More from @aiwithmayank
Dec 29, 2025
Anthropic engineers don’t prompt like everyone else.
I reverse-engineered their internal techniques from leaked docs and demos.
The difference is insane.
Here are the 10 insider methods they don’t want you to know:
1. The "Recursive Logic" Loop
Most prompts ask for an answer. This forces the model to doubt itself 6 times before committing.
Template: "Draft an initial solution for [TOPIC]. Then, create a hidden scratchpad to intensely self-critique your logic. Repeat this ‘think-revise’ cycle 5 times. Only provide the final, bullet-proof version."
2. The "Context Architect" Frame
Stop stuffing your AI with info. Use "Just-in-Time" retrieval to stop "context rot."
Template: "I am going to provide [DATA]. Do not p…
More from @aiwithmayank
Dec 29, 2025
Anthropic engineers don’t prompt like everyone else.
I reverse-engineered their internal techniques from leaked docs and demos.
The difference is insane.
Here are the 10 insider methods they don’t want you to know:
1. The "Recursive Logic" Loop
Most prompts ask for an answer. This forces the model to doubt itself 6 times before committing.
Template: "Draft an initial solution for [TOPIC]. Then, create a hidden scratchpad to intensely self-critique your logic. Repeat this ‘think-revise’ cycle 5 times. Only provide the final, bullet-proof version."
2. The "Context Architect" Frame
Stop stuffing your AI with info. Use "Just-in-Time" retrieval to stop "context rot."
Template: "I am going to provide [DATA]. Do not process everything. Use a ‘minimal high-signal’ approach to extract only the facts necessary to solve [PROBLEM]. Discard all redundant noise."
Read 13 tweets
Dec 26, 2025
Stop using Perplexity or ChatGPT for market research.
I tested Gemini 3 and it’s on a whole different level for data analysis.
Here are 5 prompts that turn it into your research team:
(Comment "Gem" and I’ll DM you my 500 mega prompts list)
1/ THE MARKET MAP PROMPT
Everyone starts with “what’s the market size lol” but winners map the entire battlefield first.
Prompt to steal:
“Give me a complete market map for [industry]. Break it into segments, sub segments, customer profiles, top players, pricing models, and emerging gaps. Highlight where new entrants have the highest odds of success.”
This gives you clarity fast.
2/ THE COMPETITOR AUTOPSY PROMPT
Stop guessing what your competitors are doing.
Gemini can literally dissect them.
Prompt to steal:
“Analyze the top 5 competitors in [space]. Break down their features, pricing, positioning, value props, moat, weaknesses, customer complaints, and hidden advantages. Summarize as if you’re preparing a strategy memo for a CEO.”
You’ll find angles they missed.
Read 8 tweets
Dec 20, 2025
OPENAI, ANTHROPIC, AND GOOGLE DON’T PROMPT LIKE YOU
They use internal techniques that turn LLMs into precision machines Accuracy jumps. Hallucinations drop.
Here are 10 of those techniques (Bookmark this for later):
Technique 1: Role-Based Constraint Prompting
The expert don’t just ask AI to "write code." They assign expert roles with specific constraints.
Template:
You are a [specific role] with [X years] experience in [domain]. Your task: [specific task] Constraints: [list 3-5 specific limitations] Output format: [exact format needed]
Example:
You are a senior Python engineer with 10 years in data pipeline optimization. Your task: Build a real-time ETL pipeline for 10M records/hour Constraints:
- Must use Apache Kafka
- Maximum 2GB memory footprint
- Sub-100ms latency
- Zero data loss tolerance Output format: Production-ready code with inline documentation
This gets you 10x more specific outputs than "write me an ETL pipeline."
Watch the OpenAI demo of GPT-5 and see how they were prompting ChatGPT... you will get the idea.
Technique 2: Chain-of-Verification (CoVe)
Google’s research team uses this to eliminate hallucinations.
The model generates an answer, then generates verification questions, answers them, and refines the original response.
Template:
Task: [your question]
Step 1: Provide your initial answer Step 2: Generate 5 verification questions that would expose errors in your answer Step 3: Answer each verification question Step 4: Provide your final, corrected answer based on verification
Example:
Task: Explain how transformers handle long-context windows
Step 1: Provide your initial answer Step 2: Generate 5 verification questions that would expose errors in your answer Step 3: Answer each verification question Step 4: Provide your final, corrected answer based on verification
Accuracy jumps from 60% to 92% on complex technical queries.
Read 14 tweets
Dec 18, 2025
Everyone quotes Sun Tzu.
Nobody actually uses his decision framework.
I turned his 2,500-year-old strategy system into copy-paste AI prompts that tell you exactly when to fight, when to retreat, and when to win without fighting at all.
Real strategy looks nothing like hustle culture 👇
1. The Terrain Analysis Prompt
Before making any move, Sun Tzu mapped the terrain.
Most people jump into decisions blind. This prompt forces you to see the entire battlefield first.
Copy this:
"You are a strategic advisor trained in Sun Tzu’s principles.
I’m facing this situation: [describe your challenge]
Analyze the terrain using these dimensions:
- Strengths I control that others don’t
- Weaknesses that could be exploited
- External forces I can’t control
- Hidden opportunities most people miss
- The real competition (not the obvious one)
Give me the strategic map before I make any moves."
2. The Asymmetric Advantage Finder
"All warfare is based on deception." - Sun Tzu
Translation: compete where you’re strong, they’re weak.
This prompt finds your unfair advantages others can’t copy.
Copy this:
"I’m competing in [your market/field] against [competitors].
My unique resources/skills: [list them] Their obvious strengths: [list them]
Using Sun Tzu’s principle "attack where undefended":
- Where are they NOT competing?
- What advantage do I have that they can’t quickly replicate?
- What market position looks weak but is actually uncontested?
Show me the asymmetric battlefield."
Read 11 tweets
Dec 16, 2025
CHAIN OF THOUGHT IS OFFICIALLY DEAD ☠️
I’ve been testing Atom of Thought prompting and it’s casually boosting reasoning accuracy by 30-40% on hard problems.
This is how people will use ChatGPT and Claude going forward.
Here’s how to use it:
The problem with Chain of Thought: it forces linear thinking.
Real problem-solving doesn’t work that way. Your brain doesn’t solve physics problems by thinking step 1 → step 2 → step 3.
You break complex problems into atomic components, then recombine them.
Atom of Thought (AoT) works differently.
Instead of sequential steps, you decompose problems into their smallest logical units. Each "atom" is an independent piece of reasoning that can be validated separately.
Think LEGO blocks instead of a chain.
Read 14 tweets
Dec 15, 2025
CHATGPT JUST TURNED PROJECT MANAGEMENT INTO A ONE PERSON SUPERPOWER
You are wasting time on Status updates, task breakdowns, timelines, scope creep, follow ups.
ChatGPT can run the entire thing for you like a project manager if you use these 6 prompts.
Here’s how:
1/ ASSIGN IT THE ROLE (THIS MATTERS)
PMs don’t just answer questions.
They own outcomes.
Prompt to steal:
“Act as a senior project manager. Your goal is to deliver [project] on time and within scope. Ask me any clarifying questions before proceeding.”
Instant ownership.
2/ TURN IDEAS INTO A REAL PLAN
Vibes don’t ship products.
Plans do.
Prompt to steal:
“Turn this idea into a project plan. Break it into phases, tasks, dependencies, risks, and milestones. Estimate timelines and flag unknowns.”
This replaces whiteboards.
Read 10 tweets