vokh (@v0khid) header
vokh (@v0khid) profile picture

creator Ai & Fintech ┃degen since 2017 ┃advisor & writer

Collabs · DMs open
Joined October 2013
Photo by vokh (@v0khid)Photo by vokh (@v0khid)Photo by vokh (@v0khid)Photo by vokh (@v0khid)Photo by vokh (@v0khid)Photo by vokh (@v0khid)Photo by vokh (@v0khid)Photo by vokh (@v0khid)Photo by vokh (@v0khid)Photo by vokh (@v0khid)
Former p*rn actress Christy Mack recounts a brutal attack by her ex-boyfriend, an MMA fighter known as War Machine. In August 2014, Jonathan Koppenhaver broke into her Las Vegas home and brutally attacked her and another man. Christy, who had previously worked in the p*rn industry, suffered multiple facial fractures, lost several teeth, and barely survived. She later revealed that she genuinely believed he was going to kill her that night. During her court testimony, Christy broke down in tears while recalling the terrifying details. Her former boyfriend, who once fought in the UFC and Bellator, was sentenced in 2017 to life in prison with the possibility of parole after 36 years. Do you think someone capable of such brutality deserves even one chance to walk free again?
34
79
1
829
346,421
vokh retweeted
stop prompting. seriously. 5 prompts. 5 copy-pastes. 5 checks. for ONE task. you’re not using AI. you’re its assistant. give it the whole job instead: >one contract >one schedule >one review at 07:00 no babysitting. just results. how to set it up in 7 days 👇
20
4
36
1,298
stop prompting. seriously. 5 prompts. 5 copy-pastes. 5 checks. for ONE task. you’re not using AI. you’re its assistant. give it the whole job instead: >one contract >one schedule >one review at 07:00 no babysitting. just results. how to set it up in 7 days 👇
20
4
36
1,298
vokh retweeted
Everyone is turning Grok Bot into a team of 6 bots. That's the fastest way to BURN YOUR WEEKLY USAGE LIMIT 🔥 Multi-agent setups use ~15× more tokens than a regular chat. These rules save your tokens so you never hit the limit: → Connectors before the browser → Stay on task, ask only once → 3 failures = stop and report → Long results: save to files, not the chat → Big data is read by code, not the model → Routines stay silent if nothing has changed Paste them once, save, and drop them into Grok Bot. Done. I also recommend my article on the habits that save hundreds of thousands of tokens 👇
23
5
33
1,003
Everyone is turning Grok Bot into a team of 6 bots. That's the fastest way to BURN YOUR WEEKLY USAGE LIMIT 🔥 Multi-agent setups use ~15× more tokens than a regular chat. These rules save your tokens so you never hit the limit: → Connectors before the browser → Stay on task, ask only once → 3 failures = stop and report → Long results: save to files, not the chat → Big data is read by code, not the model → Routines stay silent if nothing has changed Paste them once, save, and drop them into Grok Bot. Done. I also recommend my article on the habits that save hundreds of thousands of tokens 👇
23
5
33
1,003
vokh retweeted
I took 5 permissions away from my AI, and it got better: ✅ Read 32 sources ✅ Cut 11 duplicates ✅ Rank the stories ✅ Verify every claim ✅ Draft 3 briefs ✅ Run 5 quality checks ✅ Write the morning report ⬜ Publish ⬜ Send emails ⬜ Pay for anything ⬜ Delete files ⬜ Guess when unsure the empty boxes are why I trust it !
17
3
1
33
764
I took 5 permissions away from my AI, and it got better: ✅ Read 32 sources ✅ Cut 11 duplicates ✅ Rank the stories ✅ Verify every claim ✅ Draft 3 briefs ✅ Run 5 quality checks ✅ Write the morning report ⬜ Publish ⬜ Send emails ⬜ Pay for anything ⬜ Delete files ⬜ Guess when unsure the empty boxes are why I trust it !
17
3
1
33
764
vokh retweeted
my AI agent blocked a $240 refund at 3am. it was right to 6 jobs ran overnight. I made 1 decision: ai news brief -> PASS inbox reply -> ASK competitor price change -> PASS $240 refund -> BLOCK x post with 1 unsourced claim -> HOLD weekly report -> PASS no prompts. just one contract: > lcode filters > lmodels draft > lI decide anything that sends, publishes or pays stops and waits for me whole run: ~$0.41 (illustrative) the best agent isn't the one that does the most it's the one that knows when to stop
22
4
1
34
757
vokh retweeted
3,000 items in. 1 decision out. my agent reads everything while I sleep. I read one card with my coffee. the trick isn't a smarter model. it's knowing which job deserves one: duplicates → plain code · $0 relevance → small model facts → strong model · only 40× writing → general model publishing → me $28.50 a run → $0.38 (illustrative) stop paying for intelligence you don't need. the best agent isn't the smartest one. it's the one that knows when to stop and ask.
16
2
30
723
Holy sh*t. AI agent tip: stop asking it for answers, give it a stopline collect → dedupe → rank → verify → build → audit → stopline Write the rules once, let it run all night, and wake up to 1 card instead of 40 prompts In the demo, one night run hit 3 moments where a normal chatbot would guess: → 11 duplicate stories: plain code cut them, $0, zero tokens → claim #4 was only a repost: no primary source, so it got parked, not guessed → "publish brief_1.md": permission is ask_first, so it paused and pinged me 32 items read. 4 ranked. 3 briefs written. 1 claim on hold. 0 published without me. Code filters. Models draft. You decide. - the full night run setup > Cheap steps go to code (dupes, dates, formats) > Small model only for "is this relevant?" > Strong model only for claims, and only against primary sources > Every brief passes 5 checks before you see it > Anything that publishes, sends or pays stops and asks Paste this into your agent: "You run my research overnight. Follow this contract: 1. Mission: [WHAT YOU WANT EVERY MORNING] 2. Sources: only [APPROVED SOURCES] > Cut duplicates with code before using any model 3. For every claim: > Find the primary source > No primary source → park it and tell me why 4. Before you hand me anything, check: > Objective intact, claims sourced, no contradictions, inside scope, next step safe 5. Boundaries: > Read, rank, draft: do it alone > Publish, send, pay: stop and ask me If unsure, stop and ask. Never guess."
8
2
1
24
638
my AI agent blocked a $240 refund at 3am. it was right to 6 jobs ran overnight. I made 1 decision: ai news brief -> PASS inbox reply -> ASK competitor price change -> PASS $240 refund -> BLOCK x post with 1 unsourced claim -> HOLD weekly report -> PASS no prompts. just one contract: > lcode filters > lmodels draft > lI decide anything that sends, publishes or pays stops and waits for me whole run: ~$0.41 (illustrative) the best agent isn't the one that does the most it's the one that knows when to stop
22
4
1
34
757
3,000 items in. 1 decision out. my agent reads everything while I sleep. I read one card with my coffee. the trick isn't a smarter model. it's knowing which job deserves one: duplicates → plain code · $0 relevance → small model facts → strong model · only 40× writing → general model publishing → me $28.50 a run → $0.38 (illustrative) stop paying for intelligence you don't need. the best agent isn't the smartest one. it's the one that knows when to stop and ask.
16
2
30
723
stop prompting AI every 5 minutes. give it the whole job most people still use AI like a vending machine: prompt → wait → review → prompt again 5 requests. 5 handoffs. 1 human managing every step the model isn't the bottleneck anymore. you are here's the system I'd build instead: 1. the contract: write it once, not every morning. mission, sources, process, boundary, done-when. → lesson: "you are a world-class researcher" changes the tone. a contract changes the behavior. 2. the agent: one agent, four jobs. collector finds evidence. builder drafts. auditor checks. operator delivers or pauses. → lesson: don't build 4 agents before you need 1. 3. the record: every story becomes a structured record before it becomes prose. source, what happened, verified, uncertain, decision. → lesson: good automation keeps the reasoning trail, not just the answer. 4. the checkpoint: errors compound. one misread announcement turns into a confident wrong article 3 steps later. → lesson: an AI checking its own work is not independent verification. 5. the stopline: allow / ask first / block. if uncertain, stop and ask. → lesson: autonomy without permissions is just risk. 6. the router: you don't need a top model to spot a duplicate. code → small model → general model → strong model → human. → lesson: use the cheapest method that can safely finish the task. the shape: 07:00 trigger → 1 agent → 4 checks → 1 folder waiting for you at 08:10. and the metric that matters (example week): system A: 120 tasks, 28 fixes, 3 hours saved system B: 75 tasks, 4 fixes, 8.5 hours saved the busiest agent loses. the edge is not a smarter prompt. the edge is a workflow you trust enough to sleep through. bookmark this before your next "can you also…" prompt. which step in your workflow still waits on you?
11
17
606
🚨URGENT: Claude can now act like your personal tutor. It can take you from complete beginner to actually understanding a skill in just 4 hours. I put together 10 prompts that guide Claude through the entire learning process. Use them in order, starting with #1
14
2
28
793
9. Practice a real-life situation with me «Pretend you are [a client, interviewer, colleague, or another person]. I want to practice [skill] in this situation: [scenario]. Keep your responses short and realistic. During the role-play, don’t tell me what I should say. After 5 exchanges, stop, give me specific feedback, and show me how I could respond better in the most difficult moment».
1
2
38
10. Help me reinforce what I’ve learned «I’ve been studying [skill] for [amount of time]. Now I can [list]. Help me come up with a small project that I can complete in [amount of time] where I can apply what I’ve learned. Create a list of steps, suggest a simple way to check the result, and make a 7-day practice schedule. Start by asking me to complete the first step». Want to dive deeper into AI? Follow me for more.
3
31