
The first arbitrage bot in the vastness of Aptos that is ready to conquer Aptos Network Discord: https://teapawt.net/t.co/S6B1cfnzmR #Aptos #AptosEcosystem #AptosNetwork
Aptos Network
Joined January 2023
- Tweets30
- Following35
- Followers16
- Likes205
A Senior Google engineer just dropped a 10-page PDF on "the future of AI-powered agents"
Some of the biggest takeaways:
• AI identifies missing evidence and unsupported claims.
• It checks methodology, experiments, and citations against review criteria.
• It generates structured review feedback instead of free-form comments.
Research it now, then read the article below.
A Architect Anthropic engineer just dropped a 15-page PDF on "LOOPer: A Learned Automatic Code Optimizer" for LLMs and agentic systems.
This paper argues they should understand its structure first.
The workflow looks like this:
Identify sections → Recognize their role → Generate the summary
The key idea:
Structure comes before summarization.
Most agentic systems are getting better at generating.
Very few are getting better at understanding before they generate.
That's probably the more important capability.
Book & research it, than read an article below
Anthropic research team just dropped a 7-page PDF on "Agent-in-the-Loop" for agentic systems
The shift: you stop annotating data in big offline batches. You build live agent feedback directly into operations instead
Annotate → Feed back → Retrain → Deploy → Repeat
Every loop runs on four feedback signals:
• Preferences: agents pick the better of two responses - pairwise signal, not a vague score.
• Adoption: the system logs whether the agent actually used the answer, and why - the "thing that can say no."
• Knowledge checks: each retrieved doc gets flagged as relevant or not, so retrieval stops drifting
• Gap-finding: agents mark what knowledge was missing - the loop discovers its own blind spots
The key insight: a model graded only by offline batches never sees how it fails in the real world
In a live US support pilot this loop delivered +11.7% recall, +14.8% precision, +8.4% helpfulness, +4.5% adoption - and cut retraining from months to weeks
This paper changed how I think about improving LLM systems in production
Read it now, then explore the article below
Andrej Karpathy in 5 min reveal ALL TRUTH about AI: "Neural networks are complicated alien artifacts"
The man who built the foundations of modern AI just admitted we have no idea what we actually created
Evolution ran a multi-agent self-play system for billions of years
5 min. free. professional look from OpenAI co-founder
bookmark & watch
Anthropic's Head of Agents engineer:
"Don't build agents for everything"
The future of agents in 3 bullets:
→ Budget-aware agents (cost + latency control in production)
→ Self-evolving tools (agents that design their own tool stack)
→ Multi-agent collaboration (async, parallel, by end of 2025)
Barry from Anthropic's team revealed actual agents setup for daily using
Book & watch it today
Andrej Karpathy in 70 min reveal how LLM Quants actually uses AI: "i just tell the machine what i want, in plain words"
no 40-line system prompts. no magic.
by 2026 the engineer who dismisses LLMs loses to the junior who configured one right
70 min. free. professional look from OpenAI co-founde
bookmark & watch
Anthropic Head of Product and Financial Services:
"Anthropic engineers fire off swarms of 300+ agents a day.
You can hand your agents 100+ tools - just keep them out of the context window until they're needed."
The Anthropic team just spent 12 minutes explaining how to take agents all the way to production
Watch this & read the article below
Boris Cherny, Claude Code creator: "now i use agents and model writing 90% code for me"
That's how the guy who built Claude Code thinks about every task
"Right now im runing hundreds of agents with /loops and dynamic workflows”
in just 18 minutes, Boris revealed his daily setup
That’s worth more than a $500 course on loops engineering
Anthropic Head of Engineering for the Cloud Platform:
"Claude Fable 5 is not just another model drop"
"most developers are optimizing for the wrong version of Claude"
Anthropic literally said it on stage in Tokyo today
This is not a chatbot
This is infrastructure that gets better while you sleep
Worth more than a $500 agent building course
Live from the last Anthropic stage in Japan. Unpublished
Boris Cherny (Claude creator): "prompting" is the new "googling" - table stakes, not a skill
Meanwhile the guy who built Claude Code just explained why the chat window is already dead
AI isn't sitting in a tab waiting for your prompt anymore
it's on your machine. running code. using your other software. and increasingly - building itself
most people are still typing questions into a box
the ones paying attention moved somewhere else entirely:
→ they let the model run commands, not just suggest them
→ they hand it a terminal, not a text box
→ they build systems the AI operates inside of
the command line is the battlefield now. that's where "user" turns into "builder"
19 minutes that reframe what the next two years actually look like
you're not behind on access. you're behind on where the work moved
Boris Cherny (Claude Code's creator): "I deleted my IDE in November and hasn't opened it since. I writes more code than ever"
In 18 minutes , Boris breaks down his real daily Claude Code setup.
"I don't prompt Claude anymore. I create loops - and the loops do the work. My job is to create loops."
Worth more than a $500 vibe-coding course.
Boris Cherny (Claude Code's creator): "I deleted my IDE in November and hasn't opened it since. I writes more code than ever"
In 18 minutes , Boris breaks down his real daily Claude Code setup.
"I don't prompt Claude anymore. I create loops - and the loops do the work. My job is to create loops."
Worth more than a $500 vibe-coding course.
Boris Cherny (Claude Code's creator): "By 2026, 'good at prompting' is the new 'good at typing' - useless on its own."
the engineer who actually built Claude Code just dropped a free video that teaches you to write prompts that genuinely work - and i've seen $300 courses that don't cover half of what he gets through in the first 10 minutes
his whole point: the skill isn't typing nicer prompts anymore, it's building a system that runs itself:
• CLAUDEmd files done right (most people bloat them and quietly burn context before they even start)
• memory shortcuts that stop you repeating yourself every session
• parallel sessions - running multiple Claudes at once instead of babysitting one
• the prompting patterns that actually change the output, not the cosmetic ones
if you've been paying $20–200/month and still using Claude like a fancier ChatGPT, you've been driving a race car in first gear
skip whatever you were going to watch tonight - put this on instead
bookmark it now before your feed eats it
Karpathy said something most senior engineers didn't want to hear
the experts who dismiss LLMs will lose to the experts who learn to use them right
he's not talking about capability. he's talking about discipline.
the model isn't the problem. the way you talk to it is.
agents fail in ways nobody anticipated during training. you can't fix that by switching models. you fix it by routing around the failure modes.
that's what CLAUDE.md is actually for. it doesn't make Claude smarter. it raises the floor of what Claude won't do wrong.
Karpathy wrote 4 rules. they cut error rate from 41% to 11%.
I added 8 more, tuned to everything that broke after January. got it to 3%.
the gap between "AI is useless" and "AI prints money" is not the model. it's the config file sitting next to it.
Ex-Citibank quant trader built system that turned him $2,217 → $199,881 in one month
he's making $7,000–$10,000 every single day on autopilot
18,372 predictions, but no one knows he exists
the system runs a complete-set delta model across BTC, ETH, SOL and XRP simultaneously:
delta = P(UP) + P(DOWN) - 1.00
when delta < 0 - both sides sum below $1.00 - the system buys both and locks guaranteed profit on the gap
when one side drops below 20¢ - directional engine kicks in:
EV = (P_win x payout) - (P_loss x stake)
if EV > 0 and trend confirms - size with Kelly, enter before the crowd adjusts
right now he has 9 open positions across 4 assets in the same 5-minute window. running both engines at once. the arb leg covers downside, the directional leg prints on top
most traders pick one direction and hope. he removed hope from the equation entirely
Profile: polymarket.com/@0xce25e214d5…
🤝 Paid partnership
This trader joined today and turned $433 into $43,398 on a single bet with 9,900% return
0 views. No name - He's the one nobody knows or talks about and he quietly printed $52,258 in one day
Here's exactly what he did:
His wallet: polymarket.com/@0x676cdce125…
→ BTC Up or Down · May 25 · 4:00AM–4:05AM ET
→ Bought 43,398 shares of UP at 1¢ per share
→ Risked: $433
→ Won: $43,398
→ Return: +9,900%
The market was pricing BTC going up in the next 5 minutes at 1% probability
He disagreed. He bought 43,000 shares at a penny each
BTC went up. Every share paid out $1. He walked away with $43K
He's looking for moments when the market severely misprices a short-term BTC direction
🤝 Paid partnership
Jane Street pays $650,000 a year to quants who trade using neural networks
This 1 hour lecture of a Jane Street talk on real ML infrastructure > 6 months of any course on YouTube
You'll finally see what tier-1 quants actually optimize
Bookmark & watch, no matter what
Ex-meteorologist turned $400 → $143,076 with 1 strategy and quit forecasting weather for betting on it
He bets only on weather. And he reads it better than the crowd
BEST TRADES
$29.88 → $14,602 · Wellington highest temp 16°C · Yes @ 0.1¢
$25.41 → $12,452 · Tokyo highest temp 16°C · Yes @ 0.1¢
$24.88 → $12,398 · Chicago highest temp 54°F · Yes @ 0.1¢
His model: GFS + ECMWF before the market checks the forecast
He enters Yes or No at 98–99¢ when the weather models already know the answer. The market hasn't caught up yet
Same strategy as an insurance company
He sells certainty to the crowd that's unsure. Buys No at 99.1¢. It resolves at 100¢. +0.86%. But at $3,977 position size - that's $33 per trade, risk-free.
High-certainty weather positions: tiny return, zero variance, stacked in size
Profile: polymarket.com/@coldmath?r=j…
82 minutes of MIT real analysis on logs, exp and power series > 6 months of quant crypto alpha YouTube slop
You'll finally understand what convergence and continuity actually mean - the stuff Shannon, Kelly and Thorp built their entire edge on
Bookmark & watch, no matter what
The Quant Who Turned Shannon's 77-Year-Old Formula Into +$555K in 30 Days
He's not betting on crypto. He's compressing its uncertainty - one bit at a time
His strategy: Shannon → Kelly → Crypto Up/Down
The Formula
D_KL(P ∥ Q) = Σ p(xᵢ) · log₂ [ p(xᵢ) / q(xᵢ) ]
P = his model. Q = the market price. The result = how many bits of information he has that the market doesn't. That number IS the edge
KL > 0.10 bits → Pro zone. Trade
KL > 0.30 bits → Insane edge or bad math
Every position is a binary, ultra-short crypto market: BTC, ETH, SOL, XRP - Up or Down on 5-min and hourly windows
The same playground Shannon's math was built for
Fastest way to copy-trade him even with $10 using: kreo.app/@0xRicker
Kelly Sizing
Position size scales with edge in bits:
- High conviction (BTC Down 58.9¢): 2,999 shares
- Medium (ETH Down 63.9¢): 900 shares
- Long-tail (XRP Up 17.1¢): 158 shares
Same recipe Thorp used to run 20 seasons with zero losing years
While retail argues vibes on 5-min candles, this wallet runs KL-divergence and Kelly - 33,795 times in a row
polymarket.com/profile/%400x…
You could literally:
> open Polymarket
> find BTC or ETH window under 15¢
> ask Claude when Markov persistence ≥ 0.87
> enter
> risk $8.7 to make $91
The math that prints $17–20K every week for this bot
Why do people still not doing this?
This HFT bot prints $17–20K every week using one strategy and turned $215 → $199,214
$17,492 last week alone. 53,571 total predictions. Still going
The entry range is the widest of any bot I've tracked: 8.7¢ to 97¢
Two modes running at once:
Mode 1 - High-confidence scalping (81–97¢)
Enters when the market has already committed a direction but still leaves 3–19¢ on the table. Small spread, small size, extremely high win rate
Mode 2 - Extreme discount entries (8.7¢–57¢)
Fires when the Markov matrix shows a committed state that the market is pricing as near-impossible. One position at 8.7¢ is currently sitting at 9.5¢ - unrealized but loaded
Fastest way to copy-trade him even with $10 using: kreo.app/@0xRicker
350 trades/day × 30 days = 10,500 trades per month
Mean return per trade implied: $17,492 / ~2,200 weekly trades ≈ $7.95 net per prediction
The bot doesn't bet big on any single position - $1,309 biggest win confirms small sizing
Profile: polymarket.com/@gabigol?r=jo…
Does that remind you of anyone from @Polymarket?
82 minutes of an MIT lecture on Volatility Modeling > 6 months of any trading course or guides on YouTube
You'll finally understand what conditional variance, GARCH and volatility clustering actually mean
Bookmark & watch, no matter what
This bot made $84,708 last month and 32k last week
Has 0 profile views. Nobody is watching
9,323 predictions. $17.6K biggest win. $117.7K deployed right now
P(Xₙ₊₁ = j | Xₙ = i) = pᵢⱼ ≥ 0.87 → ENTER
When that condition holds at 47¢, the edge is +112%. When it holds at 73¢, the edge is +37%
Mean entry across active positions ≈ 63¢ → avg r ≈ +58% per resolution
The formula works regardless of which direction the market has committed to
Fastest way to copy-trade him even with $10 using: kreo.app/@0xRicker
Pure BTC directional scalping across hourly and 5-minute windows
Entry zone 47¢–87¢ - the bot reads both sides simultaneously, betting Up and Down on different time windows at the same time.
$117.7K deployed × 58% avg return = $68K theoretical monthly yield on active capital
Profile: polymarket.com/@0x04b6d7e930…
This 1 hour MIT lecture will teach you more about Portfolio Management than 20 years inside any hedge fund, investment bank or financial institution
Bookmark & watch, no matter what. It'll be the most productive thing you do this week
A trader who hasn't lost a single trade in last 158 trades and has made $286,403 profit
He only bets on No - and always win
I've never seen $286K deployed with zero losses
He doesn't trade price. He trades impossibility
His secret?
The math that makes this work:
EV = P(No) × Payout − Cost
When the market says 99¢ on "BTC dips to $30K" - it's free money
Fastest way to copy-trade him even with $10 using: kreo.app/@0xRicker
• Find the extreme targets the crowd fears but math says won't happen
• Bet "No" when odds swell to 89–99¢ — the premium is the profit
• Stack across multiple dip levels: $15K, $20K, $25K, $30K, $35K, $40K, $45K
• Size up - $46K to $149K per position
$26.8K biggest single win. $579K active right now. Still compounding.
Profile: polymarket.com/@lucasmeow?r=…

























