Part 1 · Cockpit

1. 2 Cockpits, 1 Discipline

Claude aur ChatGPT ko sath sath khola jaye to dono dekhne mein kaafi similar lagte hain. Dono mein previous chats ki list, message box, model ya reasoning control, files aur tools ka menu, projects ya workspaces, memory aur instructions, aur search-research features shamil hain.

Jo skills dono jagah kaam aati hain, wo ye hain:

• Task ko clearly describe karna
• Useful context provide karna
• Kya delegate karna hai, ye decide karna
• Answers ko check karna
• Kab stronger model ki taraf jana hai, ye pata hona
• Sensitive information protect karna

Jo cheez different hai wo hai product ke names, layout, aur strengths, aur ye waqt ke sath badalti rehti hain.

Dono ko jaanna isliye zaroori hai kyunki Claude aur ChatGPT har task pe same performance nahi dete. Jo model ek type ke kaam mein excellent hai, wo doosre type ke kaam mein kamzor ho sakta hai. Isay platform awareness kehte hain, yani ye pata hona ke har tool kis cheez mein achha hai, ek hi interface se loyal ban jane ki jagah.

Dono companies chat se aage bhi desktop apps, coding tools, aur browser agents dete hain, lekin ye course sirf chat window tak limited rehta hai.

Sabse Important Beginner Habit

“Assistant se aise baat karo jaise ek capable colleague se karte ho.”

Koi bhi serious request bhejne se pehle 3 cheezein cover karo:

1. 1

   Stage Set Karo

   Aap kaun ho aur actually kya achieve karna chahte ho?

2. 2

   Task Define Karo

   Assistant ko exactly kya karna chahiye?

3. 3

   Rules Set Karo

   Kaunsa tone, format, constraints, ya examples follow karne chahiyen?

“Pattern seekho, button ki location nahi.”

2. Models Aur Thinking Modes

Claude aur ChatGPT, dono aapko ye choose karne dete hain ke kisi task pe kitni capability spend karni hai. Wrong starting point model names hain, kyunki wo fast badalte hain. Right starting point 3 tiers wala pattern hai.

Claude ke model names: Haiku (sabse fast aur cheap), Sonnet (everyday choice), aur Opus (heavy flagship). Haiku aur Sonnet fast tier mein aate hain, Opus flagship tier mein. Thinking ek separate switch hai, model ka naam nahi.

ChatGPT ke model names: Instant (everyday) se le kar Pro (hardest kaam) tak, aur beech mein thinking ke levels.

2 Simple Rules

1. 1

   Rule 1 · Fast Se Shuru Karo

   Zyada tar everyday kaam ke liye normal fast mode use karo. Chhote rewrites, summaries, ya simple explanations ke liye heavy reasoning ka wait mat karo.

2. 2

   Rule 2 · Hard Task Pe Escalate Karo

   Multi-step logic, careful comparison, mathematics, mushkil debugging, ya code jo correct hona zaroori ho, in ke liye thinking mode ya stronger model use karo.

Wo rule jo har new model release ke baad bhi qaim rehta hai: model ko task ke hisab se pick karo, habit ke hisab se nahi. Everyday ke liye fast, hard reasoning ke liye thinking, aur genuinely hard kaam ke liye flagship. Pehle reasoning ka level choose karo, phir current model names seekho.

3. Wo Context Jo Aap Attach Karte Ho

Ek language model sirf usi information pe kaam kar sakta hai jo uske saamne ho. Isi liye attachments matter karte hain. Supported file types mein PDFs, Word documents, spreadsheets, CSV files, images, screenshots, aur code files shamil hain.

Jab aap koi file attach karte ho, to aap usi conversation ke liye assistant ko zyada context de rahe hote ho. Ye comparison dekho:

Words almost same hain. Doosra request isliye useful hai kyunki assistant actually contract dekh sakta hai. Agar aapko kisi dashboard, error message, chart, ya interface mein help chahiye, to usay dikhana, usay memory se describe karne se behtar hai. Dono products live web search bhi kar sakte hain, jab model ki apni knowledge kaafi na ho to search current information le aati hai.

Lambi Conversations Ko Handle Karna

Dono products current conversation ka record rakhte hain, jo baad ke answers ko shape karta hai. Is record ki bhi ek limit hoti hai, aur kisi lambe kaam mein aap us limit tak pahunch jaoge.

Ek example: aap ne poora din ek hi chat mein quarterly board report likhne mein spend kiya. 3 signs batate hain ke chat purani ho chuki hai:

• Subah di gayi koi rule (jaise "bullet points na likho") bhool jati hai aur bullets wapis aa jate hain
• Pehle diye gaye kisi figure se contradict hota hai
• Answers generic advice ki taraf drift ho jate hain, jo kisi bhi report pe apply ho sakti hai

Aisa hone pe na to force karte raho, aur na hi chat abandon karo. Pehle ek state summary mangwao, yani ek short record ke ab tak kya decide hua, kya still open hai, aur di gayi har rule:

Summarize what we have decided so far, what is still open, and every
rule I gave you. Keep it short enough to paste into a new chat.

Summary parho, mistakes fix karo, phir usay files ke sath ek new chat mein paste kar ke aage badho. Agar aap khud ko har hafte wahi summary paste karte huye paate ho, to yehi signal hai ke ab ek project ki zaroorat hai, jo Concept 4 hai.

Behtar context usually ek clever prompt se zyada answer behtar banati hai.

Part 2 · Workspace Ko Apna Banana

4. Projects, Ek Kaam Ke Liye Ek Room

AI ko kuch weeks use karne ke baad, sidebar messy ho jata hai. Ek hi client ya topic ke baare mein kai chats sidebar mein alag alag pari hoti hain, kaam ki files sirf ek chat mein hoti hain, aur background baar baar dobara explain karna parta hai.

Ek project 3 cheezein hold karta hai:

• Chats: usi kaam se related saari conversations
• Knowledge: usi project ke liye upload ki gayi files
• Instructions: project ke andar assistant kaise kaam kare, iski standing guidance

Ek example, Quarterly Board Reporting Project mein ye shamil ho sakta hai: pichle quarter ka board deck, current financial model, reporting ka style guide, tone aur audience ke baare mein instructions, aur board report se related har conversation. Ab aapko wahi files dobara upload karne ya wahi instructions dobara dene ki zaroorat nahi. Yehi poori baat hai: ek project ek stream of work ke liye context ko persistent bana deta hai. Book ki language mein, ek project ek chhota System of Context hai.

Claude Aur ChatGPT Ka Farq

Claude ka approach (retrieval): jab project knowledge context window se badi ho jati hai, to Claude automatically project knowledge ko search karna shuru kar deta hai, isse capacity roughly 10x badh jati hai. Ye switch automatically hota hai, aur ye ek paid-plan feature hai.

ChatGPT ka approach (memory): project memory add karta hai, jo 2 tarah se kaam kar sakti hai: default memory mein wider memory bhi participate kar sakti hai, jabke project-only memory mein project ek real boundary ban jata hai, aur outside memory andar nahi aati. Project-only memory hi actually isolation banati hai.

Yani ek project hamesha kaam ko organize karne mein help karta hai. Project-only memory hi wo cheez hai jo usay isolation mein badalti hai.

Project Kab Banao

“Projects us kaam ke liye hain jo continue rehta hai.”

5. Memory Aur Standing Instructions

Beginners in teenon cheezon ko constantly mix up kar dete hain.

Standing Instructions: Stable Rules

Standing instructions wo rules hain jo ek baar likhi jati hain aur broadly apply honi chahiyen, jaise “explanation se pehle answer do”, “jab tak main technical detail na maangoon, simple language use karo”, ya “main business leaders ke liye likhta hoon, software engineers ke liye nahi.” Claude mein ye settings aur styles mein rakhi jati hain, ChatGPT mein Custom Instructions mein (Settings aur Personalization ke andar). Ye un cheezon ke liye hain jo stable hain.

Memory: Aapke Baare Mein Evolving Context

Memory wo cheez hai jo assistant aapki conversations se pickup karta hai aur baad mein use kar sakta hai, jaise expressed preferences, recurring goals, aapke kaam karne ke tareeke se related facts, ya koi bhi cheez jo aap ne remember karne ko bola ho. Dono products aapko memory review, edit, aur delete karne dete hain.

Dono products ek aisa mode bhi dete hain jo conversation ko history aur memory se bahar rakhta hai, Claude mein incognito chat, ChatGPT mein temporary chat. Lekin dono companies safety aur abuse review ke liye roughly 30 din ke liye ek copy phir bhi retain karti hain. Isliye inhe memory-free samjho, trace-free nahi.

Projects: Ek Stream Of Work Ki Context

Ye dono se different hai. Projects kisi specific client, course, research topic, product, ya ongoing kaam ke gird context rakhte hain.

Stored Context Ek Responsibility Hai

2 concerns hain, convenience aur sensitivity. Stored context mein sensitive ya private information bhi ho sakti hai. Ye purani bhi hoti rehti hai, aur purani context silently fail hoti hai, koi error message nahi aata, sirf ek aisa answer aata hai jo confident lagta hai aur galat hota hai.

Example ke taur pe, agar pichle quarter ka financial model abhi bhi Quarterly Board Reporting project mein para hai, to is quarter ka board deck pichle quarter ke numbers ko neatly format kar ke quote kar dega. Assistant ye nahi bata sakta ke aap ne jo diya hai wo purana ho chuka hai.

Stored context isliye useful hai kyunki wo waqt ke sath compound hoti hai. Ye ek aisi jagah bhi hai jahan care matter karti hai. Ek achi tarah configured assistant weeks ke use ke baad zyada useful ho jati hai. Model nahi badla, uske gird ki context behtar ho gayi.

“Persistence tabhi help karti hai jab right information right jagah ho.”

6. Artifacts Aur Writing Blocks: Wo Kaam Jo Aap Le Ja Sakte Ho

Chat conversation ke liye achi hai, lekin finished work rakhne ke liye ek weak container hai. Agar aap 2,000 words ki report, ek working calculator, ek diagram, ya ek chhoti application maangte ho, to result ek separate object ki tarah chahiye hota hai, scroll mein bikhri ek message ki tarah nahi.

Claude Mein: Artifacts

Artifact types: document, code, web page, diagram, vector image, calculator, dashboard, aur ek chhoti interactive application. Aap Claude se baat karte raho jabke artifact wahin maujood rehta hai, jise build aur revise kiya ja raha ho. Claude ke paas Word, Excel, PowerPoint, aur PDF files banane ki separate abilities bhi hain.

ChatGPT Mein: Writing Blocks Aur Code Blocks

Purani tutorials “canvas” aur ek side-by-side editing panel ka zikr karti hain. 2026 mein OpenAI ne apne current models se canvas retire kar diya aur wahi kaam khud conversation ke andar le aaya, writing blocks aur code blocks ki shape mein, yani thread ke andar hi maujood, editable regions. Agar koi tutorial aapko canvas open karne ko kahe aur wo na mile, to yehi reason hai.

Design different hai. Idea same hai: work product ko discussion se separate rakhna.

Behtar Results Kaise Lo

“ek dashboard” ya “ek report” mat maango. Batao ke finished cheez ko actually kya karna chahiye.

Good:

Build a monthly budget tracker. I should be able to enter expenses
by category, see a pie chart, and get a warning when I go over budget.

Weak:

Build a budget tracker.

Ye bhi batao ke ye kis ke liye hai, kyunki new employees ke liye ek flowchart, experienced engineers ke liye flowchart jaisi cheez nahi hoti. Format bhi batao: agar answer data hai, to table ya spreadsheet file mango, paragraph nahi. Data se matlab hai jaise 12 vendors with prices and lead times, ek delivery schedule, ya koi bhi aisi cheez jise colleague Excel mein load karega. 12 vendors ka ek paragraph sort, filter, ya sheet mein paste nahi ho sakta. Ek table ho sakta hai. Agar aap maango, dono assistants spreadsheet file bana denge, jise download kiya ja sakta hai. Phir ek waqt mein ek change kar ke revise karo.

“Chat conversation hai, artifacts, writing blocks, aur code blocks hi asal kaam hold karte hain.”

7. Skills Aur Plugins: Kaam Karne Ke Packaged Tareeke

Projects is sawal ka answer dete hain ke knowledge ko sath kaise rakha jaye, lekin repetition ek aur sawal khara karti hai: assistant se har baar wahi method kaise dohrwaya jaye? Answer hai: Skills.

Isme instructions, examples, supporting resources, aur kabhi kabhi code shamil hota hai. Jab koi matching task aata hai, assistant khud us skill ko load kar leta hai.

Aap ek skill in cheezon ke liye bana sakte ho:

• Quarterly business reviews likhna
• Ek contract ko checklist ke against check karna
• Customer meeting ka brief prepare karna
• Raw notes ko preferred report format mein badalna

Har chat mein method dobara sikhane ki jagah, aap method ko ek hi baar pack kar dete ho. Claude ke paas skills hain, ChatGPT ke paas bhi skills hain. Products isme differ karte hain ke skills kahan available hain aur kaise manage hoti hain, lekin underlying idea ab dono mein shared hai.

Agent Skills Standard

Dono vendors ek open standard, Agent Skills (agentskills.io), pe kaam karte hain. Ek skill Markdown files ka ek folder hoti hai, jiske peeche na koi server hota hai na koi runtime, aur isi wajah se ye ek jagah se doosri jagah travel kar sakti hai. Kai companies ke dus-hazaar tools ab isi format ko read kar sakte hain, isliye ek product ke liye likhi hui skill doosre product mein bhi install ho sakti hai. Ye ek workflow ko ek prompt se zyada durable cheez bana deta hai, jo kisi ek vendor mein trapped na ho. Yehi is course ka opening lesson bhi hai: discipline hamesha constant hai, tool variable hai.

Plugin Kya Hai?

Isme kai skills, external apps se connections, aur kabhi kabhi commands ya sub-agents shamil hote hain. Ye ek installable package hota hai jo kisi job function ke liye banaya jata hai.

Custom GPTs Ke Baare Mein

ChatGPT ke paas custom GPTs bhi hain. Ek custom GPT ek separately configured assistant hai jise aap purpose se open karte ho, aur iski apni instructions, persona, knowledge, aur tools hote hain. Ye ek skill se different cheez hai.

“Jab koi method repeat ho, to us method ko package kar do.”

Part 3 · Reach Extend Karna

8. Connectors, Apps, Aur Sawal Ko Route Karna

Ab tak assistant zyada tar usi information pe kaam kar rahi thi jo aap ne uske saamne rakhi. Lekin aapki real information zyada tar kahin aur hoti hai: email, calendar, cloud storage, project-management tools, chat systems, aur company ke knowledge bases. Iska solution connectors aur apps hain.

Claude is ke liye connector word use karta hai, ChatGPT app (pehle connector bhi kehta tha). Dono names ek hi type ki cheez ke liye hain, sirf product ka naam different hai. Ek connector ya app assistant ko doosre system mein search, read, aur kabhi kabhi act karne deta hai, jitni permission aap dein.

• "Wo email dhoondo jahan hum ne vendor contract discuss kiya tha."
• "Kal mere kaunse meetings hain?"
• "Pichle week ke project notes ka summary do."
• "Mere sabse high-priority tasks kaunse hain?"

Ab assistant aapke real systems se answer de sakta hai, har cheez chat mein copy karne ki jagah.

MCP: Shared Integration Standard

Dono ecosystems ek standard, Model Context Protocol (MCP), ko support karte hain. Isay usually “AI tools ke liye USB-C” kaha jata hai, yani ek hi standard plug, jisse ek baar banaya gaya tool kai different assistants use kar sakte hain. Claude ke custom connectors MCP pe chalte hain, ChatGPT ke apps bhi Apps SDK ke through MCP pe bante hain.

Connections Permission Ke Decisions Hain

Kisi tool ko connect karna sirf productivity ka decision nahi, ye security aur governance ka decision hai, yani ye decision ke kis ko kya dekhne aur karne ki permission hai. Connect karne se pehle ye poocho:

• Ye kya read kar sakta hai?
• Kya ye outside world mein write kar sakta hai, send kar sakta hai, delete kar sakta hai, buy kar sakta hai, ya koi aur action le sakta hai?
• Kya mujhe is source pe trust hai?
• Kya mujhe is account ya system ko connect karne ki permission hai?

Connectors aur apps ko waise hi treat karo jaise software install karne ko treat karte ho: useful, aur permission-gated on purpose.

Ek connector ya app sirf wahi kar sakta hai jiske liye wo bana ho. Agar aapke kaam ko assistant se koi cheez send karwani hai, to pehle confirm karo ke sending us tool ke actions mein shamil hai. Permissions un actions mein se choose karti hain jo tool ke paas pehle se hain, wo new actions add nahi karti.

Search, Thinking, Ya Research?

Beginners aksar task ke liye wrong mode pick kar lete hain.

Pehla sawal ye poochna chahiye: main assistant se kis type ka kaam maang raha hoon? Ek fact pe jo search seconds mein de sakta hai, deep research waste mat karo. Jab asal problem reasoning ho to search ki taraf mat jao. Jab kaam ko multi-source investigation chahiye ho to sirf normal chat use mat karo.

Research Mode Ek Alag Cheez Hai

Research mode “better search” nahi hai. Assistant ek investigation plan karta hai, kai searches run karta hai, leads follow karta hai, sources parhta hai, aur citations ke sath ek structured report deta hai. Isme seconds nahi, minutes lagte hain. Aap ek investigation ka kaam de rahe hote ho, sirf ek quick question nahi.

“Sawal bhejne se pehle usay route karo.”

9. Us Kaam Pe Prove Karo Jo Aap Pehle Se Jaante Ho

Aapko kaise pata chalega ke assistant actually aapke kaam mein achhi hai? Right benchmark koi benchmark ya demo nahi, balke aapka apna task, aapka apna data, aapka apna standard hai.

Example: Program Director

Ek program director har quarter attendance aur employment outcomes ka analysis karta hai. Wo agle report ke liye AI se help chahta hai. Usay new data pe pehle se trust nahi karna chahiye. Iski jagah, wo assistant ko pichle quarter ka raw data deta hai, kyunki usay pehle se pata hai ke correct analysis kaisa dikhta hai. Phir wo assistant se wo kaam dobara karwata hai. Ab uske paas compare karne ke liye ek real cheez hai. Agar assistant koi zaroori pattern miss kare, to wo instructions behtar bana sakta hai. Agar data mein hi koi problem nikle jo analysis ke liye zaroori hai, to usay ek data problem mil gayi. Agar assistant kaam ka koi hissa reliably nahi kar pata, to wo hissa human ke paas rehta hai. Teenon outcomes useful hain.

5-Step Proving Loop

1. 1

   Ek Repeating Task Pick Karo

   Exactly clear raho.

2. 2

   Ek Old Example Dhoondo

   Aisa jiska correct result aapko pehle se pata ho.

3. 3

   Assistant Se Dobara Banwao

   Usay wahi context do jo aapke paas normally hoti.

4. 4

   Compare Karo

   Kya match hua? Kya fail hua? Kaunsi instruction missing thi?

5. 5

   Improve Karo Aur Dobara Run Karo

   Jab tak kaafi achha na ho, ya ye decide ho jaye ke ye task delegate nahi hona chahiye.

Pass Hona Actually Kya Prove Karta Hai?

Ye aapko similar future kaam ke liye tested confidence deta hai. Ye prove nahi karta ke assistant har agle case mein bhi correct hoga. Aur ye responsibility transfer nahi karta.

Ab bhi in cheezon ki zaroorat rehti hai:

• New results check karo ke wo make sense karte hain
• Final kaam ke zimmedar khud raho
• Jahan appropriate ho, AI ki help disclose karo
• High-risk decisions ko proper human oversight mein rakho

Ye chhoti si exercise aapki pehli evaluation suite hai.

“Trust known work ke against earn karna chahiye, kisi good-looking answer se assume nahi kar lena chahiye.”

Chhota Recap, Prompts Se Pehle

Ab aapke paas dono products ka map hai. Claude aur ChatGPT alag tools hain, lekin kaafi hisson mein same soch pe bane hain.

• Models aur thinking modes, kitni capability spend karni hai, ye control karte hain
• Attachments, is waqt zaroori context provide karte hain
• Projects, ek stream of work ke liye context sath rakhte hain
• Standing instructions, stable rules rakhti hain
• Memory, aapke baare mein evolving context rakhti hai
• Artifacts, writing blocks, aur code blocks, finished work ko chat scroll se bahar rakhte hain
• Skills, repeating methods pack karti hain
• Plugins, poore type ke kaam ke liye capabilities bundle karte hain
• Claude mein connectors aur ChatGPT mein apps, external systems tak reach dete hain
• MCP, dono ke neeche shared integration standard hai
• Search, thinking, aur research, 3 alag questions ke 3 alag routes hain
• Jani-pehchani kaam pe test karna, trust earn karne ka tareeka hai

Products badalte rahenge. Ye mental model current interface se kahin zyada der tak chalega.

Ab Khud Try Karo: 6 Prompts

Cockpit ke baare mein parhna, usme baith kar fly karne jaisa nahi hai. In 6 exercises mein roughly 25 minute lagte hain.

1. 1

   1. Ek Guided Tour Maango

   Run karne se pehle, 2 features ka guess laga lo jo assistant ke mention karne ki expectation hai. Phir Claude ya ChatGPT se poocho ke wo aaj ke workspace ka tour de, models badalne, file upload, projects, memory, aur research ke baare mein. Dhyan do: apni pehle wali guess se compare karo, phir jo screen pe dikh raha hai usse. Jahan assistant apne hi interface ke baare mein wrong ho, wahi sabse interesting hissa hai.

2. 2

   2. Ek Task Dono Assistants Mein Run Karo

   Apne week ka koi chhota, real task pick karo. Run karne se pehle guess lagao ke kaunsa assistant behtar karega, aur kyun. Phir dono se wahi task karwao aur poocho ke unhe behtar karne ke liye kaunsi context chahiye thi. Dhyan do: outputs compare karo, phir dono ke jawab compare karo ke unhe kya chahiye tha.

3. 3

   3. Apna Pehla Project Setup Karo

   Kaam ka koi stream pick karo jahan aap pehle se wahi background ya files repeat kar chuke ho. Assistant se poocho ke wo aapse short interview kare, phir 2 cheezein de: project instructions, aur upload karne layak documents ki priority-ordered list. Dhyan do: jo sawal wo poochta hai, wahi context hai jo aap har week hath se type kar rahe the.

4. 4

   4. Jo Remember Kiya Gaya Hai Uska Audit Karo

   Assistant se poocho ke wo aapke baare mein jo bhi remember karta hai, ek plain list ki tarah dikhaye. Har item ke liye khud decide karo: keep karo, correct karo, ya delete karo. Phir poocho ke settings mein memory kahan manage ki ja sakti hai. Dhyan do: kya kisi cheez se aapko embarrass hota, aur kya kuch purana ho chuka hai.

5. 5

   5. 3 Real Sawal Route Karo

   Apne kaam se 3 sawal pick karo: ek jise current fact chahiye, ek jise hard reasoning chahiye, aur ek jo proper report deserve kare. Har sawal poochne se pehle route choose karo, search, thinking, ya research. Dhyan do: wrong routing, jaise research mode jise search hona chahiye tha, dono products mein sabse common daily waste hai.

6. 6

   6. Assistant Ko Apni Old History Pe Test Karo

   Assistant ko batao ke aap ek repeating task ko test karna chahte ho, jiska correct result aapko pehle se pata hai. Old input material do, aur assistant se wahi kaam dobara karwao jaise aap khud karte. Apna original answer tab tak hold karo jab tak kaam finish na ho. Baad mein 3 cheezein likh lo: kya match hua, kaunsi extra detail chahiye thi, aur kya human ke paas rehna chahiye.

Is Chapter Ke Naye Terms

Exam ke liye ye poori glossary yaad rakho, koi bhi term skip mat karo:

Self-Test

Khud Se Poocho

Pehle khud answer do, phir sawal pe click kar ke answer check karo. Peeche dekhe bina answer dene ki koshish karo.