Prompt muhandisligi sahifasining manbasini koʻrish

{{Qisqa tavsif|Generativ sunʼiy intellektga kiritiladigan koʻrsatmalarni strukturalash va optimallashtirish jarayoni}}

'''Prompt muhandisligi''' (inglizcha: ''prompt engineering'') — [[generativ sunʼiy intellekt]] (GenAI) modelidan muayyan va aniq natijalarni olish uchun [[tabiiy til]]dagi [[kiritish-chiqarish|kiritish maʼlumotlarini]] ('''promptlar''' deb ataladi) strukturalash va shakllantirish jarayonidir. '''Kontekst muhandisligi''' (inglizcha: ''context engineering'') — bu dasturiy taʼminot muhandisligining turdosh yoʻnalishi boʻlib, u GenAI modeliga taqdim etiladigan prompt va promptdan tashqari kontekstlarni (tizimiy koʻrsatmalar, [[metamaʼlumotlar]], [[API]] vositalari hamda tokenlarni) boshqarishga qaratilgan.

Shuningdek, u generativ AI modeliga beriladigan kiritish koʻrsatmalarini yanada aniqroq, tegishliroq yoki foydaliroq natijalar olish uchun loyihalash hamda takomillashtirish amaliyoti sifatida ham taʼriflanadi. Samarali prompt muhandisligi modelning tilni qanday talqin qilishini tushunishni talab etadi. Ushbu jarayon bir nechta namunalik promptlash (few-shot prompting), fikrlash zanjiri (chain-of-thought prompting) va modellarga muayyan rollarni biriktirish kabi usullarni oʻz ichiga oladi.

2020-yillardagi [[Sunʼiy intellekt bumu|AI bumu]] davrida prompt muhandisligi turli korporatsiyalar hamda sanoat tarmoqlarida muhim biznes koʻnikmasi sifatida tan olindi. Tashkilotlarda mehnat undorligi va samaradorlikni oshiruvchi promptlarni yaratish uchun '''prompt muhandisi''' (inglizcha: ''prompt engineer'') lavozimiga xodimlar ishga olindi. Biroq, keyinchalik insonlarga qaraganda yaxshiroq promptlar yarata oladigan AI modellarining paydo boʻlishi va umumiy xodimlar uchun kiritish koʻrsatmalari boʻyicha korporativ treninglar yoʻlga qoʻyilishi natijasida bu alohida lavozimga boʻlgan talab biroz pasaydi.

Keng tarqalgan promptlash usullariga multi-shot, fikrlash zanjiri (chain-of-thought) va fikrlar daraxti (tree-of-thought) usullari, shuningdek, modelga maxsus rollarni yuklash kiradi. [[Axborotni qidirish bilan toʻldirilgan generatsiya]] (RAG) kabi avtomatlashtirilgan prompt generatsiyasi usullari prompt muhandislari uchun yuqori aniqlik hamda kengroq funksional imkoniyatlarni taʼminlaydi. [[Prompt injeksiyasi]] esa [[kiberxavfsizlik]] sohasidagi hujum turi boʻlib, u [[mashinali oʻrgatish]] modellariga gʻarazli promptlar yuborish orqali ularni chalgʻitishga qaratilgan.

== Atamashunoslik ==
[[Oksford ingliz tili lugʻati]] (Oxford English Dictionary) prompt muhandisligiga quyidagicha taʼrif beradi: "Sunʼiy intellekt dasturi, algoritm va boshqalar uchun uning natijasini optimallashtirish yoki kutilgan natijaga erishish maqsadida promptlarni shakllantirish va takomillashtirish harakati yoki jarayoni; ushbu faoliyat bilan shugʻullanuvchi soha yoki kasb."<ref>{{Cite book |url=https://www.oed.com/dictionary/prompt-engineering |title=Oxford English Dictionary |publisher=[[Oxford University Press]] |year=2025 |chapter=prompt engineering}}</ref> 2023-yilda "prompt" soʻzi ("sunʼiy intellekt dasturi yoki algoritmiga beriladigan, u yaratadigan kontentni belgilaydigan yoki unga taʼsir koʻrsatadigan koʻrsatma") Oksfordning [[yil soʻzi]] nominatsiyasida ikkinchi oʻrinni egalladi.<ref>{{Cite web |title=Oxford Word of the Year 2023 |url=https://languages.oup.com/word-of-the-year/2023/ |access-date=2026-02-06 |website=Oxford Languages |publisher=[[Oxford University Press]] |language=en-GB |archive-date=January 31, 2024 |archive-url=https://web.archive.org/web/20240131001517/https://languages.oup.com/word-of-the-year/2023/ |url-status=live }}</ref>

=== Prompt ===
Prompt — bu sunʼiy intellekt (AI) bajarishi kerak boʻlgan vazifani tasvirlaydigan hamda belgilaydigan [[tabiiy til]]dagi matndir.<ref name="language-models-are-multitask">{{cite web |last1=Radford |first1=Alec |last2=Wu |first2=Jeffrey |last3=Child |first3=Rewon |last4=Luan |first4=David |last5=Amodei |first5=Dario |author-link5=Dario Amodei |last6=Sutskever |first6=Ilya |author-link6=Ilya Sutskever |year=2019 |title=Language Models are Unsupervised Multitask Learners |url=https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf |publisher=OpenAI |quote="We demonstrate language models can perform down-stream tasks in a zero-shot setting – without any parameter or architecture modification" |access-date=June 19, 2023 |archive-date=February 6, 2021 |archive-url=https://web.archive.org/web/20210206183945/https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf |url-status=live }}</ref> Matndan matn hosil qiluvchi [[katta til modeli|katta til modeli]] uchun prompt oddiy soʻrov, [[Buyruq (hisoblash)|buyruq]] yoki kontekst, koʻrsatmalar va suhbat tarixiga havolalarni oʻz ichiga olgan batafsil bayonot boʻlishi mumkin. Prompt muhandisligi aniq soʻrovlarni loyihalash, soʻz birikmalarini takomillashtirish, tegishli kontekstni taqdim etish, chiqish uslubini koʻrsatish hamda AI modelini aniqroq, foydaliroq va izchil javoblar berishga yoʻnaltirish uchun maxsus xarakterni taqlid qilishni oʻz ichiga oladi.<ref name="Genkina">{{cite web |last=Genkina |first=Dina |date=2024-03-06 |title=AI Prompt Engineering is Dead: Long live AI prompt engineering |url=https://spectrum.ieee.org/prompt-engineering-is-dead |access-date=2025-01-18 |website=[[IEEE Spectrum]] |archive-date=January 17, 2025 |archive-url=https://web.archive.org/web/20250117210904/https://spectrum.ieee.org/prompt-engineering-is-dead |url-status=live }}</ref><ref name="auto1">{{Cite book |last1=Wahle |first1=Jan Philip |title=Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing |last2=Ruas |first2=Terry |last3=Xu |first3=Yang |last4=Gipp |first4=Bela |date=2024 |publisher=Association for Computational Linguistics |editor-last=Al-Onaizan |editor-first=Yaser |location=Miami, Florida, USA |pages=11004–11033 |chapter=Paraphrase Types Elicit Prompt Engineering Capabilities |doi=10.18653/v1/2024.emnlp-main.617 |editor2-last=Bansal |editor2-first=Mohit |editor3-last=Chen |editor3-first=Yun-Nung |chapter-url=https://aclanthology.org/2024.emnlp-main.617/ |arxiv=2406.19898}}</ref>

Matndan tasvir yoki matndan audio hosil qiluvchi model bilan muloqot qilishda odatiy prompt kutilayotgan natijaning tasvirini oʻz ichiga oladi. Masalan: "Ot minayotgan kosmonavtning yuqori sifatli fotosurati"<ref>{{Cite web |last=Heaven |first=Will Douglas |date=April 6, 2022 |title=This horse-riding astronaut is a milestone on AI's long road towards understanding |url=https://www.technologyreview.com/2022/04/06/1049061/dalle-openai-gpt3-ai-agi-multimodal-image-generation/ |access-date=2023-08-14 |website=MIT Technology Review}}</ref> yoki "Tadbirkorlik namunalariga ega past templi elektron musiqa".<ref>{{cite web |last=Wiggers |first=Kyle |date=2023-06-12 |title=Meta open sources an AI-powered music generator |url=https://techcrunch.com/2023/06/12/meta-open-sources-an-ai-powered-music-generator/ |access-date=2023-08-15 |publisher=TechCrunch |quote=Next, I gave a more complicated prompt to attempt to throw MusicGen for a loop: "Lo-fi slow BPM electro chill with organic samples."}}</ref> Matndan tasvir yaratuvchi modellarda prompt muhandisligi keraksiz obyekt, uslub, maket, yoritish va estetikaga erishish uchun qoʻllaniladi.<ref name=":3">{{Cite web |last=Mittal |first=Aayush |date=2023-07-27 |title=Mastering AI Art: A Concise Guide to Midjourney and Prompt Engineering |url=https://www.unite.ai/mastering-ai-art-a-concise-guide-to-midjourney-and-prompt-engineering/ |access-date=2025-05-09 |website=Unite.AI |language=en-US |archive-date=May 9, 2025 |archive-url=https://web.archive.org/web/20250509164100/https://www.unite.ai/mastering-ai-art-a-concise-guide-to-midjourney-and-prompt-engineering/ |url-status=live }}</ref>

=== Usullar ===
Prompt muhandisligining turli usullarini ifodalash uchun qoʻllaniladigan asosiy atamalarga ''fikrlash zanjiri'' (chain-of-thought),<ref name="weipaper" /> ''fikrlar daraxti'' (tree-of-thought),<ref name=":5">{{Cite conference |date=2023 |title=Tree of Thoughts: Deliberate Problem Solving with Large Language Models |arxiv=2305.10601 |conference=NeurIPS}}</ref> hamda [[axborotni qidirish bilan toʻldirilgan generatsiya]] (RAG) kiradi.<ref name=":6">{{Cite web |date=31 May 2024 |title=Why Google's AI Overviews gets things wrong |url=https://www.technologyreview.com/2024/05/31/1093019/why-are-googles-ai-overviews-results-so-bad/ |access-date=7 March 2025 |website=MIT Technology Review |archive-date=May 3, 2025 |archive-url=https://web.archive.org/web/20250503021705/https://www.technologyreview.com/2024/05/31/1093019/why-are-googles-ai-overviews-results-so-bad/ |url-status=live }}</ref> 2024-yilda ushbu sohada oʻtkazilgan tadqiqot 50 dan ortiq matnli promptlash usullarini, 40 ta koʻp modalli variantlarni va 33 ta asosiy atamani aniqladi. Bu esa sohada standartlashtirilgan terminologiyaning hali shakllanmaganligini koʻrsatadi.<ref name=":4">{{Cite arXiv |eprint=2406.06608 |class=cs.CL |first1=Sander |last1=Schulhoff |first2=Michael |last2=Ilie |title=The Prompt Report: A Systematic Survey of Prompt Engineering Techniques |date=2024 |last3=Balepur |first3=Nishant |last4=Kahadze |first4=Konstantine |last5=Liu |first5=Amanda |last6=Si |first6=Chenglei |last7=Li |first7=Yinheng |last8=Gupta |first8=Aayush |last9=Han |first9=HyoJung |last10=Schulhoff |first10=Sevien |author11=Pranav Sandeep Dulepet |last12=Vidyadhara |first12=Saurav |last13=Ki |first13=Dayeon |last14=Agrawal |first14=Sweta |last15=Pham |first15=Chau |last16=Kroiz |first16=Gerson |last17=Li |first17=Feileen |last18=Tao |first18=Hudson |last19=Srivastava |first19=Ashay |author20=Hevander Da Costa |last21=Gupta |first21=Saloni |last22=Rogers |first22=Megan L. |last23=Goncearenco |first23=Inna |last24=Sarli |first24=Giuseppe |last25=Galynker |first25=Igor |last26=Peskoff |first26=Denis |last27=Carpuat |first27=Marine |last28=White |first28=Jules |last29=Anadkat |first29=Shyamal |last30=Hoyle |first30=Alexander |display-authors=1}}</ref>

Vibe coding (Vayb kodlash) — bu [[sunʼiy intellekt yordamida dasturiy taʼminot yaratish]] usuli boʻlib, unda foydalanuvchi katta til modeliga oʻzi xohlagan funksiyani tavsiflab beradi va kodni generatsiya qilishni AI modeliga topshiradi. 2025-yilda "vibe coding" iborasi [[Kollinz ingliz tili lugʻati]] (Collins Dictionary) tomonidan yil soʻzi deb topildi.<ref>{{Cite web |last=Kolirin |first=Lianne |date=2025-11-06 |title='Vibe coding' named Collins Dictionary's Word of the Year |url=https://www.cnn.com/2025/11/06/vibe-coding-collins-word-year-scli-intl |access-date=2026-02-07 |website=CNN |language=en |archive-date=November 14, 2025 |archive-url=https://web.archive.org/web/20251114002226/https://www.cnn.com/2025/11/06/vibe-coding-collins-word-year-scli-intl |url-status=live }}</ref>

=== Kontekst muhandisligi ===
Kontekst muhandisligi (inglizcha: ''context engineering'') — foydalanuvchi promptlariga hamroh boʻladigan kontekst unsurlarini (tizimiy koʻrsatmalar, qidirib olingan bilimlar, vositalar taʼrifi, suhbat xulosalari hamda vazifa metamaʼlumotlari) boshqarishga qaratilgan turdosh jarayondir. Kontekst muhandisligi Ishlab chiqarish tizimlarida LLM modellarning ishonchliligini, axborot manbalarining aniqligini va tokenlar samaradorligini oshirish uchun amalga oshiriladi.<ref>{{cite web |last=Casey |first=Matt M. |date=2025-11-05 |title=Context Engineering: The Discipline Behind Reliable LLM Applications & Agents |url=https://www.comet.com/site/blog/context-engineering/ |access-date=2025-11-10 |publisher=Comet |archive-date=November 11, 2025 |archive-url=https://web.archive.org/web/20251111224338/https://www.comet.com/site/blog/context-engineering/ |url-status=live }}</ref><ref>{{cite web |date=2025-07-02 |title=Context Engineering |url=https://blog.langchain.com/context-engineering-for-agents/ |access-date=2025-11-10 |publisher=LangChain |archive-date=November 11, 2025 |archive-url=https://web.archive.org/web/20251111102526/https://blog.langchain.com/context-engineering-for-agents/ |url-status=live }}</ref> Bu konseptsiya tokenlar byudjetini rejalashtirish, manba teglarini biriktirish, kontekst vositalarini versiyalash va observability (tizimga qanday kontekst berilganini qayd etish) kabi amaliyotlarga alohida eʼtibor qaratadi.<ref>{{cite arXiv |eprint=2507.13334 |class=cs.CL |first=Lingrui |last=Mei |title=A Survey of Context Engineering for Large Language Models |date=2025-07-17}}</ref>

== Nazariy va amaliy asoslar ==
Oʻtkazilgan tadqiqotlar shuni koʻrsatadiki, [[katta til modeli|katta til modellarining]] (LLM) samaradorligi promptdagi misollarning tartibi, koʻrsatilgan namunalar sifati va hatto jumlalarning kichik oʻzgarishlariga ham juda taʼsirchandir. Baʼzi hollarda, prompt ichida misollar ketma-ketligini oʻzgartirish natijaning aniqligini 40 foizdan ortiqqa oʻzgartirishi kuzatilgan.<ref name=":4" />

=== Kontekst ichidagi oʻrganish (In-context learning) ===
Modelning kiritilgan promptlardan vaqtincha oʻrganish qobiliyati ''kontekst ichidagi oʻrganish'' (in-context learning) deb ataladi. Bu xususiyat katta til modellarining [[katta til modellarining favqulodda qobiliyatlari|favqulodda qobiliyati]] (emergent ability) hisoblanadi.<ref name=":2">{{Cite journal |last1=Wei |first1=Jason |last2=Tay |first2=Yi |last3=Bommasani |first3=Rishi |last4=Raffel |first4=Colin |last5=Zoph |first5=Barret |last6=Borgeaud |first6=Sebastian |last7=Yogatama |first7=Dani |last8=Bosma |first8=Maarten |last9=Zhou |first9=Denny |last10=Metzler |first10=Donald |last11=Chi |first11=Ed H. |last12=Hashimoto |first12=Tatsunori |last13=Vinyals |first13=Oriol |last14=Liang |first14=Percy |last15=Dean |first15=Jeff |date=October 2022 |title=Emergent Abilities of Large Language Models |journal=Transactions on Machine Learning Research |arxiv=2206.07682 |quote="In prompting, a pre-trained language model is given a prompt (e.g. a natural language instruction) of a task and completes the response without any further training or gradient updates to its parameters... The ability to perform a task via few-shot prompting is emergent when a model has random performance until a certain scale, after which performance increases to well-above random" |last16=Fedus |first16=William}}</ref> Kontekst ichida oʻrganish model oʻlchami ortishi bilan namoyon boʻladigan xususiyat boʻlib, kattaroq modellarda uning samaradorligi kichik modellarga qaraganda tezroq oʻsadi.<ref name=":2" /><ref>{{Cite journal |last1=Caballero |first1=Ethan |last2=Gupta |first2=Kshitij |last3=Rish |first3=Irina |last4=Krueger |first4=David |date=2023 |title=Broken Neural Scaling Laws |journal=ICLR |arxiv=2210.14891}}</ref> Tizimiy oʻrgatish va [[sozlash (chuqur oʻrgatish)|nozik sozlashdan]] (fine-tuning) farqli ravishda, kontekst ichidagi oʻrganish vaqtinchalik xarakterga ega.<ref>{{cite web |last1=Musser |first1=George |author-link=George Musser |title=How AI Knows Things No One Told It |url=https://www.scientificamerican.com/article/how-ai-knows-things-no-one-told-it/ |access-date=17 May 2023 |website=[[Scientific American]] |quote="By the time you type a query into ChatGPT, the network should be fixed; unlike humans, it should not continue to learn. So it came as a surprise that LLMs do, in fact, learn from their users' prompts—an ability known as in-context learning." |archive-date=July 3, 2023 |archive-url=https://web.archive.org/web/20230703174606/https://www.scientificamerican.com/article/how-ai-knows-things-no-one-told-it/ |url-status=live }}</ref> Modellarni kontekst ichida oʻrganishga oʻrgatish jarayonini [[meta-oʻrganish]] yoki "oʻrganishni oʻrganish" shakli sifatida koʻrish mumkin.<ref>{{Cite journal |last1=Garg |first1=Shivam |last2=Tsipras |first2=Dimitris |last3=Liang |first3=Percy |last4=Valiant |first4=Gregory |date=2022 |title=What Can Transformers Learn In-Context? A Case Study of Simple Function Classes |url= |journal=NeurIPS |arxiv=2208.01066 |quote=Training a model to perform in-context learning can be viewed as an instance of the more general learning-to-learn or meta-learning paradigm}}</ref>

=== Model taʼsirchanligini baholash ===
Tadqiqotlar shuni koʻrsatadiki, LLM modellar prompt formati, strukturasi hamda lingvistik xususiyatlardagi kichik oʻzgarishlarga ham oʻta taʼsirchandir. Baʼzi tadqiqotlarda format oʻzgartirilganda aniqlik darajasi 76 foizgacha farq qilishi aniqlangan.<ref name="auto">{{Cite conference |date=2024 |title=Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting |conference=ICLR |arxiv=2310.11324}}</ref> Morfologik, sintaktik va leksik-semantik oʻzgarishlar kabi lingvistik xususiyatlar prompt samaradorligiga sezilarli taʼsir koʻrsatadi.<ref name="auto1" /><ref>{{Cite journal |last1=Leidinger |first1=Alina |last2=van Rooij |first2=Robert |last3=Shutova |first3=Ekaterina |date=2023 |editor-last=Bouamor |editor-first=Houda |editor2-last=Pino |editor2-first=Juan |editor3-last=Bali |editor3-first=Kalika |title=The language of prompting: What linguistic properties make a prompt successful? |url=https://aclanthology.org/2023.findings-emnlp.618/ |journal=Findings of the Association for Computational Linguistics: EMNLP 2023 |location=Singapore |publisher=Association for Computational Linguistics |pages=9210–9232 |arxiv=2311.01967 |doi=10.18653/v1/2023.findings-emnlp.618 |archive-date=December 13, 2024 |access-date=November 18, 2024 |archive-url=https://web.archive.org/web/20241213140628/https://aclanthology.org/2023.findings-emnlp.618/ |url-status=live }}</ref> Masalan, ergashgan gapli sintaksis axborotni qidirishda izchillikni oshiradi va noaniqlikni kamaytiradi.<ref>{{Cite book |last1=Linzbach |first1=Stephan |title=Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) |last2=Dimitrov |first2=Dimitar |last3=Kallmeyer |first3=Laura |last4=Evang |first4=Kilian |last5=Jabeen |first5=Hajira |last6=Dietze |first6=Stefan |date=June 2024 |publisher=Association for Computational Linguistics |editor-last=Duh |editor-first=Kevin |location=Mexico City, Mexico |pages=3645–3655 |chapter=Dissecting Paraphrases: The Impact of Prompt Syntax and supplementary Information on Knowledge Retrieval from Pretrained Language Models |doi=10.18653/v1/2024.naacl-long.201 |editor2-last=Gomez |editor2-first=Helena |editor3-last=Bethard |editor3-first=Steven |chapter-url=https://aclanthology.org/2024.naacl-long.201/ |arxiv=2404.01992 |archive-date=November 3, 2024 |access-date=November 18, 2024 |archive-url=https://web.archive.org/web/20241103184605/https://aclanthology.org/2024.naacl-long.201/ |url-status=live }}</ref> Ushbu taʼsirchanlik hatto model hajmi kattalashtirilganda yoki qoʻshimcha namunalar berilganda ham saqlanib qoladi.

Modellarning taʼsirchanligini baholash uchun maxsus usullar taklif etilgan. FormatSpread tizimi turli xil prompt formatlarini baholash orqali tizimli tahlilni amalga oshiradi.<ref name="auto" /> PromptEval esa turli promptlar boʻyicha samaradorlik taqsimotini baholaydi va cheklangan byudjet sharoitida aniq baholash imkonini beradi.<ref>{{Cite conference |date=2024 |title=Efficient multi-prompt evaluation of LLMs |conference=NeurIPS |arxiv=2405.17202}}</ref>

== Promptlash usullari ==

=== Multi-shot (Koʻp namunalik promptlash) ===
Prompt modelga kontekst ichida oʻrganish uchun bir nechta misollarni oʻz ichiga olishi mumkin, bu LLMlar uchun [[bir nechta namunalik oʻrganish]] (few-shot learning) usulidir.<ref name="weipaper" /><ref>{{Cite journal |last1=Brown |first1=Tom |last2=Mann |first2=Benjamin |last3=Ryder |first3=Nick |last4=Subbiah |first4=Melanie |last5=Kaplan |first5=Jared D. |last6=Dhariwal |first6=Prafulla |last7=Neelakantan |first7=Arvind |year=2020 |title=Language models are few-shot learners |journal=Advances in Neural Information Processing Systems |volume=33 |pages=1877–1901 |arxiv=2005.14165}}</ref> Masalan, prompt modeldan quyidagi ketma-ketlikni toʻldirishni soʻrashi mumkin: "''maison'' → uy, ''chat'' → mushuk, ''chien'' → ", va kutilayotgan javob "it" soʻzi boʻladi.<ref>{{Cite journal |last1=Garg |first1=Shivam |last2=Tsipras |first2=Dimitris |last3=Liang |first3=Percy |author3-link=Percy Liang |last4=Valiant |first4=Gregory |date=2022 |title=What Can Transformers Learn In-Context? A Case Study of Simple Function Classes |journal=NeurIPS |arxiv=2208.01066}}</ref>

=== Chain-of-thought (Fikrlash zanjiri) ===
''Fikrlash zanjiri'' (Chain-of-thought / CoT) — bu [[katta til modeli|katta til modellariga]] (LLM) yakuniy javob berishdan oldin muammoni ketma-ket oraliq bosqichlar koʻrinishida hal qilish imkonini beruvchi usuldir. 2022-yilda [[Google Brain]] tadqiqotchilari fikrlash zanjiri usuli modelni fikrlash bosqichlari bilan javob berishga undash orqali uning [[avtomatlashtirilgan mantiqiy fikrlash]] qobiliyatini yaxshilashini maʼlum qildi.<ref name="weipaper" /><ref>{{Cite web |author=Narang |first1=Sharan |last2=Chowdhery |first2=Aakanksha |date=2022-04-04 |title=Pathways Language Model (PaLM): Scaling to 540 Billion Parameters for Breakthrough Performance |url=https://ai.googleblog.com/2022/04/pathways-language-model-palm-scaling-to.html |website=ai.googleblog.com |access-date=August 14, 2023 |archive-date=April 4, 2022 |archive-url=https://web.archive.org/web/20220404161447/https://ai.googleblog.com/2022/04/pathways-language-model-palm-scaling-to.html |url-status=live }}</ref> Ushbu usul LLM modellarga arifmetik yoki mantiqiy masalalarni bosqichma-bosqich yechishda yordam berish uchun ishlab chiqilgan.<ref>{{cite web |last1=Dang |first1=Ekta |date=8 February 2023 |title=Harnessing the power of GPT-3 in scientific research |url=https://venturebeat.com/ai/harnessing-the-power-of-gpt-3-in-scientific-research/ |access-date=10 March 2023 |website=VentureBeat |archive-date=March 18, 2023 |archive-url=https://web.archive.org/web/20230318050303/https://venturebeat.com/ai/harnessing-the-power-of-gpt-3-in-scientific-research/ |url-status=live }}</ref>

540 milliard parametrga ega [[PaLM]] til modelida qoʻllanilganda, CoT usuli modelga GSM8K matematik mezonlarida (benchmark) eng yuqori natijalarni koʻrsatishga yordam berdi.<ref name="weipaper" /> Modellarni CoT maʼlumotlar toplamida nozik sozlash orqali ushbu imkoniyatni yanada oshirish mumkin.<ref>{{Cite journal |date=2024 |title=Scaling Instruction-Finetuned Language Models |url=https://jmlr.org/papers/volume25/23-0870/23-0870.pdf |journal=Journal of Machine Learning Research |archive-date=May 8, 2025 |access-date=May 9, 2025 |archive-url=https://web.archive.org/web/20250508091413/http://jmlr.org/papers/volume25/23-0870/23-0870.pdf |url-status=live }}</ref> Shuningdek, promptga shunchaki "Keling, buni bosqichma-bosqich oʻylab koʻraylik" (Let's think step-by-step) iborasini qoʻshish ham CoT usulini nol namunalik (zero-shot) rejimida samarali qoʻllash imkonini beradi.<ref name=":1">{{Cite journal |last1=Kojima |first1=Takeshi |author2=Shixiang Shane Gu |last3=Reid |first3=Machel |last4=Matsuo |first4=Yutaka |last5=Iwasawa |first5=Yusuke |date=2022 |title=Large Language Models are Zero-Shot Reasoners |journal=NeurIPS |arxiv=2205.11916}}</ref>

==== Self-consistency (Oʻz-oʻziga moslik) ====
''Self-consistency'' usuli bir nechta fikrlash zanjirlarini amalga oshiradi va ularning orasidan eng koʻp takrorlangan va mantiqan toʻgʻri kelgan yakuniy xulosani tanlab oladi.<ref>{{Cite conference |date=2023 |title=Self-Consistency Improves Chain of Thought Reasoning in Language Models |conference=ICLR |arxiv=2203.11171}}</ref>

=== Tree-of-thought (Fikrlar daraxti) ===
''Tree-of-thought'' usuli fikrlash zanjirini umumlashtirib, bir vaqtning oʻzida bir nechta fikrlash yoʻnalishlarini (shoxlarini) generatsiya qiladi hamda ortga qaytish yoki boshqa yoʻllarni oʻrganish imkonini beradi. U [[kenglik boʻyicha qidirish]], [[chuqurlik boʻyicha qidirish]] yoki [[nurli qidirish]] (beam search) kabi [[daraxt boʻyicha qidirish algoritmlaridan]] foydalanadi.<ref name=":5" />

=== Matndan tasvir yaratuvchi promptlash ===
2022-yilda [[DALL-E 2]], [[Stable Diffusion]] va [[Midjourney]] kabi [[matndan tasvir yaratuvchi]] modellar ommaga taqdim etildi. Ushbu modellar matnli promptlarni qabul qiladi va unga mos tasvirlarni generatsiya qiladi.<ref>{{Cite web |last=Goldman |first=Sharon |date=2023-01-05 |title=Two years after DALL-E debut, its inventor is "surprised" by impact |url=https://venturebeat.com/ai/two-years-after-dall-e-debut-its-inventor-is-surprised-by-impact/ |access-date=2025-05-09 |website=VentureBeat |language=en-US |archive-date=January 18, 2025 |archive-url=https://web.archive.org/web/20250118182528/https://venturebeat.com/ai/two-years-after-dall-e-debut-its-inventor-is-surprised-by-impact/ |url-status=live }}</ref>

Tasviriy prompt odatda sanʼat obyektining tavsifi, vosita (masalan, raqamli tasvir yoki fotosurat), uslub (masalan, giperrealizm), yoritish va teksturani oʻz ichiga oladi.<ref>{{Cite web |date=2023-05-14 |title=Stable Diffusion prompt: a definitive guide |url=https://stable-diffusion-art.com/prompt-guide/ |access-date=2023-08-14 |archive-date=August 14, 2023 |archive-url=https://web.archive.org/web/20230814211136/https://stable-diffusion-art.com/prompt-guide/ |url-status=live }}</ref> Soʻzlarning ketma-ketligi ham tasvir natijasiga taʼsir qiladi: prompt boshida kelgan soʻzlarga koʻproq urgʻu beriladi.<ref name="diab">{{Cite web |last1=Diab |first1=Mohamad |last2=Herrera |first2=Julian |last3=Chernow |first3=Bob |date=2022-10-28 |title=Stable Diffusion Prompt Book |url=https://cdn.openart.ai/assets/Stable%20Diffusion%20Prompt%20Book%20From%20OpenArt%2011-13.pdf |access-date=2023-08-07 |quote="Prompt engineering is the process of structuring words that can be interpreted and understood by a ''text-to-image'' model." |archive-date=March 30, 2023 |archive-url=https://web.archive.org/web/20230330010200/https://cdn.openart.ai/assets/Stable%20Diffusion%20Prompt%20Book%20From%20OpenArt%2011-13.pdf |url-status=live }}</ref>

==== Rassom uslublari ====
Baʼzi matndan tasvir yaratuvchi modellar muayyan rassomlarning uslubiga taqlid qilish imkoniyatiga ega. Masalan, ''Greg Rutkovski uslubida'' iborasi polshalik raqamli rassom [[Greg Rutkowski|Greg Rutkovskining]] oʻziga xos uslubida tasvirlar yaratish uchun ishlatilgan.<ref>{{Cite web |last=Heikkilä |first=Melissa |date=2022-09-16 |title=This Artist Is Dominating AI-Generated Art and He's Not Happy About It |url=https://www.technologyreview.com/2022/09/16/1059598/this-artist-is-dominating-ai-generated-art-and-hes-not-happy-about-it/ |access-date=2023-08-14 |website=MIT Technology Review |archive-date=January 14, 2023 |archive-url=https://web.archive.org/web/20230114125952/https://www.technologyreview.com/2022/09/16/1059598/this-artist-is-dominating-ai-generated-art-and-hes-not-happy-about-it/ |url-status=live }}</ref> Shuningdek, [[Vinsent van Gog]] va [[Salvador Dalí]] kabi mashhur rassomlar uslubidan ham keng foydalaniladi.<ref>{{Cite web |last=Solomon |first=Tessa |date=2024-08-28 |title=The AI-Powered Ask Dalí and Hello Vincent Installations Raise Uncomfortable Questions about Ventriloquizing the Dead |url=https://www.artnews.com/art-news/issue/salvador-dali-vincent-van-gogh-ai-installations-ethics-1234714954/ |access-date=2025-01-10 |website=ARTnews.com |language=en-US |archive-date=January 24, 2025 |archive-url=https://web.archive.org/web/20250124074321/https://www.artnews.com/art-news/issue/salvador-dali-vincent-van-gogh-ai-installations-ethics-1234714954/ |url-status=live }}</ref>

==== Matn inversiyasi va embeddinglar ====
Matndan tasvir yaratuvchi modellarda ''matn inversiyasi'' (textual inversion) namunaviy tasvirlar toʻplami asosida yangi soʻz embeddingini loyihalashtiradi. Ushbu embedding vektori prompt tarkibida misollar uslubini aks ettiruvchi "psevdo-soʻz" vazifasini oʻtaydi.<ref>{{Cite journal |last1=Gal |first1=Rinon |last2=Alaluf |first2=Yuval |last3=Atzmon |first3=Yuval |last4=Patashnik |first4=Or |last5=Bermano |first5=Amit H. |last6=Chechik |first6=Gal |last7=Cohen-Or |first7=Daniel |date=2023 |title=An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion |journal=ICLR |arxiv=2208.01618}}</ref>

=== Tasviriy promptlash ===
2023-yilda [[Meta Platforms]] kompaniyasi Segment Anything deb nomlangan [[kompyuterli koʻrish]] modelini taqdim etdi. Ushbu model promptlar orqali [[tasvir segmentatsiyasini]] bajarishi mumkin. Matnli promptlarga muqobil ravishda, u chegaralovchiramkalar va segmentatsiya niqoblarini qabul qiladi.<ref name="Kirillov">{{Cite conference |date=2023 |title=Segment Anything |url=https://openaccess.thecvf.com/content/ICCV2023/papers/Kirillov_Segment_Anything_ICCV_2023_paper.pdf |conference=ICCV |access-date=May 9, 2025 |archive-date=April 28, 2025 |archive-url=https://web.archive.org/web/20250428221529/https://openaccess.thecvf.com/content/ICCV2023/papers/Kirillov_Segment_Anything_ICCV_2023_paper.pdf |url-status=live }}</ref>

=== Cheklovlar ===
Prompt yozish va uni takomillashtirish jarayoni muhandislik loyihalash amaliyotiga oʻxshasa-da, bir qancha fundamental cheklovlar mavjud. Samarali prompt strategiyalari muayyan modelga oʻta darajada bogʻliq boʻlib, bir modelda yaxshi ishlagan usul boshqasida natijani yomonlashtirishi mumkin. Shuningdek, soʻzlarning joylashuvi yoki tinchlik belgilaridagi kichik oʻzgarishlar ham model natijasini keskin oʻzgartirib yuborishi mumkin.<ref>{{Cite journal |last=Berry |first=David M. |date=2026-06-02 |title=Prompt anxiety and the algorithmic politics of uncertainty |url=https://doi.org/10.1007/s00146-026-03093-8 |journal=AI & Society |language=en |doi=10.1007/s00146-026-03093-8 |issn=1435-5655}}</ref>

Bundan tashqari, AI modellarining foydalanuvchi maqsadini toʻgʻridan-toʻgʻri tushunish qobiliyati oshib borayotgani sababli, koʻplab qoʻlda bajariladigan promptlash usullari oʻz ahamiyatini yoʻqotmoqda.<ref>{{Cite news |last=Bousquette |first=Isabelle |date=2025-04-25 |title=The Hottest AI Job of 2023 Is Already Obsolete |url=https://www.wsj.com/articles/the-hottest-ai-job-of-2023-is-already-obsolete-1961b054 |access-date=2026-05-24 |work=The Wall Street Journal |language=en-US |issn=0099-9660 |archive-date=May 7, 2025 |archive-url=https://web.archive.org/web/20250507053042/https://www.wsj.com/articles/the-hottest-ai-job-of-2023-is-already-obsolete-1961b054 |url-status=live }}</ref>

== Avtomatlashtirilgan prompt generatsiyasi ==
Soʻnggi tadqiqotlar inson aralashuvisiz promptlarni generatsiya qilish va optimallashtirish uchun avtomatlashtirilgan algoritmlarni oʻrganmoqda. Ushbu avtomatik usullar model gradientlari hamda evolyutsion jarayonlarni tahlil qilish orqali samarali prompt namuna turlarini aniqlashni maqsad qiladi.<ref>{{Cite arXiv |eprint=2502.11560 |class=cs.AI |first1=Wenwu |last1=Li |first2=Xiangfeng |last2=Wang |title=A Survey of Automatic Prompt Engineering: An Optimization Perspective |date=2025 |last3=Li |first3=Wenhao |last4=Jin |first4=Bo}}</ref>

=== Axborotni qidirish bilan toʻldirilgan generatsiya (RAG) ===
[[Axborotni qidirish bilan toʻldirilgan generatsiya]] (Retrieval-augmented generation / RAG) — GenAI modellariga yangi axborotlarni qidirib topish hamda ularni javobga biriktirish imkonini beruvchi texnologiyadir. U foydalanuvchi soʻrovlariga javob berishda tashqi hujjatlar bazasiga murojaat qiladi va modelning mavjud [[oʻrgatish maʼlumotlari|oʻrgatish maʼlumotlarini]] boyitadi.<ref name=":6" />

RAG modeli javob generatsiya qilishdan oldin [[axborot qidiruvi]] (information retrieval) bosqichini amalga oshiradi. U maʼlumotlar bazalaridan va yuklangan hujjatlardan tegishli matnlarni dinamik ravishda tortib oladi. Bu esa AI modelining qayta oʻrgatishsiz aniqroq javob berishiga va [[sunʼiy intellekt gallyutsinatsiyasi|AI gallyutsinatsiyalarini]] kamaytirishga xizmat qiladi.<ref>{{Cite web |date=6 June 2024 |title=Can a technology called RAG keep AI models from making stuff up? |url=https://arstechnica.com/ai/2024/06/can-a-technology-called-rag-keep-ai-models-from-making-stuff-up/ |access-date=7 March 2025 |website=Ars Technica |archive-date=June 6, 2024 |archive-url=https://web.archive.org/web/20240606121243/https://arstechnica.com/ai/2024/06/can-a-technology-called-rag-keep-ai-models-from-making-stuff-up/ |url-status=live }}</ref>

=== Grafli axborot qidiruvi bilan toʻldirilgan generatsiya (GraphRAG) ===
GraphRAG ([[Microsoft Research]] tomonidan kiritilgan) — RAG texnologiyasini [[bilimlar grafi]] bilan kengaytiruvchi usul boʻlib, u modelga turli maʼlumotlar toʻplamlarini bogʻlash hamda umumlashtirilgan semantik tushunchalarni tahlil qilish imkonini beradi.<ref>{{citation |last1=Larson |first1=Jonathan |title=GraphRAG: Unlocking LLM discovery on narrative private data |date=February 13, 2024 |url=https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data/ |publisher=Microsoft |last2=Truitt |first2=Steven |access-date=May 7, 2024 |archive-date=May 7, 2024 |archive-url=https://web.archive.org/web/20240507061745/https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data/ |url-status=live }}</ref><ref>{{Cite web |title=An Introduction to Graph RAG |url=https://www.kdnuggets.com/an-introduction-to-graph-rag |access-date=2025-05-09 |website=KDnuggets |language=en-US |archive-date=August 20, 2025 |archive-url=https://web.archive.org/web/20250820103627/https://www.kdnuggets.com/an-introduction-to-graph-rag |url-status=live }}</ref>

=== Prompt yaratishda til modellaridan foydalanish ===
Til modellarining oʻzidan boshqa til modellari uchun promptlar tayyorlashda foydalanish mumkin.<ref>{{Cite conference |date=2023 |title=Explaining Patterns in Data with Language Models via Interpretable Autoprompting |url=https://aclanthology.org/2023.blackboxnlp-1.3.pdf |conference=BlackboxNLP Workshop |arxiv=2210.01848 |access-date=May 9, 2025 |archive-date=May 9, 2025 |archive-url=https://web.archive.org/web/20250509161046/https://aclanthology.org/2023.blackboxnlp-1.3.pdf |url-status=live }}</ref> "Avtomatik prompt muhandisi" (Automatic Prompt Engineer) algoritmi quyidagicha ishlaydi:
* Ikkita LLM qoʻllaniladi: maqsadli LLM va promptlovchi LLM.
* Promptlovchi LLMga namuna kiritish-chiqarish juftliklari beriladi hamda ushbu natijalarni keltirib chiqarishi mumkin boʻlgan koʻrsatmalarni generatsiya qilish buyuriladi.
* Har bir koʻrsatgich baholanadi va eng yuqori ball olgan koʻrsatgichlar qayta takomillashtiriladi.

Auto-CoT usulida esa savollar kutubxonasi [[BERT (til modeli)|BERT]] kabi modellar orqali vektorlarga oʻtkaziladi hamda [[klaster tahlili|klasterlanadi]]. Har bir klaster markaziga yaqin savollar tanlab olinadi va model ularga zero-shot CoT rejimida javob beradi.<ref>{{Cite conference |date=2023 |title=Automatic Chain of Thought Prompting in Large Language Models |conference=ICLR |arxiv=2210.03493}}</ref>

=== Avtomatik prompt optimallashtirish ===
Avtomatik prompt optimallashtirish usullari maxsus maʼlumotlar toʻplamidan foydalanib, muqobil prompt matnlarini qidirish orqali natijani yaxshilaydi:
* MIPRO (Multi-prompt Instruction Proposal Optimizer) koʻp bosqichli til modellari koʻrsatmalarini optimallashtiradi.<ref>{{Cite conference |last1=Opsahl-Ong |first1=Krista |last2=Ryan |first2=Michael J. |last3=Purtell |first3=Josh |last4=Broman |first4=David |last5=Potts |first5=Christopher |last6=Zaharia |first6=Matei |last7=Khattab |first7=Omar |date=2024 |title=Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs |conference=Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP) |location=Miami, Florida |publisher=Association for Computational Linguistics |arxiv=2406.11695 |doi=10.18653/v1/2024.emnlp-main.525}}</ref>
* GEPA (Genetic-Pareto) evolyutsion qidiruv orqali murojaatlarni tahlil qiladi va [[mustahkamlovchi oʻrgatish]] (reinforcement learning) usullariga qaraganda yuqori samaradorlik beradi.<ref>{{Cite arXiv |eprint=2507.19457 |class=cs.CL |first=Lakshya A. |last=Agrawal |title=GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning |date=2025}}</ref>
* DSPy va Opik kabi ochiq kodli freymvorklar ushbu optimallashtirish vositalarini dasturlash konveyerlariga biriktirish imkonini beradi.<ref>{{Cite arXiv |eprint=2310.03714 |class=cs.CL |first=Omar |last=Khattab |title=DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines |date=2023}}</ref>

=== Gradient tushishi orqali promptlarni qidirish ===
Prefix-tuning yoki prompt tuning usullarida suzuvchi nuqtali vektorlar bevosita [[gradient tushishi]] (gradient descent) orqali optimal koʻrinishga keltiriladi.<ref>{{cite book |last1=Li |first1=Xiang Lisa |title=Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) |last2=Liang |first2=Percy |year=2021 |pages=4582–4597 |chapter=Prefix-Tuning: Optimizing Continuous Prompts for Generation |doi=10.18653/V1/2021.ACL-LONG.353 |s2cid=230433941}}</ref><ref>{{cite book |last1=Lester |first1=Brian |title=Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing |last2=Al-Rfou |first2=Rami |last3=Constant |first3=Noah |year=2021 |pages=3045–3059 |chapter=The Power of Scale for Parameter-Efficient Prompt Tuning |doi=10.18653/V1/2021.EMNLP-MAIN.243 |s2cid=233296808 |arxiv=2104.08691}}</ref> Maskalangan til modellarida esa AutoPrompt algoritmi tokenlar ketma-ketligini izlash uchun gradient tushishi gʻoyasidan foydalanadi.<ref>{{Cite book |last1=Shin |first1=Taylor |title=Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) |last2=Razeghi |first2=Yasaman |last3=Logan IV |first3=Robert L. |last4=Wallace |first4=Eric |last5=Singh |first5=Sameer |date=November 2020 |publisher=Association for Computational Linguistics |location=Online |pages=4222–4235 |chapter=AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Prompts |doi=10.18653/v1/2020.emnlp-main.346 |chapter-url=https://aclanthology.org/2020.emnlp-main.346 |doi-access=free |s2cid=226222232 |archive-date=June 11, 2023 |access-date=June 11, 2023 |archive-url=https://web.archive.org/web/20230611194552/https://aclanthology.org/2020.emnlp-main.346/ |url-status=live }}</ref>

== Tarixi ==
Strukturalangan muloqot tizimlarining dastlabki namunalari 1990-yillardagi korporativ avtomatlashtirish dasturlarida koʻringan. Masalan, Krishna K. Mukerji tomonidan yaratilgan Intelligent Filling Manager (1999) tizimi hujjatlarni avtomatik yaratish uchun savol-javob interfeysidan foydalangan.<ref>{{Cite conference |last=Mukherjee |first=Krishna C. |year=1999 |title=Automating Forms Publishing with the Intelligent Filing Manager |conference=IEEE International Conference on Systems, Man, and Cybernetics |pages=378–383 |doi=10.1109/ICSMC.1999.814108 }}</ref> 2018-yilda tadqiqotchilar [[tabiiy tilni qayta ishlash]] (NLP) sohasidagi barcha alohida vazifalarni umumiy savol-javob shakliga oʻtkazish mumkinligini taklif qildi va koʻp vazifali ilk modelni oʻrgatdi.<ref>{{Cite conference |last1=McCann |first1=Bryan |last2=Keskar |first2=Nitish |last3=Xiong |first3=Caiming |last4=Socher |first4=Richard| date=2018-06-20 |title=The Natural Language Decathlon: Multitask Learning as Question Answering |arxiv=1806.08730 |conference=ICLR}}</ref>

[[Sunʼiy intellekt bumu|AI bumu]] davrida modeldan kutilgan natijani olish hamda [[sunʼiy intellekt gallyutsinatsiyasi|gallyutsinatsiyalarning]] oldini olish uchun promptlash usullarini qoʻllashga eʼtibor oshdi.<ref>{{Cite journal |last1=Knoth |first1=Nils |last2=Tolzin |first2=Antonia |last3=Janson |first3=Andreas |last4=Leimeister |first4=Jan Marco |date=2024-06-01 |title=AI literacy and its implications for prompt engineering strategies |journal=Computers and Education: Artificial Intelligence |volume=6 |article-number=100225 |doi=10.1016/j.caeai.2024.100225 |issn=2666-920X|doi-access=free }}</ref> 2022-yil noyabr oyida [[ChatGPT]] taqdim etilgach, prompt muhandisligi tez orada muhim biznes koʻnikmasiga aylandi va korxonalar maxsus mutaxassislarni yollay boshladi.<ref name="Genkina"/> [[The Wall Street Journal]] maʼlumotlariga koʻra, prompt muhandisi kasbi 2023-yilda eng xaridorgir kasblardan biri boʻlgan boʻlsa-da, modellar intuitsiyasining rivojlanishi tufayli u oʻzining avvalgi dolzarbligini biroz yoʻqotdi.<ref>{{Cite news |last=Bousquette |first=Isabelle |date=2025-04-25 |title=The Hottest AI Job of 2023 Is Already Obsolete |url=https://www.wsj.com/articles/the-hottest-ai-job-of-2023-is-already-obsolete-1961b054 |access-date=2025-05-07 |work=Wall Street Journal |language=en-US |issn=0099-9660 |archive-date=May 7, 2025 |archive-url=https://web.archive.org/web/20250507053042/https://www.wsj.com/articles/the-hottest-ai-job-of-2023-is-already-obsolete-1961b054 |url-status=live }}</ref>

2022-yilda [[Google]] tadqiqotchilari tomonidan ''fikrlash zanjiri'' (chain-of-thought) usuli taklif etildi.<ref name="weipaper">{{cite conference |last1=Wei |first1=Jason |last2=Wang |first2=Xuezhi |last3=Schuurmans |first3=Dale |last4=Bosma |first4=Maarten |last5=Ichter |first5=Brian |last6=Xia |first6=Fei |last7=Chi |first7=Ed H. |last8=Le |first8=Quoc V. |last9=Zhou |first9=Denny |date=31 October 2022 |title=Chain-of-Thought Prompting Elicits Reasoning in Large Language Models |conference=Advances in Neural Information Processing Systems (NeurIPS 2022) |language=en |volume=35 |arxiv=2201.11903}}</ref><ref>{{Cite web |last=Brubaker |first=Ben |date=2024-03-21 |title=How Chain-of-Thought Reasoning Helps Neural Networks Compute |url=https://www.quantamagazine.org/how-chain-of-thought-reasoning-helps-neural-networks-compute-20240321/ |access-date=2025-05-09 |website=Quanta Magazine |language=en |archive-date=May 9, 2025 |archive-url=https://web.archive.org/web/20250509153705/https://www.quantamagazine.org/how-chain-of-thought-reasoning-helps-neural-networks-compute-20240321/ |url-status=live }}</ref> 2023-2024 yillarda esa koʻplab ochiq prompt bazalari ommaga taqdim etildi.<ref>{{Cite web|url=https://www.nytimes.com/2023/06/23/technology/ai-chatbot-life-coach.html|title=How to Turn Your Chatbot Into a Life Coach|last=Chen|first=Brian X.|date=2023-06-23|website=The New York Times|access-date=August 14, 2023|archive-date=August 14, 2023|archive-url=https://web.archive.org/web/20230814211132/https://www.nytimes.com/2023/06/23/technology/ai-chatbot-life-coach.html|url-status=live}}</ref><ref>{{Cite book |last1=Chen |first1=Zijie |last2=Zhang |first2=Lichao |last3=Weng |first3=Fangsheng |last4=Pan |first4=Lili |last5=Lan |first5=Zhenzhong |chapter=Tailored Visions: Enhancing Text-to-Image Generation with Personalized Prompt Rewriting |date=2024-06-16 |title=2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) |chapter-url=https://doi.org/10.1109/cvpr52733.2024.00738 |publisher=IEEE |pages=7727–7736 |doi=10.1109/cvpr52733.2024.00738|arxiv=2310.08129 |isbn=979-8-3503-5300-6 }}</ref>

== Prompt injeksiyasi ==
{{Asosiy maqola|Prompt injeksiyasi}}

Prompt injeksiyasi (inglizcha: ''prompt injection'') — bu [[kiberxavfsizlik]] sohasidagi zaiflik boʻlib, unda tajovuzkorlar kiritadigan gʻarazli maʼlumotlar [[mashinali oʻrgatish]] va [[katta til modeli|katta til modellarining]] kutilmagan harakatlarni bajarishiga olib keladi. Ushbu hujum modelning dasturchi koʻrsatmalari bilan foydalanuvchi kiritgan maʼlumotlarni ajrata olmasligidan foydalanib, xavfsizlik cheklovlarini aylanib oʻtadi.<ref>{{Cite web |last=Vigliarolo |first=Brandon |date=19 September 2022 |title=GPT-3 'prompt injection' attack causes bot bad manners |url=https://www.theregister.com/2022/09/19/in_brief_security/ |access-date=2023-02-09 |website=[[The Register]] |language=en |archive-date=March 29, 2023 |archive-url=https://web.archive.org/web/20230329110712/https://www.theregister.com/2022/09/19/in_brief_security/ |url-status=live }}</ref><ref>{{Cite web |date=26 March 2024 |title=What is a prompt injection attack? |url=https://www.ibm.com/think/topics/prompt-injection |access-date=7 March 2025 |website=IBM |archive-date=March 3, 2025 |archive-url=https://web.archive.org/web/20250303201800/https://www.ibm.com/think/topics/prompt-injection |url-status=live }}</ref>

== Manbalar ==
{{Manbalar}}

[[Turkum:Chuqur oʻrgatish]]
[[Turkum:Mashinali oʻrgatish]]
[[Turkum:Tabiiy tilni qayta ishlash]]
[[Turkum:Generativ sunʼiy intellekt]]
[[Turkum:Sunʼiy intellekt]]