Kagi retweeted
this is f*cking wild.
i simulated a polymarket copy bot for 27 days. $200 in paper money became $11,460.
the whole stack:
• Anthropic's coding agent
• 4 open-source repos
• $25 a month
step 1 - find the smart money
one prompt: "find wallets with 100+ trades and 70%+ win rate. rank by profit."
14,212 wallets scanned. 47 kept.
step 2 - cut the markets
a Rust CLI pulls 500 live markets. it checks spread, liquidity, timing and whale moves.
500 became 35. 93% rejected before i looked.
step 3 - 3 agents vote
arbitrage, convergence, whale copy. no shared memory.
2 agree = full size. 1 agrees = half size. 0 = no trade.
this cut 40% of the losing trades.
step 4 - leave early
91% of smart wallets exit before the end.
so the bot exits first, at 85% of the expected move.
after 27 days:
271 trades
74.2% win rate
sharpe 2.47
max drawdown -$61
it still took a -$23 loss on Oil $90. the agents said full size. remember that part.
most people show you the green line. nobody shows the 214 trades the bot refused to take.
the stack costs less than Netflix.
next run: 1,000 markets instead of 500. what breaks first?
Kagi retweeted
it's getting f*cking cheap to run Claude Code. one flag puts Sonnet 5.5 on the code and calls Opus 5.5 in for 3 decisions
claude --model sonnet --advisor opus
Opus stays on call and writes 0 lines of code
→ launch the session with that flag, Sonnet 5.5 takes every routine turn
→ hand the searching to 3 Haiku 5.5 subagents: search, grep, docs
→ Sonnet usually calls Opus before a plan locks
→ it calls Opus again when the same error comes back twice
→ and a 3rd time before it says "done"
→ Opus reads the full session each time, every tool call and every result, and sends advice back
→ add 1 line to your prompt for more checks: "consult the advisor before you lock a plan"
every Opus call re-reads the whole session from zero, with no cache. a long session with many calls eats the savings, remember that part
it's also experimental and runs only on Anthropic's API, not on Bedrock or Google Cloud
most setups run Opus on every turn, file reads included
$0 extra tools. on a subscription, advisor tokens count toward the plan you already pay for
the video shows a mock session with 41 turns. 3 of them go to Opus
how many of your turns today needed Opus?
i show you how to let AI do your work, daily.
Kagi retweeted
9 prompts. 1 skill. a six-legged crane walks, lifts, grasps and comes apart into labeled parts, and every move reads one 2,400 rpm input. you are still keyframing one gear by hand
the skill is Elvar's teardown. free, runs in Claude Code, ships the whole machine as one offline HTML page
the rule inside it is simple. every part is a function of one input. on this board that input is a V8 power pack at 2,400 rpm. the crankshaft with 8 pistons, the 96-tooth slewing ring, 6 leg drives, the boom and the hook all read the same model. so walking, lifting, grasping, the exploded view and the lidar sweep are one simulator in six modes, not six animations glued together
here is the gist:
→ type /teardown, name the machine, attach one real reference. without it you get a toy
→ split it into assemblies you can isolate: hull, boom, legs, manipulator, optics
→ give every part a real number before it gets a shape: teeth, sections, drives, pistons
→ drive everything from that one input
→ add modes as views of the same model: walk, lift, grasp, explode, scan
→ press any part and it slides out with its label while the rest steps aside
→ play it at 1/20 speed and watch for the one leg that is out of step
six modes, one model. fix one number and every mode fixes with it
the 9 prompts are in the quoted article
legs on a tripod gait or a boom slewing 360° on 96 teeth, which assembly breaks first when you build yours?
↳ concept board, crane is illustrative, method from the article
it's getting f*cking cheap to run Claude Code. one flag puts Sonnet 5.5 on the code and calls Opus 5.5 in for 3 decisions
claude --model sonnet --advisor opus
Opus stays on call and writes 0 lines of code
→ launch the session with that flag, Sonnet 5.5 takes every routine turn
→ hand the searching to 3 Haiku 5.5 subagents: search, grep, docs
→ Sonnet usually calls Opus before a plan locks
→ it calls Opus again when the same error comes back twice
→ and a 3rd time before it says "done"
→ Opus reads the full session each time, every tool call and every result, and sends advice back
→ add 1 line to your prompt for more checks: "consult the advisor before you lock a plan"
every Opus call re-reads the whole session from zero, with no cache. a long session with many calls eats the savings, remember that part
it's also experimental and runs only on Anthropic's API, not on Bedrock or Google Cloud
most setups run Opus on every turn, file reads included
$0 extra tools. on a subscription, advisor tokens count toward the plan you already pay for
the video shows a mock session with 41 turns. 3 of them go to Opus
how many of your turns today needed Opus?
i show you how to let AI do your work, daily.
Kagi retweeted
Grok Bot DJ'd an entire night set, and I think this is the most honest way anyone has ever shown trading.
The idea is simple. The bot is a DJ. Every coin is a track. Two turntables, Deck A plays the trade that's live right now, and Deck B is already cueing up the next one while it gets scored. In the middle there's a mixer, and every knob is an agent: Scout, Filter, Analyst, Grok, Sniper, Guard.
But the part I keep coming back to is the playlist.
$OKRA, played, +$1,544
$WREK, skipped, dev rug ×3
$JOLT, playing now
$FLOP, skipped, volume $3k
$MUDD, skipped, mint live
$GHOUL, skipped, LP 0%
$DREG, skipped, top10 47%
By morning the count said 11 played, 250 skipped.
Any good DJ will tell you the set isn't about the tracks you play. It's about the hundreds you don't. Same here. Grok Bot spent most of the night saying no, and that's exactly why the set worked.
The waveform at the top is the whole night as a song. Every colored block is a trade, and every arrow shows where it came in and where it dropped out. You can scroll through the set and see exactly where it hit: +$1,862 on $RAVN, +$2,655 on $ZENO with a big "DROP!" right on the spectrum. And you can see where it missed: -$415, -$128. The misses are short. The hits play out.
$4,180 at 23:00. $14,746 by the end of the set. 8 wins, 3 losses, 3.53× on paper.
My honest opinion:
I love this concept because it fixes something every trading dashboard gets wrong. Normal screens make you stare at one number going up and down. Here you see a set with a rhythm: build up, drop, transition, next track. Trading done right actually feels like that. Patience, then a quick, clean move, then patience again.
And the losses feel like a bad track, not a disaster. A DJ plays a song that doesn't land and just mixes into the next one. That's exactly how a bot should treat a loss: no revenge, no panic, just the next track.
The one thing I'd turn down is the leverage. Some of those tracks run at 40x, and that's the volume knob cranked to the max. One wrong note at 40x is a lot louder than at 5x. Keep the playlist rules, lower the volume.
Comment "SET" and I'll DM you how each knob on the mixer works and the exact skip rules behind the playlist
And tell me in the replies: which skipped track name made you laugh the most? I'm going with $WREK, dev rug ×3, because that dev clearly has a whole discography
Kagi retweeted
THIS IS F*CKING DANGEROUS. AN EX-CURSOR ENGINEER SAYS SHE DOESN'T WRITE CODE ANYMORE. SHE RUNS 50+ AI AGENTS
1 chief of staff, 3 PMs, 40+ workers. they build while she sleeps
most people open one AI chat, ask one question and close the tab. she runs 50 at once
how her setup works:
→ the chief of staff takes the goal and plans the day
→ 3 PMs split the plan into tasks
→ 40+ workers build at the same time, one task each
→ in the morning she checks the results over coffee
50+ agents
1 human
"the output of 20 engineers", in her words
SpaceXAI already runs agents like this overnight. they only come back to a human when someone has to decide
start small tonight: 1 planner, 2 workers, 1 goal. check the results in the morning
her agents worked all night. yours is still waiting for your next question
THIS IS F*CKING DANGEROUS. AN EX-CURSOR ENGINEER SAYS SHE DOESN'T WRITE CODE ANYMORE. SHE RUNS 50+ AI AGENTS
1 chief of staff, 3 PMs, 40+ workers. they build while she sleeps
most people open one AI chat, ask one question and close the tab. she runs 50 at once
how her setup works:
→ the chief of staff takes the goal and plans the day
→ 3 PMs split the plan into tasks
→ 40+ workers build at the same time, one task each
→ in the morning she checks the results over coffee
50+ agents
1 human
"the output of 20 engineers", in her words
SpaceXAI already runs agents like this overnight. they only come back to a human when someone has to decide
start small tonight: 1 planner, 2 workers, 1 goal. check the results in the morning
her agents worked all night. yours is still waiting for your next question
Kagi retweeted
this is f*cking wild.
a 20-year-old Japanese student built a trading bot with Anthropic's AI in 48 hours. $68 turned into $750K.
that's the story going around AI twitter right now.
the video below is my version of that dashboard. same idea, simulated numbers.
here is how a bot like that makes money:
→ watches 50 short BTC up/down markets on Polymarket at once
→ pulls the BTC price from Binance every second
→ when Binance moves and Polymarket is still slow, it buys the side that's about to win
→ the gap closes in seconds and the bot keeps the difference
→ then it does it again. 297 trades in my sim, 2 losses
it never reads a chart. it never gets scared after a loss. it never guesses where BTC goes, it reacts to a price that already moved.
now look at the bottom right corner of the chart.
it says SIMULATED. the original video says it too.
a simulation has no fees, no slippage, no rejected orders. a live market has all three.
and the "free setup, just comment" offer under those posts is not a secret being shared. it's how the account gets reach.
most people sharing it never looked.
Kagi retweeted
it is getting f*cking weird: Anthropic's new agent Duet plugs into 6,000+ apps, and it still can't trade or move money between your accounts. even if you say yes
a business owner with about 110 employees says one agent brings it all together: no more opening 100 tabs
the 3 levels Anthropic built in:
1. does it alone: reads, drafts, prepares
2. asks you first: send, pay, sign, delete
3. never, even with your yes: trading, moving money between accounts
what the day looks like:
→ every few hours: checks your connected apps
→ morning: one short brief on what matters
→ all day: pings you only when something needs you
→ on its own cloud computer, so your laptop can be off
→ before it sends, pays, signs or deletes: waits for you
6,000+ apps → 4 actions it asks about → 0 trades → 1 yes from you
honest part: Anthropic says it can misread a thread or misjudge urgency, and you can't watch its browser yet. it's a research preview, US Pro and Max users first, through a waitlist
steal the rules. paste this into your own assistant:
"sort every task into 3 levels. level 1: do it alone and tell me after: read, summarize, draft. level 2: show me the exact action and wait for my yes: send, pay, sign, delete. level 3: never do it, even if i say yes: trading, moving money between accounts. if you are not sure, use level 2"
the app has a waitlist. the rules don't.
Kagi retweeted
Your ChatGPT chats are training data by default.
Your Gemini chats might be read by a human.
Your Claude chats can be stored for 5 years.
Here's how to turn all 3 off in 2 minutes.
Think about what you've pasted in there.
The medical question you were too embarrassed to ask a doctor.
The bank statement you asked it to explain.
Your ex's texts you asked it to analyze.
Your CV with your home address. Your salary. Your contract.
The code with your API key still in it.
Nobody asked you. It was just on.
ChatGPT trains on Free and Plus chats unless you switch it off. Most people never open that menu.
Google says it plainly in its own privacy notice: a subset of Gemini chats is reviewed by humans, including outside contractors. Reviewed chats are kept for up to 3 years. Even if you delete your activity.
And there's a trap. Even with activity off, if you hit thumbs up or thumbs down on a Gemini answer, that chat gets reviewed and kept for up to 3 years.
Claude asked everyone to pick a setting last year. If "Help improve Claude" is on, your chats are kept for 5 years. If it's off, 30 days.
It gets worse. Last year shared ChatGPT conversations started showing up in Google search. Mental health confessions. Job candidate reviews. Private source code. All one search away. OpenAI pulled the feature, but old shared links can still be live.
Here's the full fix. Takes 2 minutes.
ChatGPT → Settings → Data Controls → turn off "Improve the model for everyone"
ChatGPT → Settings → Data Controls → Shared Links → delete every link you don't need
ChatGPT → Settings → Personalization → Memory → read what it remembers about you. Delete what you don't want there.
Claude → Settings → Privacy → turn off "Help improve Claude"
Gemini → myactivity.google.com/produc… → turn off "Keep Activity"
For anything sensitive from now on, use Temporary Chat in ChatGPT. It doesn't train the model.
Turning these off only covers future chats. Go back and delete the old ones you'd regret.
You'll spend more time reading this post than fixing it.
Bookmark this so you don't lose it, and send it to anyone who uses AI every day.
it is getting f*cking weird: Anthropic's new agent Duet plugs into 6,000+ apps, and it still can't trade or move money between your accounts. even if you say yes
a business owner with about 110 employees says one agent brings it all together: no more opening 100 tabs
the 3 levels Anthropic built in:
1. does it alone: reads, drafts, prepares
2. asks you first: send, pay, sign, delete
3. never, even with your yes: trading, moving money between accounts
what the day looks like:
→ every few hours: checks your connected apps
→ morning: one short brief on what matters
→ all day: pings you only when something needs you
→ on its own cloud computer, so your laptop can be off
→ before it sends, pays, signs or deletes: waits for you
6,000+ apps → 4 actions it asks about → 0 trades → 1 yes from you
honest part: Anthropic says it can misread a thread or misjudge urgency, and you can't watch its browser yet. it's a research preview, US Pro and Max users first, through a waitlist
steal the rules. paste this into your own assistant:
"sort every task into 3 levels. level 1: do it alone and tell me after: read, summarize, draft. level 2: show me the exact action and wait for my yes: send, pay, sign, delete. level 3: never do it, even if i say yes: trading, moving money between accounts. if you are not sure, use level 2"
the app has a waitlist. the rules don't.
Kagi retweeted
this is f*cking wild: Anthropic deleted over 80% of its own agent’s prompt. it got no worse. you’re still adding rules to yours.
an Anthropic engineer explained why: older models needed hard rules and repeats. the new ones have better judgement.
your agent reads your whole CLAUDE.md at the start of every chat. every old rule is noise it reads again and again.
one command finds them: /doctor prompt-audit
it splits the work:
→ reads your CLAUDE.md, AGENTS.md and rules files, so nothing hides in a side file
→ flags lines written for older models, so you see what the new ones don’t need
→ finds rules that fight each other, so your agent stops guessing. two rules clash and it may just pick one
→ finds links to files that don’t exist, so your agent stops reading about nothing
→ shows the file and line for every flag. you fix each one in one edit
the numbers:
→ over 80% of the system prompt removed
→ no measurable loss on coding evals
→ 1 command. Claude Code v2.1.283 or newer
how: run claude update, open your project, type /doctor prompt-audit
new models need fewer rules. every extra line is noise your agent reads every time.
Anthropic already cleaned up its own prompt.
yours is still written for the old models.
Kagi retweeted
$15,000 A MONTH FOR AN EMPLOYEE OR $200 FOR AI. HE PICKED AI
his whole point is pretty simple. don't hire anyone at the start, pay for ChatGPT Pro and do the work yourself. and start posting content from day one, every day, because followers barely mean anything now. the For You page decides who sees you
honestly I agree with most of it
but nobody talks about the annoying part. paying $200 takes 10 seconds. posting every day when like 30 people see your stuff is way harder, and thats where most people just stop
I've been running my account since early august and yeah. some days a post still gets 9 views lol
AI saved me money for sure. but it can't make you keep going when nobody's watching
what actually works for me is pretty boring. show up every day, jump on trends while they're fresh, and stop checking the numbers every hour. the posts that blew up for me were never the ones I expected
this is f*cking wild: Anthropic deleted over 80% of its own agent’s prompt. it got no worse. you’re still adding rules to yours.
an Anthropic engineer explained why: older models needed hard rules and repeats. the new ones have better judgement.
your agent reads your whole CLAUDE.md at the start of every chat. every old rule is noise it reads again and again.
one command finds them: /doctor prompt-audit
it splits the work:
→ reads your CLAUDE.md, AGENTS.md and rules files, so nothing hides in a side file
→ flags lines written for older models, so you see what the new ones don’t need
→ finds rules that fight each other, so your agent stops guessing. two rules clash and it may just pick one
→ finds links to files that don’t exist, so your agent stops reading about nothing
→ shows the file and line for every flag. you fix each one in one edit
the numbers:
→ over 80% of the system prompt removed
→ no measurable loss on coding evals
→ 1 command. Claude Code v2.1.283 or newer
how: run claude update, open your project, type /doctor prompt-audit
new models need fewer rules. every extra line is noise your agent reads every time.
Anthropic already cleaned up its own prompt.
yours is still written for the old models.
Kagi retweeted
this is f*cking insane, Claude Code reverse-engineered a dead app and got 9 years of notes out of it
the app stopped getting updates in 2022. no export, no docs, no support email that answers. just a 2.3 GB data file in a format nobody ever wrote down
every OS update it opened a little slower. I knew one of them would be the last, and everything I'd written since 2017 would go with it
so at 23:04 I gave Claude Code a copy of that file and one rule: the original stays untouched
23:07 → readers point REA at the app and find the function that saves a note
00:52 → the format comes out: compressed 4 KB blocks chained together, images in a separate pool linked by ID
02:14 → mappers finish the spec. 31 fields, each one traced to the code that writes it
03:21 → first exporter run. 200 notes out, 6 of them missing their images
04:06 → the bug: two notes can share one picture, and the exporter only copied it once. fixed in one line
05:49 → full run. 14,212 notes as Markdown, 4,517 attachments, every original date kept
05:58 → checkers compare 200 random notes with the old app. 0 mismatches
handwriting and audio stayed raw and marked "check by hand". 0 bytes changed in the original
the bug at 4am is the reason I trust the rest. the exporter said it was done. the checkers opened the old app and proved it wasn't
that's the whole setup in one sentence: the agent that builds the export never gets to say it works
my notes now sit in a plain folder of Markdown files. no app, no account, nothing that can stop opening next year
meet the agents that got my notes back ↓
Kagi retweeted
I CAN'T F*CKING BELIEVE IT: ANTHROPIC'S ENGINEERS PUT 1 AGENT AGAINST A TEAM OF AGENTS.
THE TEAM WON BY 90.2% ON THEIR OWN RESEARCH TEST.
i drew it: 1 window fills up, cracks, splits into 4. an illustration, not a live run.
you give 1 chat the whole research job. 1 window holds every search.
how the team works:
→ 1 lead agent (Opus 4) reads the question and makes a plan
→ it starts 3 to 5 sub-agents (Sonnet 4) at the same time
→ each one searches in its own window
→ each one hands back only the key parts
→ the lead puts it all into 1 answer
→ parallel search cuts research time by up to 90% on complex questions
remember this part: it uses about 15x more tokens than a normal chat. 1 agent alone uses about 4x.
most people look for a smarter model. on a hard web search test, token use alone explained 80% of the difference in scores.
try it today: split 1 big task into 3 small chats. paste only their summaries into a 4th.
would you pay 15x the tokens for +90.2%?
Kagi retweeted
You don't need to show your face to make money on YouTube 👇
Faceless AI "Sleep Stories" channels are blowing up. Long videos, calm voice, zero editing skills.
How it works:
Pick a calm topic (space, history, fantasy)
Claude writes a 30-min story
ElevenLabs turns it into a soft voiceover
Add slow AI visuals or one looping image
Upload 3x a week
Long watch time = better monetization chances.
Honest part: it takes 2-3 months of consistent posting before it pays anything. Most quit in week 2.
Follow @AIPulseNewsX for new AI tools 🚀
Save and retweet if this was useful 🔁
I CAN'T F*CKING BELIEVE IT: ANTHROPIC'S ENGINEERS PUT 1 AGENT AGAINST A TEAM OF AGENTS.
THE TEAM WON BY 90.2% ON THEIR OWN RESEARCH TEST.
i drew it: 1 window fills up, cracks, splits into 4. an illustration, not a live run.
you give 1 chat the whole research job. 1 window holds every search.
how the team works:
→ 1 lead agent (Opus 4) reads the question and makes a plan
→ it starts 3 to 5 sub-agents (Sonnet 4) at the same time
→ each one searches in its own window
→ each one hands back only the key parts
→ the lead puts it all into 1 answer
→ parallel search cuts research time by up to 90% on complex questions
remember this part: it uses about 15x more tokens than a normal chat. 1 agent alone uses about 4x.
most people look for a smarter model. on a hard web search test, token use alone explained 80% of the difference in scores.
try it today: split 1 big task into 3 small chats. paste only their summaries into a 4th.
would you pay 15x the tokens for +90.2%?
