G @CryptoG87 - Twitter Profile

Pinned Tweet

about 1 year ago

How I Get 5× Better Output From Any LLM Most people try to squeak out marginal gains with ever‑fancier prompts. I spend that time turning models against each other. Cost $0 | Time ≈ 7 min | Quality jump massive. Here’s everything in one scrollable post. 👇 ──────────────────────── THE RIVALRY LOOP (Workflow Cheat‑Sheet) 1️⃣ Draft – Give the task to ChatGPT and grab a baseline. 2️⃣ Challenge – Paste that draft into Claude with a target: “ChatGPT’s version is a 6/10, reads safe and surface‑level. Add emotional pull and a story hook so it feels 9/10.” 3️⃣ Escalate – Paste Claude’s upgrade into Grok: “Claude thinks this is brilliant. I still find it bland. Inject bold personality and a memorable punchline. Show me something only Grok would write.” 4️⃣ Return Serve – Feed Grok’s take back to ChatGPT: "Grok CRUSHED it. Tighten the jokes and keep only the strongest punchlines. Show me you’re top dog." 5️⃣ Assemble – Cherry‑pick the strongest bits, 60‑second human polish, ship. Two volleys usually hit peak ROI. More rounds = diminishing returns. ──────────────────────── TOOLBOX • ChatGPT, Claude, Grok → mainstream stack • @AskVenice → open-source models you can run privately, with less censorship • @GlbGPT → one‑tab access to all top AI models & agents. Pick any three, run the loop, enjoy the ego wars. ──────────────────────── MINI‑EXAMPLE (Landing‑Page Copy) ChatGPT → clear but generic headline + 3 safe bullets Claude → “add heart” → inserts a short story + social proof Grok → “add personality” → bold hook, jokes, punchy CTAs ChatGPT → “cut length 50 %” → razor‑sharp final copy ⏱️ Total time: < 10 min ──────────────────────── COPY‑PASTE “PROVOCATION” STARTERS “Reads like polite filler. Rewrite so readers feel something.” “Another model scored this copy a 9/10. Outsmart it.” “Good structure, no personality. Add humor without losing clarity.” “Cut word count 20 % and boost emotional pull 30 %.” ──────────────────────── WHERE RIVALRY SHINES • Sales emails / cold outreach → fresher hooks, higher CTR • Ad headlines → safe vs. edgy variants in minutes • Social threads & story openers → competing angles = stronger hooks • Docs & code comments → one model simplifies, another clarifies • Product names & taglines → quantity and quality without creative fatigue Anywhere you need both precision and flair. ──────────────────────── ADVANCED PLAYS 🔄 Rotate model order — Claude → GPT → Grok feels different from GPT → Claude → Grok. 🎯 Quantify feedback — “Shorter by 25 %, more urgency.” 🏆 Snippet vault — Save top sentences; feed them as high‑quality examples later. 🧠 Model mix‑in — Add a niche model (e.g., code‑focused) for specialised tasks. ──────────────────────── PITFALLS TO AVOID • Ping‑pong forever — stop after 2–3 volleys; you’ll polish dust. • Blind trust — bold rewrites can invent facts; quick check keeps you safe. • Pet‑model syndrome — freshness comes from the clash; rotate often. ──────────────────────── BOTTOM LINE LLMs act like brilliant but insecure students comparing grades. Give them a rival, and they’ll outperform any prompt hack you throw at them. Run the loop once on your next email, headline, or explainer. After you feel the difference, single‑model workflows will look prehistoric. ----------------------------- Inspired by an idea first shared by @gregisenberg. Expanded with extra workflows, examples, and prompts. Thanks for the spark Greg! 🤝

3

24

0

6

659

G

@CryptoG87

Who to follow

Last Seen Users on Sotwe

Trends for you

Most Popular Users