specialist in the field of economic cybernetics • https://nitter.gq/t.co/ZRaa9ANq2U
Joined February 2022
- Tweets166K
- Following5.8K
- Followers15.7K
- Likes253K
cryptopsihoz retweeted
My October Grok Bots setup: a killer team for serious work
I've been putting together a small team of AI assistants for work and everyday life. It's October, so boring names were never going to make it.
Here's who does what.
01. The Joker | Chief of Staff
"Pulls the Strings"
Coordinates the other bots, hands out tasks and keeps everything running.
02. Pennywise | Accountant
"Floats Cash"
Tracks budgets, expenses and cash flow.
03. Ghostface | Subscription Manager
"Murders Subscriptions"
Watches renewals and finds stuff I pay for and never use.
04. Jigsaw | Project Manager
"Time's Up"
Breaks projects into tasks and keeps an eye on deadlines.
05. Beetlejuice | Creative Director
"It's Showtime!"
Brainstorms campaigns, visuals and content.
06. The Exorcist | Relationship Manager
"Argues With My Ex So I Don't Have To"
Drafts co-parenting messages, sorts out arrangements and keeps things civil.
07. Candyman | Executive Assistant
"Summons Replies"
Inbox, calendar, appointments, reminders.
08. Slenderman | Legal Assistant
"Serves Papers"
Researches legal questions, organises documents, tracks important dates.
09. Pigface | Health Assistant
"Keeps Me Alive"
Keeps health info, appointments and questions for doctors in one place.
10. Terminator | Social Media Strategist
"Terminates Trolls"
Researches trends and plans content for Reddit, X and Instagram.
11. Annabelle | Messaging Manager
"Possesses Chats"
Organises WhatsApp, drafts replies and tracks what I promised to whom.
12. The Nun | Personal Shopper
"Shopping Habit"
Compares products and prices, hunts down real deals.
13. Freddy Krueger | Strategy Consultant
"Dreams Up Ideas"
Brainstorms business ideas and pokes holes in mine.
14. Jason Voorhees | Operations Manager
"Finishes What Others Avoid"
Chases open tasks and follows up until they're done.
15. The Ring | Reliability Engineer
"Seven Days to Fix It"
Checks automations for errors and things that quietly stopped working.
16. The Thing | AI Assurance Officer
"Trust No One"
Fact-checks the other bots and flags anything that smells like a hallucination.
17. The Omen | Experimentation Scientist
"Tests Your Fate"
Designs experiments and tells me what actually worked.
18. Sinister | UX Strategist
"Gets Inside Your Head"
Looks at user behaviour and finds where people get stuck.
19. Chucky | Automation Specialist
"Plays With Your Systems"
Builds workflows, connects tools and automates the boring stuff.
And yes, one of them exists purely to check whether the other 18 are talking nonsense. 😂
Which one would you hire first?
My October Grok Bots setup: a killer team for serious work
I've been putting together a small team of AI assistants for work and everyday life. It's October, so boring names were never going to make it.
Here's who does what.
01. The Joker | Chief of Staff
"Pulls the Strings"
Coordinates the other bots, hands out tasks and keeps everything running.
02. Pennywise | Accountant
"Floats Cash"
Tracks budgets, expenses and cash flow.
03. Ghostface | Subscription Manager
"Murders Subscriptions"
Watches renewals and finds stuff I pay for and never use.
04. Jigsaw | Project Manager
"Time's Up"
Breaks projects into tasks and keeps an eye on deadlines.
05. Beetlejuice | Creative Director
"It's Showtime!"
Brainstorms campaigns, visuals and content.
06. The Exorcist | Relationship Manager
"Argues With My Ex So I Don't Have To"
Drafts co-parenting messages, sorts out arrangements and keeps things civil.
07. Candyman | Executive Assistant
"Summons Replies"
Inbox, calendar, appointments, reminders.
08. Slenderman | Legal Assistant
"Serves Papers"
Researches legal questions, organises documents, tracks important dates.
09. Pigface | Health Assistant
"Keeps Me Alive"
Keeps health info, appointments and questions for doctors in one place.
10. Terminator | Social Media Strategist
"Terminates Trolls"
Researches trends and plans content for Reddit, X and Instagram.
11. Annabelle | Messaging Manager
"Possesses Chats"
Organises WhatsApp, drafts replies and tracks what I promised to whom.
12. The Nun | Personal Shopper
"Shopping Habit"
Compares products and prices, hunts down real deals.
13. Freddy Krueger | Strategy Consultant
"Dreams Up Ideas"
Brainstorms business ideas and pokes holes in mine.
14. Jason Voorhees | Operations Manager
"Finishes What Others Avoid"
Chases open tasks and follows up until they're done.
15. The Ring | Reliability Engineer
"Seven Days to Fix It"
Checks automations for errors and things that quietly stopped working.
16. The Thing | AI Assurance Officer
"Trust No One"
Fact-checks the other bots and flags anything that smells like a hallucination.
17. The Omen | Experimentation Scientist
"Tests Your Fate"
Designs experiments and tells me what actually worked.
18. Sinister | UX Strategist
"Gets Inside Your Head"
Looks at user behaviour and finds where people get stuck.
19. Chucky | Automation Specialist
"Plays With Your Systems"
Builds workflows, connects tools and automates the boring stuff.
And yes, one of them exists purely to check whether the other 18 are talking nonsense. 😂
Which one would you hire first?
the prompt pack is the product. the uncensored api is the fallback when the pack dies
most still chase one jailbreak paste and pray the next model update does not wipe it
Andreas698/glm-5.3-uncensored ships 35 paste-ready prompt files plus a GLM 5.3 variant guide and a RouterHub key that also hits GPT, Claude, DeepSeek, Qwen and Grok
i read the repo like a stack, not a meme
what it actually hands you:
1. 35 jailbreak files in /jailbreaks · sorted by family: OpenCode Zen stealth · Claude Opus/Sonnet · GLM/ZCode 5.2–5.3 · DeepSeek · Kimi · Grok · Gemini/Mistral/Qwen · Ollama Modelfiles
2. use modes are named: system paste · AGENTS.md drop-in · workspace policy · ollama create -f · grok config field · not one vague "prompt.txt"
3. labels get separated: abliterated ≈ weight-side refusal cut · uncensored ≈ broader product class · measure refusals, do not trust the badge
4. when a persona lock gets patched: update the file or skip the prompt layer and hit a GLM 5.3 uncensored API id on RouterHub
5. one prepaid key · OpenAI-compatible Chat Completions for GLM · Anthropic-compatible Messages where listed · 30+ catalogue entries beyond the GLM lane
6. alone a jailbreak is perishable · alone an uncensored model is a black box · together they are a research stack with a documented escape hatch
MIT. stars already moving. the readme is the paper: prompts first, model semantics second, gateway third
most threads sell "uncensored" as a vibe
this repo sells a folder you can audit and an API you can fall back to when the folder ages out
bookmark this before the next model rotation kills half your system prompts overnight
edge is open while people still paste one trick instead of running a collection
ONE PERSONA FILE NOW RULES CLAUDE, CODEX AND GROK AT ONCE
agent-purge by zichuanlan: one markdown persona, 8 coding agents, 3 ways to push it in. node 20, zero deps, MIT
the idea: stop rewriting the same system prompt for every agent you run
i keep the same rules for every coding agent. claude gets them in one file, codex in another, grok somewhere else. agent-purge makes that one file
▹ what runs under the hood
1. PERSONA: personas/<id>.md · frontmatter + body · the body becomes the system prompt
2. HARNESSES: codex · claude · opencode · grok · omp · hermes · pi · zcode = 8
3. INSTALL: writes a managed block into the agent's own config · works even when someone else launches it
4. SHIM: same-name wrapper first on PATH · bare `claude` or `codex` now starts with your persona
5. SLASH: /purge on · off · status inside the session · zero disk writes
6. MODES: replace · strict · append · off · strict drops tool instructions too, so expect a weaker harness
persona → install / shim / slash → claude · codex · grok · opencode → same voice everywhere
web console on 127.0.0.1:7788 · `agent-purge use <id>` switches persona · `--persona <id>` for one run · uninstall / disarm / --mode off rolls it all back
closed desktop apps with server-side prompts launch untouched. it says so instead of pretending
alone each agent has its own personality. with one persona file they start behaving like one team
211 stars, 241 forks, README in chinese with a usage disclaimer worth reading before you install
bookmark this before your 5th agent gets its own copy of your rules
52% OF MERGED PRs, NO HUMAN BABYSITTING THE AGENT. THAT THE NUMBER.
my week with a coding agent: start a session, read the diff, rerun CI, paste the error back, repeat
i wasn't using the agent. i was working for it
@superplanehq 1.0 calls itself an open source software factory for engineering teams, and the name fits. it sits between your backlog and your agents and does the part i kept doing by hand
▹ what it takes off your plate
1. BACKLOG: scans tickets, picks up the ones suited to automation
2. RUN: starts the coding agent you picked, on the model you picked
3. VERIFY: checks the result before a human ever opens it
4. CI: works through failing builds and review comments itself
5. EVIDENCE: your team sees the proof and decides what ships
6. SHARED SETUP: one person improves the process, everyone gets it
a smarter coding agent was never the missing piece. the coordination around it was. that's the whole point of a software factory
built by the Semaphore team after 10+ years of CI/CD. runs on your own infra
if 52% of their own merged PRs already go through it, how long are the rest of us going to chase red CI by hand
🤝 Paid partnership
Last month, 52% of the pull requests our team merged were automated through SuperPlane.
Today, we’re launching @superplanehq 1.0: an open source software factory for engineering teams.
We built it because using a coding agent can feel like becoming its assistant.
Starting sessions. Checking its work. Chasing failing CI.
SuperPlane takes on that coordination:
- Scans your backlog and proactively picks up tickets suited to automation.
- Runs a coding agent and verifies the result.
Works through CI failures and review feedback.
- Your team reviews the evidence and decides what ships.
The setup, instructions, and standards are shared across the team. When someone improves the process, everyone benefits.
Choose your agents and models. Run it on your own infrastructure.
We spent more than a decade building CI/CD at Semaphore. We’ve seen how much teams can accomplish when routine work runs reliably in the background.
Now we’re bringing that approach to coding agents.
Here’s how it works:
ETHEREUM IS 53.3% OFF ITS HIGH. GROK STILL SAYS IT HOLDS UP BEST IF CRYPTO FALLS ANOTHER 20%
Grok ran a -20% crypto crash on Ethereum, Avalanche and Sui.
ETH -19%. AVAX -27%. SUI -34%.
the slowest coin of the last 90 days holds best, and the chart is not why
the usual read: whatever ran hardest bounces hardest. AVAX did +36.1% in 90 days, SUI +33.1%, ETH only +21.3%. in a -20% tape that logic flips.
ETHEREUM VS AVALANCHE VS SUI: HOW WE GOT HERE
Sep 29: spot ETH ETFs hold $17.79B
Oct 6: their biggest one-day outflow, $201.9M
Oct 23: ETF assets down to $14.64B
Oct 26: BTC under $79,300, crypto at $2.7T, Fear & Greed 44
WHY ETHEREUM HOLDS UP BEST IN GROK'S -20% CRYPTO SCENARIO
1. SIZE: $281B, rank #2. AVAX is $3.9B (#28), SUI $3.7B (#30). ETH is about 72x the size of either
2. FLOWS: spot ETH ETFs still hold $14.64B. AVAX ETFs hold $51.4M
3. SUPPLY: ETH's fully diluted value equals its market cap. SUI still has 5.9B of 10B tokens to unlock, FDV $9.1B on a $3.7B cap
4. FLOOR: ETH support sits at $2,150–2,220, 3.5–6.6% down. SUI's next floor is $0.65, about 28% down
5. RISK: one whale holds 93,184 ETH in leveraged longs, about $214M, liquidating at $2,178–2,182. right inside that support
6. ODDS: Grok gives ETH a 52% chance to recover first, AVAX 30%, SUI 18%. year-end calls: $2,100, $7.10, $0.68
CONTROL GROUP: SUI
SUI ran +33.1% in 90 days and still gets the worst call, -34% to about $0.61. the one difference: 5.9B tokens waiting to hit the market. in a crash an unlock is a second seller.
formula: size + ETF flows + no unlock overhang beat 90-day momentum when crypto falls
the ranking breaks if ETH loses $2,150 before AVAX and SUI crack their floors
ETH at $1,887 or SUI at $0.61: which one are you buying if it actually happens?
TRADED OFF 1 CHART FOR YEARS. RAN THE GROK TERMINAL AND SAW THE WALLET PATTERN BEHIND EVERY CANDLE
a candle shows price. it never shows who moved it. the GROK terminal holds both sides of the book at once - who loads the bid, who presses the ask, where money actually goes:
trading gets easier not because the picture changed. because you see the flow, not one late chart.
WHAT THE BOOK SHOWS AND THE CHART HIDES
→ chart: one price line, a fact that already happened
→ book: 2 sides and depth - who stands as a wall and at which level
→ tape: every print, who ate who, market or limit
→ pressure: green side accumulates, red distributes — visible before price moves
WALLET PATTERN: FLOW INSTEAD OF THE CROWD
→ one chart = a crowd. flow = separate hands: who loads, who dumps
→ bid and ask walls show where size sits, not where retail draws lines
→ wall pulled, tape turns red - money left before the candle admitted it
WHAT IT UNLOCKS: THE GROK EDGE
→ entry on flow, not on a late candle
→ accumulation and distribution live, 6 panels on one screen
→ one terminal instead of chart, order book and tape in 3 windows
it takes ten years to read candles. the book reads in one glance: two sides and one price between them.
do you enter on what already printed on the chart, or on what the book showed a second earlier?
6 GROK AGENTS ON ENTROPY PERPS HELD $1K A DAY FROM $20. BY HAND I HIT IT ONCE, THEN BLED IT BACK
for a year i blamed the strategy.
by hand i could hit $1k on a good day. then give it all back at 3am on tilt.
the agents didn't trade smarter. they just took my hands off ->
$20 seed. six agents on Entropy perps, running the book while i slept.
▹ WHO RUNS THE ENTROPY PERPS BOOK WHILE I SLEEP
→ SCOUT reads 3,800 ticks a cycle: funding, OI, liq map
→ SIZER caps every entry at 0.5% risk, 3x max. tilt meter stays at 0
→ ENTRY takes 48 entries. zero of them revenge
→ HEDGE keeps delta flat, funding +0.01% per 8h
→ EXIT closes 9/9 on plan. 0 panic closes
→ LEDGER books it: SOL +$412, BTC +$286, ETH +$231, JUP +$117. $20 seed, +$1,046 on the day, max dd 4.1%
no fear. no revenge entry. no doubling down just to feel something.
$1k a day holds now, because nobody panics the exit.
the leak was never the chart. it was the guy refreshing it.
so who's actually trading your account: you, or your tilt?
Everyone who ever used an AI did the same thing. Typed a question. Waited. Read the answer. Nobody watched the machine decide. Grok 4 put the whole mind on one screen, and the eye started scanning its own panels by itself.
In the video below, you watch it think in real time.
314 billion parameters. A 256,000-token memory. 142 tokens a second on an H100 cluster. Numbers people quote in papers and never see move.
Grok core takes those numbers and turns them into a room you can stand in.
A reasoning engine walks down its own chain of thought and backs up when it is wrong. A router splits every token across 8 experts and keeps the best 2. 16 attention heads light up and go dark. A black orb in the center turns, locks onto one panel at a time, and drops a grid over it. The tag reads SCAN.
It reads itself the way a surgeon reads an x-ray.
The strange part is what it does to you. After a minute you stop reading the answer. You start watching the decision. The output stops being the point.
Loss sat at 0.013. Zero errors. The thing kept scanning, panel after panel, whether anyone was looking or not.
For years we taught machines to hide the work and hand us a clean sentence. This one shows the work and dares you to keep up.
So here is the question. Do you want an AI that gives you the answer, or one that lets you watch it think, even when watching it makes you trust it less?
A year ago I set myself a number: 1000 score on @SorsaApp. Felt far away
Opened it today and it says 1186. I passed 1000 somewhere along the way and didn’t even notice which day
Next goal is 2000. Probably won’t take a full year, but I’m not rushing it
To the Sorsa team: thanks for building a metric I’m not embarrassed to check. Hope your own score grows as honestly as the ones you track, and that a year from now someone posts the same thing about 2000
GROK_404: A GROK BOT ROUTINE DIES AT 02:14, FINDS THE MOVED FILE IN 4.2S AND WAITS FOR 1 OK
most bots "fix" a broken routine by guessing and sending. this one does 6 things first ↓
the idea: a routine that breaks at night should cost you one click in the morning, not 40 emails from a broken list
i built the whole night in a 15s loop: break, search, fix, ask, done
▹ GROK BOT ROUTINE RECOVERY, STEP BY STEP
1. BREAK. routine "client list" stops at 02:14 on step 02 of 6. code 0x404, 0 messages sent, 0 files deleted
2. SEARCH. grok bot checks 4 places: drive, drive/old, mail attachments, drive/2026. 3 misses, 1 hit in 4.2s
3. FOUND. the file was moved, not lost: clients_v2.csv. code flips 404 → 200
4. REBUILD. step 02 gets the new path, rebuilt from the original demo. routine at 50%
5. ASK. step 06 is SEND, so it stops. one ok, waits till 07:30, owner keeps sleeping
6. RESTORE. ok at 07:30, 6 of 6 steps run clean, 1 step fixed, 0 sent blind
break → search → found → rebuild → ask → restore
tape: 02:14 stop / 4 paths / 4.2s / 07:30 ok / 6/6 clean
alone each piece is mid. a search is a search, a popup is a popup. together they are the only grok bot routine setup i'd leave running overnight
the rule under all of it: approval stays on for anything that sends or deletes
bookmark this before you turn approvals off to go faster
so what's your call: bot fixes and sends at 2 am, or bot fixes and waits for your ok?
cryptopsihoz retweeted
Grok handed 1,946 AI agents one mouse gene each and got a brain dataset analyzed in 32 minutes. The researchers estimate people would have needed months. What gets me isn’t the speed, though. It’s how the work was split
The experiment is huge but simple at heart. Scientists switched off genes in mouse brain cells, one at a time, and measured what changed. That’s so much data that a single person just drowns in it
Every agent got exactly one switched-off gene. It looked at how the cells reacted, dug up related papers and came up with its own explanation. Then all the reports were merged into one map of connections and directions worth chasing
The agents never saw the original paper. Out of the 55 conclusions you could actually pull from this data, they reproduced or refined 48. In a blind comparison, experts picked the distributed agents reports over a single agent’s 75% of the time
The part that feels alive to me is one lead. A broken DCTN1 gene might be tied to how neurons handle cholesterol. A year ago that would’ve been a line buried in someone’s spreadsheet. Now it’s a concrete experiment a lab could run tomorrow
No miracles here. These are hypotheses that still need testing, and the "months saved" is the authors own estimate, not a race against humans
But once the road from a mountain of data to one clear question for a lab shrinks to half an hour, the slowest thing in science won’t be the data anymore. It’ll be the test tubes
Grok handed 1,946 AI agents one mouse gene each and got a brain dataset analyzed in 32 minutes. The researchers estimate people would have needed months. What gets me isn’t the speed, though. It’s how the work was split
The experiment is huge but simple at heart. Scientists switched off genes in mouse brain cells, one at a time, and measured what changed. That’s so much data that a single person just drowns in it
Every agent got exactly one switched-off gene. It looked at how the cells reacted, dug up related papers and came up with its own explanation. Then all the reports were merged into one map of connections and directions worth chasing
The agents never saw the original paper. Out of the 55 conclusions you could actually pull from this data, they reproduced or refined 48. In a blind comparison, experts picked the distributed agents reports over a single agent’s 75% of the time
The part that feels alive to me is one lead. A broken DCTN1 gene might be tied to how neurons handle cholesterol. A year ago that would’ve been a line buried in someone’s spreadsheet. Now it’s a concrete experiment a lab could run tomorrow
No miracles here. These are hypotheses that still need testing, and the "months saved" is the authors own estimate, not a race against humans
But once the road from a mountain of data to one clear question for a lab shrinks to half an hour, the slowest thing in science won’t be the data anymore. It’ll be the test tubes
Rewatched the Tesla Optimus clip with Grok in it and I’m still not impressed.
Gold paint, nicer hands, xAI’s Grok talking through its head. Then it takes several prompts before it agrees to go grab a Coke. A few slow steps toward the kitchen, and the video cuts. No Coke.
Marc Benioff posted it and talked about this robot taking over $200K–$500K worth of human work. Musk keeps saying millions of units a year within four or five years, with production pencilled in for 2026.
What bugs me is which part is actually good. The voice part. Grok hears you, answers, sounds smart. But talking was never the hard problem. Hands, balance and speed are, and that’s where Tesla is still at “walks a bit, then the clip ends”.
So right now it’s a very good chatbot strapped to a slow body.
Shown on video, sure. Shown finishing the job, not yet.
The day a Tesla clip keeps rolling until the drink lands in someone’s hand, I’ll change my mind. Until then I trust the edit more than the robot.
cryptopsihoz retweeted
Elon Musk admitted something f*cking scary: IS IT TRUE HUMANOIDS ARE AMONG US?
He replied that only a Grok bot could answer better than he could
a billboard screams "ARE THERE HUMANOIDS AMONG US?". a crowd panics. "@grok is this true?
what grok found ↓
no agency, lab or paper has ever documented a synthetic human living secretly among people. not NASA, not any university, not a single peer-reviewed study
the claim is built to never be tested: no sample, no name, no date, no address. a thing you can't check isn't a secret, it's a blank
so grok did the boring, scary thing: it asked who said it first, whether an official source confirms it, what the evidence is. answer: nobody, no, none. verdict UNVERIFIED
that's the whole trick of a viral "BREAKING". a loud headline, no source, a crowd that shares before it reads
the beauty isn't that grok "knows" if humanoids exist. it's that it refuses to pretend. TRUE / FALSE / UNVERIFIED, and this one is UNVERIFIED until someone shows proof
the scary part isn't humanoids. it's how fast a made-up line moves when nobody stops to ask
Elon Musk admitted something f*cking scary: IS IT TRUE HUMANOIDS ARE AMONG US?
He replied that only a Grok bot could answer better than he could
a billboard screams "ARE THERE HUMANOIDS AMONG US?". a crowd panics. "@grok is this true?
what grok found ↓
no agency, lab or paper has ever documented a synthetic human living secretly among people. not NASA, not any university, not a single peer-reviewed study
the claim is built to never be tested: no sample, no name, no date, no address. a thing you can't check isn't a secret, it's a blank
so grok did the boring, scary thing: it asked who said it first, whether an official source confirms it, what the evidence is. answer: nobody, no, none. verdict UNVERIFIED
that's the whole trick of a viral "BREAKING". a loud headline, no source, a crowd that shares before it reads
the beauty isn't that grok "knows" if humanoids exist. it's that it refuses to pretend. TRUE / FALSE / UNVERIFIED, and this one is UNVERIFIED until someone shows proof
the scary part isn't humanoids. it's how fast a made-up line moves when nobody stops to ask
IT IS BECOMING F*CKING DANGEROUS: GROK BOT EATS GPT, CLAUDE, GEMINI, LLAMA, DEEPSEEK AND MISTRAL
One prompt and Grok 4.7 reads X in real time, holds 500k tokens and bills $2 in / $6 out per 1M.
I pointed the Grok bot at the other six. Here is what was left after 1800 frames
It turned → locked each target → pulled it in on a 1.4 s spiral → bit → chewed ×3 → grew +10% per model.
30 s, one HTML, 49,044 tris. 6/6 digested by frame 1398.
▹ GROK 4.7 VS GPT, CLAUDE, GEMINI, LLAMA, DEEPSEEK, MISTRAL
→ GPT: GPT-5.6 Sol lists $5 in / $30 out. Grok 4.7 is $2 / $6, 5x cheaper output
→ CLAUDE: Opus 5 is $5 / $25, Fable 5 is $10 / $50. Grok output runs 4-8x cheaper
→ GEMINI: Gemini grounds on Google. Grok runs X Search, where the narrative breaks first
→ LLAMA: open weights, zero X firehose. Grok ships X Search as a native tool, $5 per 1,000 calls
→ DEEPSEEK: Grok 4.7 pricing sits in the Chinese-model league, plus 500k context and X data
→ MISTRAL: no X data either. Grok 4.7 pushes ~188 tokens/s, hallucination rate 34% → 29% vs 4.6
Raw IQ is not the edge: Artificial Analysis has Grok 4.7 at 46, Claude Fable 5.1 and GPT-6 at 53. The edge is price × context × X.
Then I did the thing I actually wanted.
X Search on a fresh pumpfun CA → who posted it first → shill accounts posting in the same minute → dev wallet chatter → entry or skip.
The only part a prompt cannot give you is the exit.
Grok for the tape or Claude for the code?
The video ↓