How to Prompt

LLMs are not magic. They are math, at scale. Here’s how they really work: 1. They predict, not think. → An LLM is a giant function. → It takes text in and predicts the next word. 2. How chatbots use it. → You start with a prompt. → The model adds your message. → It predicts what a helpful AI would say. → That randomness makes it sound human. 3. Why answers change. → The process is deterministic. → Randomness means no outputs are the same. 4. Training is the hard part. → The model reads everything. → Each pass tweaks billions of tiny dials. 5. Large means really large. → No one sets them. → Hundreds of billions of parameters. → The model learns them by itself. In the end, an LLM is a prediction machine. One word at a time. ✦ PS: video edited by opus.pro ✦ Start mastering AI at how-to-prompt.ai

Being rude to AI makes it work better. A new study just proved it: Researchers tested ChatGPT-4o with 250 prompts. 1. Read the full breakdown here: https://lnkd.in/d3z9eJA5 2. Master prompting here: how-to-prompt.ai The team covered math, science, and history. Every question was tested 10 times, across all tones. Aggressive prompts got the best results: • Very Rude: 84.8% accuracy • Very Polite: 80.8% accuracy Earlier models like GPT-3.5 did not show this pattern. Only the newest models, like ChatGPT-4o, responded this way. The shift is not about being mean for fun. It’s about how the AI processes urgency and directness. Polite prompts add extra words. This can distract the AI, making it less accurate.

BREAKING: Andrew NG just dropped a new course on Agentic AI. Here's what you'll learn: 1) Reflection: Teach your agent to review its own work and improve with every step. 2) Tool Use: Let your agent decide which tools to use (search the web, send emails). 3) Planning: Break down big tasks into smaller steps, and let your agent handle each one. 4) Multi-Agent Collaboration: Build teams of agents, each with a specialty, working together like a real company. Access it here: https://lnkd.in/dyHD9PRr Bonus: start mastering AI here: how-to-prompt.ai

Stop learning how to prompt. Do this instead. ☑ Copy & paste this specific prompt: (You must turn on "Extended Thinking"!!) Prompt: "I want to learn the best way to [goal/decision you need] for [product, project, or situation]. Before conducting your deep research, ask me exactly 10 short clarifying questions about my aims, limits, audience, timeline, and any key facts you need. Wait for my answers.” ☑ Now answer ChatGPT's questions to prompt it. Because is what most of your prompts lack: → Direction. → Specificity in research. → As much context as possible. Once the research is done, continue normally. ☑ Here are good follow-up prompts: → Prompt to get specific info: “Give me the most current facts on [sub-topic] inside [topic]. List at least 3 sources, link each. Show facts as short bullet points. After the list, add one line: “Main takeaway: …” → Prompt to get a counterargument: “Show the best reasons against the main claim in [topic]. Give 5 solid points, one sentence each. Under each point, add the evidence or expert that backs it (with link). End with: “Why these points matter: …” (max 3 lines)” → Prompt to get an actionable step-by-step plan: “Change these key ideas on [topic] into a clear plan for a beginner. Use a table with columns: “Week”, “Action”, “Goal-post”. Cover Week 1–12. Keep each action ≤ 20 words. After the table, add: “How to track progress: …” → Prompt to find ressources/courses about the topic: “List good ways to learn more about [topic]. Give 5–7 items across books, papers, free online courses. For each, add: link, cost (free / paid), and one-line value note. Sort by beginner → advanced.” Read the full breakdown here: https://lnkd.in/ex8fXGQW

10 prompt techniques that make a difference. This is worth saving for later: 1. Examples beat instructions 2. Version control your prompts like code 3. Test coverage matters more than prompt quality 4. Domain expertise > prompt tricks 5. Temperature tuning is underrated 6. Model-specific optimization required 7. Chain-of-thought isn’t always better 8. Use AI to write prompts for AI 9. System prompts are your foundation 10. Prompt injection defense from day one ✦ Bonus: master prompt basics at: https://lnkd.in/dUQEec5v

I’ve found this FREE Sora 2 prompt library. See how to (exactly) prompt it: Check out the Notion here → https://lnkd.in/dkciN9zt ✦ Bonus: master prompt basics at: https://lnkd.in/dUQEec5v

You think ChatGPT is the best AI to search. But it's not. Here are 6 others to use instead: 1. Perplexity, made for search. 2. Grok, from xAI (which has access to Twitter). 3. Deepseek, the Chinese sensation that just got an upgrade. 4. Claude, that also got an upgrade this week with Sonnet 4.5. 5. Gemini, the ChatGPT of Google, known for its image editing. 6. Kimi, another AI tool from China, known for being the least biased. Read the full breakdown here: https://lnkd.in/dbD_qGV8

OpenAI released a Sora 2 prompt guide. Here's how to (exactly) prompt the new viral AI: "[Prose scene description in plain language. Describe characters, costumes, scenery, weather, and other details. Be as descriptive to generate a video that matches your vision.] Cinematography: Camera shot: [framing and angle, e.g., wide establishing shot, eye level] Mood: [overall tone, e.g. cinematic and tense, playful and suspenseful, luxurious anticipation] Actions: - [Action 1: a clear, specific beat or gesture] - [Action 2: another distinct beat within the clip] - [Action 3: another action or dialogue line] Dialogue: [If the shot has dialogue, add short natural lines here or as part of the actions list. Keep them brief so they match the clip length.]" Help others prompting better by ♻️ reposting this. Start mastering AI at → how-to-prompt.ai

These AI prompt tricks work so well it feels like cheating. They make AI smarter, faster, and more human: 1. Start with “Let’s think about this differently” ↳ It stops giving you generic answers. ↳ You get creative, out-of-the-box ideas. 2. Use “What am I not seeing here?” ↳ This finds blind spots and hidden assumptions. ↳ You get a wider, smarter view of your problem. 3. Say “Break this down for me” ↳ Even for simple things, this works. ↳ AI explains the steps, the science, and the why. 4. Ask “What would you do in my shoes?” ↳ AI stops being neutral and gives real opinions. ↳ You get advice that’s specific and actionable. 5. Use “Here’s what I’m really asking” ↳ Add this after your main question. ↳ You get answers that fit your true needs. 6. End with “What else should I know?” ↳ This is the secret weapon. ↳ You get a complete answer, not just a quick fix. To learn more about how to prompt chatgpt, go to → https://lnkd.in/d7xTY3AX

Stop using ChatGPT to seek information. This is a much better combo: 1. Use Perplexity Pro. 2. Make sure to use "Research". 3. Copy and paste this prompt below: “I need [specific fact/criteria] about [subject] as of [date/today/timeframe]. Give the answer, key caveats, and what to double-check.” For example, I love using this for viral topics: “I need the top 3 viral AI startup stories as of today. Give the answer, key caveats (such as disputed details or unconfirmed reports), and what to double-check before posting.” Perplexity will search real-time news, link direct sources, and flag possible issues (like if a stat is outdated or a report comes from a single source). --- Ruben Hassid wrote an entire blog on how to (best) search with AI. Read the full breakdown here: https://lnkd.in/dbD_qGV8