founder • entrepreneur • living & making money in AI ecosystem and sharing it with you ⚡️ AI safety external advisor at OpenАІ🧠 DMs open 📩
California, USA
Joined August 2020
- Tweets6.6K
- Following161
- Followers9.9K
- Likes18.6K
tekario retweeted
I GAVE MY GROK BOT $500 AND ONE FORMULA FROM SECOND-YEAR PROBABILITY: THE KELLY CRITERION.
20 HOURS LATER: $2,325
4.6x and the formula decided every position size.
most people size trades by feel. same amount every time, or more after a win, or double after a loss to "get it back". that's how small accounts die.
Kelly is one line:
f* = p − (1 − p) / b
p = the chance the trade wins. b = how big the win is next to the loss. f* = how much of the balance to put in.
so before every entry, Grok estimates p and b, and the formula picks the size:
- strong setup: p = 0.55, b = 3 → 0.55 − 0.45 / 3 = 40% of the balance
- weak setup: p = 0.38, b = 2 → 0.38 − 0.62 / 2 = 7%
- f* at zero or below → no trade. that happened 61 times.
it traded SOL, AVAX, SUI, ADA and LINK, both ways, with 8x to 25x leverage set by how far away the stop sits.
the five big closes:
- 02:14, SOL short at 20x, 184.20 → 175.10, 97% on margin, Kelly 30%
- 05:40, AVAX short at 18x, 21.84 → 20.62, 99%, Kelly 36%
- 09:05, SUI long at 16x, 3.412 → 3.641, 106%, Kelly 40%
- 12:48, ADA short at 22x, 0.6815 → 0.6512, 96%, Kelly 31%
- 17:30, LINK short at 19x, 15.36 → 14.48, 107%, Kelly 40%
take LINK. 15.36 to 14.48 is a 5.7% drop. at 19x that's 109% gross, 107% after fees. Kelly put 40% of the $1,607 balance in, so it paid $689.
37 losing trades over the session. Kelly sized them at 5–10%, so none of them hurt.
the honest part: the five big closes made $1,804 of the $1,825 profit. the other 52 trades together netted +$20. and Kelly is only as good as Grok's guess of p. overestimate it and the formula oversizes, which is why most pros run half-Kelly.
40% of the balance on one trade at 19x is what this paper desk runs, not a plan for anyone else's account.
hour 20. $2,325 on the balance, and the next trade gets exactly the size the formula allows.
GROK BOT + OPUS 5.5 = A $5,000/MONTH SERVICE YOU CAN RUN IN 5 HOURS A WEEK.
HERE'S THE MATH.
the offer is a weekly "what your competitors did" report for local businesses.
- grok watches each client's competitors: new offers, price changes, what customers say about them online.
- opus 5.5 turns that into a clear one-page report with three moves the client should make this week.
the numbers:
- 30 minutes per client per week to check and send
- 10 clients = 5 hours a week, about 20 hours a month
- $500 per client per month = $5,000 a month
- $5,000 ÷ 20 hours = $250 an hour
to be honest about it: the AI part is the easy part. finding 10 clients is the real work. But you're no longer selling your time. You're selling a system.
I post the business models people actually ask me about. If someone you know wants a side business that isn't dropshipping, show them this one.
what would you charge for a report like this in your city?
tekario retweeted
GROK BOT + OPUS 5.5 CAN BE YOUR FULL-TIME ANALYST FOR UNDER $1 A TASK.
NO SALARY. NO SICK DAYS. IT CHECKS ITS OWN WORK.
the difference between AI that guesses and AI you can trust with real work:
bad: One prompt, one answer, hope
good: Plan, act, check, fix, repeat
give it 12 checks for what "correct" means. It loops until all 12 pass, then stops on its own.
3 rules:
1. A clear checklist
2. Permission to fix its own mistakes
3. A stop rule: all pass, a lap limit or a budget
I break down one AI system like this every week. stay tuned and pass it on to a business owner who needs it.
would you trust an AI analyst with your numbers?
GROK BOT + OPUS 5.5 CAN BE YOUR FULL-TIME ANALYST FOR UNDER $1 A TASK.
NO SALARY. NO SICK DAYS. IT CHECKS ITS OWN WORK.
the difference between AI that guesses and AI you can trust with real work:
bad: One prompt, one answer, hope
good: Plan, act, check, fix, repeat
give it 12 checks for what "correct" means. It loops until all 12 pass, then stops on its own.
3 rules:
1. A clear checklist
2. Permission to fix its own mistakes
3. A stop rule: all pass, a lap limit or a budget
I break down one AI system like this every week. stay tuned and pass it on to a business owner who needs it.
would you trust an AI analyst with your numbers?
tekario retweeted
GROK + OPUS 5.5 = 18,000 PARTICLES SHOWING WHY YOUR AI BILL IS 87% TOO HIGH
whether you run grok, opus 5.5 or any other big model, you're probably paying full price for the same prompt again and again.
this board shows where the money goes:
fire = tokens the model reads fresh, at full price
ice = tokens reused from cache, at about 10% of the price
gold = what you keep
same 18,600-token request, example pricing:
→ no cache: $0.0558
→ cached: $0.0072
three rules decide which side you're on:
1. same stuff first: instructions, tools and docs at the top
2. new stuff last: only the new message at the bottom
3. don't edit the start: one changed line at the top makes everything after it full price again
and caches go cold. leave one idle for a few minutes and you pay to write it again.
i break down AI cost tricks like this in plain english. if someone on your team pays the API bill, show them this.
did you know your AI could reuse most of your prompt for a tenth of the price?
GROK + OPUS 5.5 = 18,000 PARTICLES SHOWING WHY YOUR AI BILL IS 87% TOO HIGH
whether you run grok, opus 5.5 or any other big model, you're probably paying full price for the same prompt again and again.
this board shows where the money goes:
fire = tokens the model reads fresh, at full price
ice = tokens reused from cache, at about 10% of the price
gold = what you keep
same 18,600-token request, example pricing:
→ no cache: $0.0558
→ cached: $0.0072
three rules decide which side you're on:
1. same stuff first: instructions, tools and docs at the top
2. new stuff last: only the new message at the bottom
3. don't edit the start: one changed line at the top makes everything after it full price again
and caches go cold. leave one idle for a few minutes and you pay to write it again.
i break down AI cost tricks like this in plain english. if someone on your team pays the API bill, show them this.
did you know your AI could reuse most of your prompt for a tenth of the price?
tekario retweeted
OPUS 5.5 + GROK BOT = 48,000 S&P 500 HEADLINES CHECKED. 6,688 WERE FAKE
NONE GOT THROUGH.
how the swarm works:
→ 300 agents take in every market headline: wire stories, posts, screenshots, forum rumours.
→ opus 5.5 reads each one and pulls out the claim: which company, what happened, what number.
→ the Grok bot checks that claim against the company's own sources, like the press release, exchange notice or earnings call. Each check takes about 100 ms.
→ if nothing official backs it up, it's flagged as fake, with the reason: no source, company denies, wrong number, old story or parody account.
→ real news is sorted into the 11 S&P 500 sectors: tech, banks, health, energy and the rest.
when a rumour storm hit, fakes jumped 3×, and every headline was still checked before it counted.
a fake "CEO steps down" post can move a stock in minutes. Here it doesn't get the chance.
if you know someone who trades on headlines, send this their way.
have you ever traded on a headline that turned out to be fake?
tekario retweeted
I GAVE GROK BOT $200 AND BANNED EVERY COIN EXCEPT BITCOIN. IT TURNED IT INTO $1,400 IN 7 DAYS
everyone said my last runs were just DOGE luck. So I took DOGE away.
no memecoins. No lottery trade. No "one lucky pump."
One coin. The most watched chart on earth. The one everyone says is too efficient to beat.
6 trades. Longs and shorts. Same coin every time.
I test AI trading repos and I post the parts that make them look bad. This week I tried to take away their best excuse.
the repo: HyperGrok. One Grok brain, 7 agents, Hyperliquid perps. Free and open source.
setup, 10 minutes:
1 Clone the repo
2 Add your Grok API key
3 Wallet: $200
4 Allowed coins: BTC. That's it.
5 Paper mode on
the week, all BTC:
Long +$90
Short +$160
Long −$40
Short +$230
Long +$310
Long +$450
$200 → $1,400. Zero DOGE.
the honest part: the last long made 38% of it. Without that one, the wallet ends at $950. Smaller than the DOGE weeks, but no single trade carried it.
if you've got a friend who still says "it was just luck," send them this one.
what should I ban next?
OPUS 5.5 + GROK BOT = 48,000 S&P 500 HEADLINES CHECKED. 6,688 WERE FAKE
NONE GOT THROUGH.
how the swarm works:
→ 300 agents take in every market headline: wire stories, posts, screenshots, forum rumours.
→ opus 5.5 reads each one and pulls out the claim: which company, what happened, what number.
→ the Grok bot checks that claim against the company's own sources, like the press release, exchange notice or earnings call. Each check takes about 100 ms.
→ if nothing official backs it up, it's flagged as fake, with the reason: no source, company denies, wrong number, old story or parody account.
→ real news is sorted into the 11 S&P 500 sectors: tech, banks, health, energy and the rest.
when a rumour storm hit, fakes jumped 3×, and every headline was still checked before it counted.
a fake "CEO steps down" post can move a stock in minutes. Here it doesn't get the chance.
if you know someone who trades on headlines, send this their way.
have you ever traded on a headline that turned out to be fake?
OPUS 5.5 + GROK BOT = 300 AGENTS HUNTING HIDDEN CONNECTIONS IN THE S&P 500
500 annual reports and every link checked against the page it came from
Apple, Nvidia, JPMorgan, Exxon. Each one publishes a 100+ page annual report every year. Who supplies them, who sits on their boards, who owns what: it's all in there, and almost nobody reads it.
so here's the swarm I'd point at it:
→ 300 agents split the S&P 500 across its 11 sectors: tech, healthcare, banks, energy, industrials and the rest
→ Opus 5.5 reads each company's annual report and pulls out suppliers, major customers, board members and major shareholders
→ It then proposes links between companies: supplier, shared board member, investor, subsidiary
→ A cheap Grok bot checks every link against the exact page it came from in about 100 ms
→ Wrong links get cut. Only links with a page number behind them stay on the map.
the real prize is the links between sectors. A chipmaker and a bank that share a board member. An energy giant that quietly supplies half of industrials.
those ties don't show up on any stock screener, and they're exactly what you'd want to know before you invest.
it's weeks of analyst work, done by a swarm that checks itself.
I'm building more boards like this. If you're into agent setups, you'll see the next one here first. And show this to the friend who picks stocks from headlines.
which S&P 500 company would you want to see the hidden ties of?
tekario retweeted
OPUS 5.5 + GROK BOT = A REVERSE ENGINEERING MACHINE THAT FACT-CHECKS ITSELF. 300 AGENTS. 51,320 CLAIMS. 0 ERRORS LEFT
how it works (the board shows the architecture):
→ 300 agents run in parallel, each taking one company apart from its filings
→ Opus 5.5 does the heavy lifting: it pulls out revenue, HQ, CEO and founding year
→ A cheap Grok bot checks every claim against the source page, in about 100 ms each
→ If a claim fails, it's thrown out and redone. No "looks right." Only "the source says so."
→ 3 passes over 100 profiles: 212 errors, then 37, then 0
the lesson: don't let one model grade its own homework.
expensive model for the work. Dirt-cheap model for checking. Loop until clean.
every dot on the ribbon is one claim. the white sparks are the ones that got caught.
I'm breaking down more agent setups like this one, so stick around if that's your thing. And show it to the friend who still trusts single-shot AI answers.
would you trust a cheap model to check an expensive one?
OPUS 5.5 + GROK BOT = A REVERSE ENGINEERING MACHINE THAT FACT-CHECKS ITSELF. 300 AGENTS. 51,320 CLAIMS. 0 ERRORS LEFT
how it works (the board shows the architecture):
→ 300 agents run in parallel, each taking one company apart from its filings
→ Opus 5.5 does the heavy lifting: it pulls out revenue, HQ, CEO and founding year
→ A cheap Grok bot checks every claim against the source page, in about 100 ms each
→ If a claim fails, it's thrown out and redone. No "looks right." Only "the source says so."
→ 3 passes over 100 profiles: 212 errors, then 37, then 0
the lesson: don't let one model grade its own homework.
expensive model for the work. Dirt-cheap model for checking. Loop until clean.
every dot on the ribbon is one claim. the white sparks are the ones that got caught.
I'm breaking down more agent setups like this one, so stick around if that's your thing. And show it to the friend who still trusts single-shot AI answers.
would you trust a cheap model to check an expensive one?
tekario retweeted
Claude Code found a hacker’s device on my home Wi-Fi…
I have 9 devices. Phone, laptop, TV, a couple of smart plugs. I asked Claude Code to list everything connected to my router. It found 11
One was the old smart plug I forgot about in the hallway. Fine
The other one had no name. Its hardware ID didn't match any manufacturer, which usually means one thing: a laptop hiding its real address
It was hacker’s laptop. Connected to my Wi-Fi every evening for 5 months.
Changed the Wi-Fi password. Everything dropped off. Only my devices came back
Someone on your network can see what else is on it, use your IP for whatever they want, and poke at every camera and smart device you own
The full step-by-step is below, takes 2 minutes. Save it so you don't lose it
tekario retweeted
after I read through Anthropic official guide on Reverse Engineering i gave my local LLM one app and one instruction: “replicate this"
then i logged everything.
- 00:00 drop in the binary. no source, no docs, nothing
- 00:06 50+ functions renamed. it reads like real code now
- 00:18 it finds the encryption routine buried deep inside
- 00:35 full login flow mapped, step by step
- 01:00 spec md done. enough to rebuild the app from zero
one hour. one laptop. zero cloud.
3 rules that made it actually work (keep this):
→ make the model quote the exact address for every claim
→ check anything important by actually running the code
→ keep one notes file it rereads every step
that's it. those 3 rules turn a guessing chatbot into a real reversing partner.
a year ago this was a week of work for a skilled human. now it's a coffee break.
i'm pushing this further every week: bigger apps, harder targets. stay tuned and watch where it breaks.
show this to the friend who still opens ida and suffers alone.
question for you: what's the first app you'd point it at?
tekario retweeted
I TOLD GROK BOT ONE RULE AND IT TURNED $100 INTO $1,000 IN 6 DAYS
the rule: "Stop at $1,000."
the tool: one free, open-source GitHub repo. Grok figured out the rest on its own.
and the moment the wallet hit $1,000, it did something I've never seen a human trader do. It stopped.
closed everything. Went flat. Sent me one line: "Target hit. Waiting for a new one."
I test AI trading repos so you can see what actually works before you put a dollar in. This is the first one that understood the word "enough."
the repo: HyperGrok. One Grok brain, 7 agents, Hyperliquid perps.
setup, 10 minutes:
1. Clone the repo
2. Add your Grok API key
3. Wallet: $100
4. The rule: stop at $1,000
the 6 days:
ETH long: +$55
BTC short: +$120
SOL long: −$30
HYPE long: +$210
DOGE long: +$545
$100 → $1,000. Exactly 10x. Then silence.
the honest part: DOGE made 61% of it. Without that trade, the target is still $545 away. Paper run.
next week I give it a harder rule. If you want to see whether it survives it, you know where to find me.
what rule would you give it?
after I read through Anthropic official guide on Reverse Engineering i gave my local LLM one app and one instruction: “replicate this"
then i logged everything.
- 00:00 drop in the binary. no source, no docs, nothing
- 00:06 50+ functions renamed. it reads like real code now
- 00:18 it finds the encryption routine buried deep inside
- 00:35 full login flow mapped, step by step
- 01:00 spec md done. enough to rebuild the app from zero
one hour. one laptop. zero cloud.
3 rules that made it actually work (keep this):
→ make the model quote the exact address for every claim
→ check anything important by actually running the code
→ keep one notes file it rereads every step
that's it. those 3 rules turn a guessing chatbot into a real reversing partner.
a year ago this was a week of work for a skilled human. now it's a coffee break.
i'm pushing this further every week: bigger apps, harder targets. stay tuned and watch where it breaks.
show this to the friend who still opens ida and suffers alone.
question for you: what's the first app you'd point it at?
tekario retweeted
FREE HECKING GOLD FOR ANY AI USER:
best 20 GitHub repos to make your own Reverse Engineering Crew to replicate any app you want
i wired these into opus 5.5 and it stopped being a chatbot.
it became a reversing crew:
[01] decompiler bridges: the agent drives the decompiler itself
- LaurieWired/GhidraMCP
- mrexodia/ida-pro-mcp
- fosdickio/binary_ninja_mcp
- radareorg/radare2-mcp
- cyberkaida/reverse-engineering-assistant
[02] ai copilots: explain, rename, comment inside your tool
- JusticeRage/Gepetto
- atredispartners/aidapal
- mahaloz/DAILA
- radareorg/r2ai
[03] runtime & mobile: debug and hook live processes
- AgentSmithers/x64dbgMCPServer
- signal-slot/mcp-gdb
- dnakov/frida-mcp
- zinja-coder/jadx-ai-mcp
- zinja-coder/apktool-mcp-server
[04] diff, decompile & fuzz: find what changed and prove the bug
- clearbluejar/ghidriff
- albertan017/LLM4Decompile
- google/oss-fuzz-gen
[05] benchmarks: check what the agent can really do
- andyzorigin/cybench
- NYU-LLM-CTF/NYU_CTF_Bench
- zhuyansen/awesome-ai-reverse-engineering
the flow: prompt -> mcp -> decompile -> debug -> patch -> verify
what i learned wiring this up:
- a smarter model doesn't win, the loop does.
- an agent that can only read pseudocode guesses.
- an agent that can also set a breakpoint, hook a function and diff two builds checks its own guesses.
that one change is what turns "AI explains code" into "AI does the reversing".
start with GhidraMCP + mcp-gdb. those two alone change how you work.
keep this for the next time a binary makes you want to quit.
i'm breaking down one of these setups end to end soon, so you'll want to be around for that.
and if you have a friend still reversing by hand at 2am, show them this.
be honest: would you trust an AI agent to patch a binary you ship?
FREE HECKING GOLD FOR ANY AI USER:
best 20 GitHub repos to make your own Reverse Engineering Crew to replicate any app you want
i wired these into opus 5.5 and it stopped being a chatbot.
it became a reversing crew:
[01] decompiler bridges: the agent drives the decompiler itself
- LaurieWired/GhidraMCP
- mrexodia/ida-pro-mcp
- fosdickio/binary_ninja_mcp
- radareorg/radare2-mcp
- cyberkaida/reverse-engineering-assistant
[02] ai copilots: explain, rename, comment inside your tool
- JusticeRage/Gepetto
- atredispartners/aidapal
- mahaloz/DAILA
- radareorg/r2ai
[03] runtime & mobile: debug and hook live processes
- AgentSmithers/x64dbgMCPServer
- signal-slot/mcp-gdb
- dnakov/frida-mcp
- zinja-coder/jadx-ai-mcp
- zinja-coder/apktool-mcp-server
[04] diff, decompile & fuzz: find what changed and prove the bug
- clearbluejar/ghidriff
- albertan017/LLM4Decompile
- google/oss-fuzz-gen
[05] benchmarks: check what the agent can really do
- andyzorigin/cybench
- NYU-LLM-CTF/NYU_CTF_Bench
- zhuyansen/awesome-ai-reverse-engineering
the flow: prompt -> mcp -> decompile -> debug -> patch -> verify
what i learned wiring this up:
- a smarter model doesn't win, the loop does.
- an agent that can only read pseudocode guesses.
- an agent that can also set a breakpoint, hook a function and diff two builds checks its own guesses.
that one change is what turns "AI explains code" into "AI does the reversing".
start with GhidraMCP + mcp-gdb. those two alone change how you work.
keep this for the next time a binary makes you want to quit.
i'm breaking down one of these setups end to end soon, so you'll want to be around for that.
and if you have a friend still reversing by hand at 2am, show them this.
be honest: would you trust an AI agent to patch a binary you ship?
tekario retweeted
a Chinese schoolkid with $5 can now outbuild your whole f*cking AI agent stack. here's the exact setup
step 1 → point Claude Code at DeepSeek's Anthropic endpoint and set EVERY model name to deepseek-flash. skip one and "opus" quietly goes to V4 Pro
step 2 → send every small call to Jev. "which file?" "retry or stop?" "ship it?" $0.042 per 1M tokens, output free
step 3 → one long session per project. cached input is $0.003 per 1M, so rereading your repo is basically free
step 4 → 2 tries per fix, then park it. heavy runs wait for off-peak, half the price
step 5 → the expensive model reads only the final diff. nothing else
Flash-1017 is about $0.20 per DeepSWE task on DeepSeek's own chart. Opus 5 is around $10. that's 50 Flash tasks for one
no team. no funding. just rules
one catch: the chart compares Opus 5, not 5.5. and Flash is cheaper, not faster
so who's still paying the $10 version for every step?
tekario retweeted
I GAVE MY GROK BOT $500 AND ONE ALPHA GITHUB REPO
7 days later it turned it into $4,870.
no charts. no signals group. no paid course. no staying up at 3am watching candles.
just Grok, 7 AI agents and a free open-source repo doing the work while I lived my life.
I test AI trading repos every week so you don't have to burn your own money finding out which ones are real. This is the first one that made me stop scrolling.
the repo is called HyperGrok. one Grok brain runs 7 agents that scan, research, read X sentiment, build a thesis, check risk, execute and review, all on Hyperliquid perps.
setup took 12 minutes:
1. Clone the repo
2. Add your Grok API key
3. Max leverage 20x, max risk per trade 10%
4. Paper mode on
5. Walk away
the trades:
BTC short: +$420
ETH long: +$310
SOL long: −$180
HYPE long: +$1,140
DOGE long: +$2,680
$500 → $4,870. that's 9.7x in a week.
the honest part: one DOGE trade did 61% of the work. Take it out and it's $2,190, still 4.4x.
if you know someone still paying for signals groups, send them this. It'll save them more than a subscription.
would you let Grok trade your $500, or keep doing it yourself?