#distributed-systems4#rag4#system-prompt4#tool-calling4#client3#embeddings3#failure-modes3#ingestion3#mcp3#message-bus3#operations3#progressive-disclosure3#ranking3#timeline3#agent-loop2#agentic-rag2#agents-sdk2#caching2#citations2#code-review2#communication2#concurrency2#consistency2#context2#context-engineering2#context-rot2#cost-control2#data-modeling2#grounding2#hybrid-search2#langgraph2#latency2#model-migration2#model-routing2#multi-agent2#ordering2#pipeline-design2#prompt-basics2#prompt-engineering2#protocol2#rate-limiting2#react2#redis-streams2#responses-api2#retrieval2#retrieval-quality2#routing2#runtime2#sse2#state-management2#statelessness2#subagents2#system-design2#tool-design2#tooling2#tracing2#trade-offs2#transport2#vector-database2#versioning2#workflow2#abstention1#access-control1#agent-design1#agent-quality1#altitude1#approvals1#async1#async-task1#atomicity1#attention-budget1#auth1#av-sync1#behavioral1#bm251#candidate-selection1#capacity-planning1#chain-of-thought1#checkpointing1#ci1#citation-verification1#claude-code1#cli-design1#cloud1#compaction1#compression1#content-hash1#context-compression1#context-window1#contextual-retrieval1#cost-optimization1#cross-model1#dag1#data-quality1#database1#decision-making1#decomposition1#degradation1#deliberate-practice1#design1#diagnostics1#dimensions1#distribution1#docker1#documentation1#engineering-judgement1#engineering-tradeoffs1#eval1#event-driven1#fallback1#fan-out1#ffmpeg1#forking1#four-elements1#framework-design1#framework-selection1#golden-set1#hallucination1#handoffs1#headless1#hnsw1#hybrid1#hyde1#image-generation1#incremental-recompute1#incremental-sync1#index-maintenance1#index-routing1#indexing1#information-retrieval1#instruction-hierarchy1#intent-routing1#interrupt-merge1#interview-prep1#invalidation1#isolation1#ivfflat1#just-in-time1#knowledge-organization1#lease1#llm-as-judge1#llm-output-quality1#long-context1#loop-guard1#media-pipeline1#metadata1#metrics1#mobile1#model-selection1#multi-tenancy1#multimodal1#nodejs1#orchestration1#pagination1#parent-child1#pdf-parsing1#performance1#permissions1#persistence1#pgvector1#pipeline-reliability1#portfolio1#prioritization1#production-readiness1#prompt-assembly1#prompt-caching1#prompt-injection1#prompt-limits1#prompt-techniques1#prompt-template1#prompt-versioning1#provider-abstraction1#quality-check1#quantization1#query-transformation1#quiet-hours1#rank-fusion1#reasoning1#recall1#redis1#reflection1#refusal1#reporting1#reproducibility1#rerank1#retrieval-failure1#retrieval-metrics1#retry1#retry-semantics1#retry-strategy1#review1#rrf1#sandbox1#sandboxing1#scalability1#scheduling1#schema-design1#scoping1#scripts1#secrets-management1#self-assessment1#self-presentation1#self-reflection1#service-architecture1#session-management1#sessions1#sharding1#skill-authoring1#skill-description1#skills1#spec1#state-machine1#stateless1#stopping-criteria1#subtitles1#task-graph1#team-governance1#testing1#tool-budget1#tool-execution1#tool-naming1#tools1#tts1#tuning1#ux1#validation1#vector-index1#verification1#workflow-engine1#xml-tags1

14 天用 Agent 搭一条 AI 短剧生产线

D10 审片室:能预览、能改词、能重生成单镜的人机协作后台

  • 版本回滚要存什么?只存最终产物够不够?What must a version record hold for rollback? Are the final artifacts enough?
    国内高频海外高频进阶#versioning#rollback#data-modeling

    分析过程 · 先想清楚再作答

    1. 题眼在「够不够」三个字,它在提示答案是否定的。先把问题重述成一句判断:**版本不是备份**,这句话说出来这题就答对了一半。
    2. 两者的语义不一样。备份是「出事了拿回来」,只需要保留最近一份好状态;版本是「两个都在」,要能并排对比、来回切换,最终选哪个由人定。审核场景要的是后者。
    3. 所以每个版本要存三类东西:产物本身(按版本分目录,一个文件都不删)、产生它的输入(那一版的台词与画面描述,否则三天后没人说得清两版差在哪)、以及这一版重跑了哪些节点与原因。
    4. 当前版本要设计成一个指针,不是一份拷贝。回滚就是把指针挪回去,不搬文件,因此是瞬时且可逆的;这也让「再切回新版本」变成理所当然的操作。
    5. 有一个连带影响必须提到,提了就说明你真做过:**回滚一镜会改变整集的时间轴**。新版配音比旧版长一秒,切回去之后后面所有镜头的起止时间都要重排。所以回滚之后要重算一次时间轴,好在这是纯本地计算,很便宜。
    6. 可预期的追问是「版本存多久」。按产物体积和业务价值定:小文本无限存,视频这种大件设一个保留期,过期只留元数据和输入,需要时可以按同样的输入重跑出来。

    How to reason about it · think before answering

    1. The hinge is are they enough, which signals the answer is no. Restate it as a claim: a version is not a backup. Saying that sentence gets you half the credit.
    2. The semantics differ. A backup means restore after an incident and only needs the latest good state. A version means both exist, side by side, switchable, with a human choosing. Review workflows need the latter.
    3. So each version stores three things: the artifacts themselves, kept in per-version directories with nothing deleted; the inputs that produced them, the line and the visual description, or nobody can explain the difference three days later; and which nodes reran plus why.
    4. The current version should be a pointer, not a copy. Rollback moves the pointer without touching files, which makes it instant and reversible, and makes switching forward again equally natural.
    5. Mention the knock-on effect, because it shows you have actually shipped this: rolling back one shot changes the whole episode timeline. If the new take of the voice is a second longer, every later shot shifts, so rollback must recompute the timeline. That part is cheap local computation.
    6. Expected follow-up: how long to keep versions. Scale it by artifact size and business value: keep small text forever, put a retention window on video, and after expiry keep only metadata and inputs so the artifact can be regenerated on demand.

    答题要点

    • 版本不是备份:备份只要最近一份好状态,版本要求新旧同时存在、能并排对比。
    • 每版要存三类:产物(按版本分目录、不删)、产生它的输入、重跑的节点与原因。
    • 当前版本是指针不是拷贝,回滚只挪指针,瞬时且可逆。
    • 回滚一镜会改变整集时间轴,回滚后要重算一次——这是纯本地计算,很便宜。
    • 保留策略按体积分级:文本长期留,大视频设保留期,过期只留元数据与输入以便按需重跑。

    Key points

    • A version is not a backup: backups keep the latest good state, versions keep old and new side by side.
    • Store three things per version: artifacts in per-version directories, the inputs that produced them, and which nodes reran and why.
    • Make the current version a pointer, not a copy, so rollback is instant and reversible.
    • Rolling back one shot shifts the episode timeline, so recompute it after rollback; it is cheap local work.
    • Set retention by size: keep text forever, expire large video and retain metadata plus inputs for regeneration.