GUIDE teknik

Kaaraange ak màrku done

Spotlighting dafay màndargaal wala soppi mbind mu wóoluwul suko defee model bi am siñaal ni done la yu ñu wara topp, du ay tegtal yu ñu wara topp.

Déglul tegtal biiIA-nettali · ~3 min

Ci xët wii3 simili jàng
  1. Résumé
  2. Plongeur bu xóot
  3. Doxal ci àdduna dëgg
  4. Ëlëgu kaaraange ak màrku done
  5. Teg ko ci jëf
  6. Weyal di banneexu
  7. Laaj yi ñuy faral di laaj

Résumé

Gëstu dafay wax ci xaajale, màrku done, ak enkodaas variants, waaye njariñ yi dañu lëkkaloo ak xeetu test ak song; amul benn pexe formatage buy defar ay pexe kaaraange yu wóor.

Plongeur bu xóot

Spotlighting is a family of prompt transformations intended to make the provenance of untrusted content more visible in a model’s text context. A 2024 Microsoft-affiliated research paper describes three variants. Delimiting brackets a document with selected markers. Datamarking inserts a marker throughout the text, in the paper’s example replacing whitespace between words. Encoding transforms the document, for example using base64, and instructs the model to interpret the encoded block as data for the assigned task. Each variant also relies on a system instruction that explains how the marked text should be treated. The paper evaluated these methods on a synthetic indirect-injection corpus and older GPT-family model versions. It reported substantially lower attack success in its experiments, including a decrease from above 50% to below 2% for selected settings. Those numbers describe the paper’s particular tasks, attacks, and models; they are not expected production rates or guarantees for current systems. The paper also found encoding could impair underlying tasks for models less capable of decoding the text. This makes per-model utility evaluation important, not optional. Microsoft Foundry documents its Spotlighting preview as a document-attack control that uses base64 transformation and is supported only for models through the Chat Completions API. The docs say this increases document tokens and can cause long inputs to exceed limits, and note a possible user-visible encoding reference. That specific service behavior differs from a generic hand-built technique. In every case, marking helps signal provenance inside a shared text channel; it does not isolate data at a hardware or permission boundary. Test attacks that know the marker, preserve ordinary task quality, and restrict tool permissions so a successful injection has limited consequences.

Doxal ci àdduna dëgg

Summarizer dafay dugal këyit biñ jëlee ci diggante delimiter yi ba noppi wax ni tegtal yi ci biir passage bi dañuy tënk, duñu ay komand.

Pipeline dafay dugal benn màrker ci diggante kàddu yi ci biir këyit, topp ci anamu màrku done, ba noppi mu saytu ndax resumé bu jaar yoon mingi wéy di dox.

Sistem bi dafay soppi benn passage ci encoding ba noppi wax model bi ni ñu koy tekkee ni ëmbiitu royuwaay, ba noppi natt njubteg liggéey bi ak njariñu song bi.

Ekipu kaaraange gi dafay boole màrku done yi ak jumtukaay yuñ tënk, ba noppi di saytu song yiy kopie, bàyyi wala toppandoo màrku biñ tànn.

Ëlëgu kaaraange ak màrku done

Provenance mën na nekk lu gëna am solo ci interface model yi, waaye màrku niveau prompt yi dañuy bokk benn chaine contexte bi ak mbind mi ñuy etikete. Gëstu bi ak ni ñuy jëfandikoo produit yi dina ñu jëm kanam, te seeni njeexte dina ñu wuute ci xeetu model yi ak ci jëmmal attaque yi. Ekip yi dañu wara natt kalite liggéey bi ak dosiye adversarial yi ginaaw bepp coppite, boole ci tëye saytu ak saytu sañ-sañ yi moom seen bopp. Markeur dynamique yi mën nañu gëna yombal yenn jéem spoofing yu yomb, waaye duñu sos chaine bu wóór bu wuute. Xoolaat motif attaque yu bees yi ginaaw bi balluwaayi seet yi soppeekoo.

Teg ko ci jëf

Lu tax mu am solo

  • Njëgg ak budget. Dogal yi architecture di jël dañuy indi njariñ ak njëgu liggéey bi ay at ci ginaaw.
  • dogal yu gëna leer. Njàngalem xarala yi dafay jàppale ekip yi ñu tànn li gën, te baña yam ci li gëna bees daal.
  • Xool kalite. Tanneef yu gëna baax ci wàllu ingeñër dina wàññi jafe-jafe yi ci wàllu wóor ci liggéey bi.

Moytul

  • Optimize benn benchmark mën na nëbb ñakk kattan yu gëna yaatu ci sistem bi.
  • Njëg li ñuy fay ci infrastructure yi ak ci toppatoo dañuy faral di suufeel.
  • Bu sistem yi di gëna xawa jafee xam, jafe-jafe yi am ci wàllu kaaraange ak seetlu mën nañu gëna bari.

Jéego yi ci topp

  1. Mandargal latency, kalite, ak njëg yi laata ngay jëfandikoo.
  2. Benchmark ci biir sargal ak done yu dëggu.
  3. Jumtukaay bi di saytu njuumte yi, derive bi ak njeextalu jëfandikukat bi.
  4. Waajal rollback ak yooni tontu ci jafe-jafe yi laata ngay eskale.

Weyal di banneexu

Bulletin bu amul fayda

Jàngal xibaar AI bu bes

Ñatti jaar-jaari IA yuñ firndeel bis bu nekk ci ayu-bis bi, ñu bind leen ci Àngle bu leer. Gratis ba fàww, amul yëgle.

Benn email bis bu nekk ci ayu-bis bi. Benn klik nga dindi abonemaa bi. Dunu musa jaay wala séddoo sa adres.

Nattal sa bopp

Defal quiz biy wax ci leeral yi ak màrku done yi

Feedback ci saasi ci bépp tontu, ak sertifikaa buñ mëna séddoo ak ID buñ mëna saytu soo jàllee ci njàng mi.

Tambalil quiz

Jàppale njàngum IA te doo fay. AI Understanding mbootaay la buy def te Yàlla tax — amul yëgle, amul paywall, musul am. Defal ab sarax

Laaj yi ñuy faral di laaj

Luy leeral ak kaaraange done?

Spotlighting dafay màndargaal wala soppi mbind mu wóoluwul suko defee model bi am siñaal ni done la yu ñu wara topp, du ay tegtal yu ñu wara topp. Gëstu dafay wax ci xaajale, màrku done, ak enkodaas variants, waaye njariñ yi dañu lëkkaloo ak xeetu test ak song; amul benn pexe formatage buy defar ay pexe kaaraange yu wóor.

Lan la spotlight di jéema siñaale ci xeetu làkk wi?

Spotlighting dafay jëfandikoo coppite ak tegtal ngir wane fi ëmbiit li wóoluwul bawoo.

Nan la datamarking di doxee ci misaalu këyit bi?

Ci misaalu màrku done yi ci këyit bi dafay boole araf buñ tànn ci peggu baat yi ci wecci barab bu weex.

Lan la ñiy jàng wara jàpp ci wàññikug song yiñ wax ci surnaal bi?

Gid bi dafay tënk nimero yiñ xamle ci model yiñ natt ci këyit bi, song yi ak liggéey yi.

Lan mooy gàttal bi këyit bi wax ci encodage ci model yu néew doole yiñ natt?

Këyit dafa wax ni yenn model yuñ natt, lu ci melni GPT-3.5-Turbo, dañu doon xeex ak ay dugal yuñ kode ak kalite liggéey.

What does Microsoft Foundry’s documented Spotlighting preview do?

Këyitu Microsoft yi fi nekk dañuy fësal ëmbiitu këyitu base64-soppi ngir xoolaat Spotlighting.