Nikos Tsoukas

[Vanja @VanjaPoker]

I've been coaching poker players for years.

Low stakes up to nosebleeds.

Here are 3 common mistakes players make and what to do instead to crush🧵

[Vanja @VanjaPoker]

Want to improve fast at poker?

Good, here is:

How to massively boost your winrate in 30 days.

[Vanja @VanjaPoker]

99% of poker players are missing this

Here are 10 simple exploits to crush before the flop:

[Forge Mode | Self-Mastery® @HoliHappiness]

This is the perfect plan for a longer, healthier life..

-Thread-

[Vanja @VanjaPoker]

Forget everything you think about poker math. Here’s what actually matters.

1. Pot odds: how often you need to win to justify a call. Break-even = Risk ÷ (Risk + Reward)

2. Alpha: how often your bluff needs to work to profit. α = Risk ÷ (Risk + Reward)

3. MDF: how often you need to defend to stop profitable bluffs. MDF = 1 − α

4. EV: (Win% × Reward) − (Loss% × Risk)

5. Value-to-bluff ratio: on the river, bluff exactly as often as your opponent’s pot odds.

Don’t overcomplicate it.

You got this champ ❤️

[Vanja @VanjaPoker]

I'm 26

I wasted years learning this the hard way.

But I'll teach you in 5 minutes.

Here are 10 uncomfortable truths I wish I knew earlie about poker:

[Vanja @VanjaPoker]

Most poker players misplay 200bb Poker

Here are 7 things you need to understand to crush deep stack:

1. 3-bet size goes down OOP and up IP:
If you reach a certain threshold (170-180bb), 3-bet sizing becomes counterintuitive. You want to put less money in if you are OOP because you want to minimize pot growth with infinite SPR behind.
The deeper you become, the worse being OOP is. You are forced to play more streets and make more decisions, so EQR goes down and down.
Vice versa for IP. Here is where you want to grow the pot; the deeper you are, the more favorable IP becomes.

2. Offsuit hands go down in value:
The deeper you become, the more important EQR is. The threshold for getting money in also goes up. At 20bb, a single top pair is getting it in; AQo outperforms ATs.
At 200bb, you need a much stronger range to stack off. You want hands that can cooler people.

3. Nut ability is more important:
Similar to the point before, you want hands that have more ability to make the nuts on all streets. If you x/r the flop, having stronger draws that can cooler weaker ones outperforms.

4. If you have a strong nut advantage, you often want to increase sizing and check more frequently:
The deeper you become, the stronger your range needs to be to stack off, so you often check more in polar spots, increase your size, and use a stronger range.

5. If you are OOP as the PFA, you often size down on very connected, middling, and lower boards:
Your overpairs are worth less and runouts change the nuts too often. You want to be more careful and wait for safe runouts. On the other hand, on boards where you have a strong advantage and want to rush money in (e.g. 223 BTN vs SB), you want very large sizes.

6. x/r frequency goes down:
Hands are worth less the deeper you become. You want a tighter, more EQ-driven range.

7. You want to implement massive sizes:
Especially in spots where you have a strong nut advantage and your range supports getting all the money in, you naturally need larger sizes in your arsenal. This is especially true in spots where action is backloaded (flop and turn didn’t go in much money).

I recommend comparing 100 and 200bb sims to spot changes yourself.
If you have questions, let me know :)

[Dami-Defi @DamiDefi]

How to use Claude Code to build your first AI product even if you’re not technical (full beginner course)

1. How to actually get good results from AI coding tools

➤ Why people get bad results

* They start coding before planning.
* They give vague prompts and expect great output.
* They let the AI make too many decisions.
* They ask for too much in one go.
* They do not answer clarifying questions properly.
* They do not test the output early.
* They do not save progress, so one bad change can ruin momentum.

➤ What works better

* Start in planning mode first
* Use planning before execution.
* Let the tool create a structure before it starts building.

* Be clear about what you want
* Say what you want built.
* Say how it should work.
* Say what matters most.
* Say what to keep simple.

* Give small, specific tasks
* One feature at a time
* One change at a time
* One revision at a time

* Answer clarifying questions
* Claude asks follow-up questions before building.
* Those questions reduce guessing and improve the output.

* Test quickly
* Run the file
* Open it in the browser
* See what actually works
* Then revise

➤ Practical prompt lesson from the transcript

Instead of:
* “Make a game”

Do this:
* describe the type of game
* explain controls
* explain the style
* explain features
* explain what should stay simple

Example:
* top-down shooter
* browser-based
* 2D retro style
* character with a gun
* enemies coming from different angles
* move with arrow keys
* aim and click with the mouse
* levels
* menu screen
* keep it relatively simple

Best takeaway

The more specific you are, the less the AI has to guess.

2. The setup most beginners skip before using Claude Code

➤ What Claude Code actually is

* A terminal / command-line coding tool from Anthropic
* You type instructions in plain language
* It generates code, edits projects, debugs, and builds apps

➤ What you need before starting

* A Claude subscription or API credits

* A terminal environment:
* PowerShell on Windows
* Terminal on Mac/Linux

➤ Installation basics

* Install Claude Code using the command for your operating system
* Run claude in the terminal to test if it works
* Authenticate in the browser
* Trust the current folder when prompted

➤ Important beginner point

* Claude Code is not a normal desktop app
* It runs inside the terminal
* Every time you want to use it, you reopen terminal and run claude

➤ Easier option for beginners

If the terminal feels intimidating:

* use the Claude desktop app
* switch into code mode
* choose project
* prompt directly from the app
* use asking / planning / coding modes there too

➤ From the transcript:

* the desktop app is easier
* but the CLI is better for serious work

➤ Best working environment

I recommend:

* open a project folder in Cursor, VS Code, or a similar editor
* open the terminal inside the editor
* run Claude there

➤ Why this matters:

* you can see the files being created live
* you can view the sidebar with project files
* you can watch changes happen in real time
* it is less confusing than using the terminal alone

➤ Useful setup actions from the transcript

* create a dedicated folder for the project
* open that folder in your editor
* open terminal inside that folder
* run claude
* trust the folder
* start building there

Best takeaway

Do not use Claude Code in a messy setup. Use a proper project folder inside an editor so you can see what the AI is doing.

3. Why Git and GitHub are non-negotiable when using AI agents

➤ Why Git matters

Git gives you a way to:

* checkpoint progress
* save versions
* revert mistakes
* avoid losing working code

➤ I explicitly recommend installing Git before doing real work.

Why this is especially important with AI agents

* AI can make mistakes
* AI can change many files quickly
* AI can take the project in the wrong direction
* without version control, it is harder to recover

➤ What I recommend

* install Git
* create a GitHub account
* connect your local setup to GitHub
* let Claude help with setup if needed

➤ What GitHub adds

GitHub gives you:

* remote backup of the code
* access from another computer
* easy sharing with others
* a way to recover the project if the folder is deleted or the machine is lost

➤ Actionable workflow

Tell Claude something like:

* make a new Git repository for this project
* create clean commit messages
* save commits locally
* push them to GitHub regularly
* keep the project backed up so we can revert anytime

➤ Why this is a big deal

Many tutorials skip this.

But this is one of the most practical parts because:

* it protects the work
* it reduces fear when making changes
* it makes longer projects safer

➤ Useful setup flow pulled from the transcript

* install Git
* create GitHub account
* install GitHub CLI if needed
* run login flow
* authorize GitHub in terminal
* let Claude create repo and start using Git/GitHub for the project

Best takeaway

AI agents are much more useful when every good version of your project is saved and recoverable.

4. The smartest workflow for building with Claude Code

➤ The workflow the transcript suggests

Step 1: Set up the folder and environment

* create a new folder for the project
* open it in Cursor / VS Code
* open terminal inside the editor
* run claude

➤ Step 2: Start in planning mode

* switch to planning mode
* think through what you want to build first
* stream your thoughts into Claude if needed

➤ Step 3: Describe the project in plain English

You do not need to know the codebase first.

* you do not need to know the languages
* you do not need to have done research
* it helps, but you can start from scratch

➤ Step 4: Ask Claude to create a plan

* give the rough project idea
* then explicitly say: create a plan

➤ Step 5: Answer follow-up questions

Claude may ask about:

* sound
* size
* other missing details

Answering these helps it produce a tighter plan.

➤ Step 6: Approve execution

* once the plan is ready
* tell it to proceed
* optionally auto-accept edits

➤ Step 7: Let it build the first version

In the transcript, the first version was a playable web shooter game.

➤ Step 8: Run the output

* open the generated file
* test it in the browser
* check if it matches the prompt

➤ Step 9: Revise in small steps

Instead of rebuilding everything:

* add one feature
* improve one part
* make one change at a time

➤ Step 10: Save with Git/GitHub

* commit progress
* push to GitHub
* keep the project protected

Extra workflow lesson from the transcript

The most important part of AI coding is:

* knowing what you want

I specifically warn:

* if you let the agent steer you, you get horrible crazy results
* if you are clear and specific and do one small task at a time, you can get fantastic results

➤ Model selection workflow

Use:

* Opus for more complex work
* Sonnet for middle-ground work
* Haiku for fast, cheaper, simpler tasks

➤ This matters because:

* usage can get expensive
* you can run through credits quickly
* not every task needs the most powerful model

Best takeaway

The real workflow is not “prompt once and hope.” It is plan, clarify, execute, test, revise, and save.

5. The Claude Code features that make power users better than beginners

Core modes

➤ Claude Code has 3 important modes:

1. Ask mode

* default mode
* asks before doing things
* safer for beginners

2. Code mode

* automatically accepts edits
* faster for building
* better when you trust the direction

3. Plan mode

* creates a detailed plan first
* gives more structure
* one of the best ways to improve output quality

➤ Useful commands and shortcuts from the transcript

* /help
* shows commands and available actions

* /model
* lets you switch between Opus, Sonnet, and Haiku

* /tasks
* shows background tasks currently running

* /init
* creates the CLAUDE.md file

* /agents
* opens agent functionality

* /exit
* exits the current session

➤ Useful keyboard shortcuts mentioned

* Shift + Tab
* cycle between modes

* Escape twice
* clear what is currently typed

* Up / down arrows
* cycle through previous prompts/commands

* Alt/Option + Enter
* add a line break without submitting

* Ctrl + O
* expand verbose view / logs

* Ctrl + B
* run as a background session

* Ctrl + T
* view tasks

* Escape
* interrupt execution

➤ Background tasks

Claude can run long tasks in the background, such as:

* web servers
* ongoing processes
* things that would otherwise block the terminal

Why this matters:

* you can keep working while the task runs
* you do not lose your main workflow

➤ CLAUDE.md

This is one of the most useful features.

Why it matters:
* Claude sessions do not keep unlimited memory
* when you restart, session context is gone
* CLAUDE.md gives Claude persistent project memory

What it can store:

* project overview
* how to run the project
* development setup
* design patterns
* workflow expectations
* rules Claude should follow

The speaker also updates it with:
* commit regularly
* push to GitHub
* keep clean commit messages
* do not lose work

➤ Agents, skills, MCP, hooks

These are more advanced features mentioned near the end.

Agents
* specialized helpers for different tasks
* can be delegated work

Skills
* repeatable tasks Claude can learn and run again

MCP
* lets Claude connect to outside services like Notion, Gmail, Google Drive, etc.

Hooks
* can trigger actions automatically

Best takeaway

Beginners use Claude like a chatbot. Power users use modes, commands, memory files, tasks, and workflows.

Strongest overall TLDR

If you want the cleanest possible summary of the whole video:

* Claude Code works best when you use it inside a proper project folder and code editor
* Start in planning mode before building
* Be very clear about what you want
* Let Claude ask clarifying questions
* Build in small steps
* Test every version
* Use Git and GitHub from the start
* Create a CLAUDE.md file so Claude remembers project rules
* Use the right model for the right task
* Treat AI coding like a workflow, not a magic prompt

Thanks to @TechWithTimm for putting this video together.

[AI Edge @aiedge_]

f*ck your weekend plans.

You NEED to:

• Learn Claude Code
• Set up Perplexity Computer
• Set up Claude Cowork (plug-ins, skills)
• Set up OpenClaw
• Experiment with agentic solutions
• Use AI to create a business plan & strategy
• Build an AI second-brain database
• Learn basic automation tools (Manus, MCP, Zapier)
• Become an elite prompt-engineer - the better you can communicate with AI, the better your Outputs
• Read AI articles
• Dive into robotics
• Research AI stocks/ETFs/investment arbitrages

The list goes on.

SO much to do.

[David @david_eng_mba]

Why Math Says This Is the Largest Pricing Error in Bitcoin History (≈105% Implied 12-Month CAGR)

Bitcoin is trading at a −35.5% deviation below its 15-year power-law trend. That is not an opinion; it is a statistical displacement the market is currently ignoring.

Power-law fair value today: $122,425
Spot price: ~$79K
That places Bitcoin firmly in the historical “oversold” regime (Z-score: −0.63).

At this depth, price doesn’t just "drift" back to trend.

It snaps.

I back tested every comparable oversold event since 2010.
Results over the following 12 months:

Win rate: 100%
Average return: +100%+

Sentiment was irrelevant every time. The Ornstein-Uhlenbeck (OU) mean reversion process was not.

This deviation has a measurable Half-Life: 133 days.
In simple terms: The market historically corrects 50% of its pricing error every 4.4 months, 100% in ~9 months.

The $43,457 gap is a compressed spring. As it relaxes, the "snap-back" velocity dictates the path:

June 2026: $113K
October 2026: $145K
January 2027: $162K

Model fit: R^2 = 0.96 (Solid)
18-month predictive correlation: 0.55
(55% of the price movement 18 months from now is statistically explained solely by the Z-score (deviation).

We are at the extreme left tail of the distribution. This is where expected value concentrates. Math supports an aggressive ~0.6x Half-Kelly allocation.

The market is offering a significant discount.

Closing the Gap

Today
Bitcoin is ~$43.5k below its power-law trend value, a −35.5% deviation. This is the extreme left tail. Historically, this is where forward returns concentrate because the error is too large to persist.

Oct 2026
The gap compresses to ~$11k (−6.8%).
That implies roughly 75% of the anomaly has already reverted. At this point, the trade is no longer “deep value” it’s transitioning into normalization.

Implied fair value ≈ $155k
Implied Bitcoin price ≈ $145k

Jan 2027
The gap shrinks to ~$7k (−4%).

Implied fair value ≈ $168k
Implied Bitcoin price ≈ $162k

CAGR ~105% (Next 12 Months)

Why This Analysis Is Robust:

The Power Law captures Bitcoin's diminishing returns and logarithmic adoption curve (R² = 0.96).

It’s Mean-Reverting: The OU Process proves that price is tethered to value. The further it stretches (Z-score), the stronger the force pulling it back.

It’s Statistically Significant: The 18-month predictive correlation is 0.55. This means 55% of Bitcoin's future price action is explained solely by today's deviation. That is an incredibly high signal-to-noise ratio for any asset class.

[Uri Peleg @UriPelegPoker]

If I had to start poker all over again, I’m not sure I would.
Not for the reason you think.
Poker is a great career choice if you have the skill and passion for it.
There are few options like it -
You progress at your own pace, purely based on merit.
Progression if you’re good can be extremely rapid - your hourly wage can increase from year to year by large margins.
The cap is also unlike a lot of mainstream jobs, especially for someone young. You can achieve extremely high income levels much faster than you would at any other profession.
Freedom - you’re your own boss. You choose your vacations, your works hours, your time off. Not feeling it - don’t go.
Socially - you meet a very special group of people. All of them off the beaten path. Maybe this is just my experience, but I’ve encountered extremely kind, smart and creative people of many different types, and benefited a lot as a result.
So why wouldn’t I do it over again?
Because right now, it’s very intimidating. Solvers - make it feel like chess. I need to learn a lot. I need to memorise a lot. I need to work a lot. And people have been doing this for years.
That’s a very daunting and unattractive situation for me to start with.
Back when I started… take a pre flop chart, read or book or two, maybe watch a few training videos - that was it. That was the knowledge you and everyone else had to go off of - and we can start playing and thinking for ourselves. Had there been these solvers that I need to study, aggregate reports, different sizing for different boards - honestly I’d never have started in the first place.
The funny thing is… from my perspective now - I understand that it’s really not about that.
If you had to start poker right now - I wouldn’t want you looking at a solver for a while. Watch a few videos, just like before. Maybe read a book. Start playing, start thinking. Stay away from solvers - you’re not equipped to deal with them in a way that will benefit you.
Learn how much your hand is worth. Learn which cards are good for which player. Learn about concepts - polarisation. Geometric sizing. Equity realisation.
Pot odds. Player types. Types of mistakes. Exploits.
What do you do if someone calls 3bets too much pre flop?
What do you do if someone folds too much?
How do you adjust against someone aggressive? Someone passive? Someone trappy? How do you recognise these tendencies fast?
Timing! Betsizing tells?
These are all things I knew before solvers. And then used solvers to help myself refine and understand better.
If you just jump straight in, and start studying what size to bet on what flop, you’re in for a long and frustrating journey.
A journey like doing a college degree. But you’re here playing poker, and not in college, so I know that’s not what you want.
Poker is awesome. But in the current climate I might never have given it a chance.`

[Edgy - The DeFi Edge 🗡️ @thedefiedge]

A few under the radar DeFi tools you should try (most of them are free):

• @Footprint_Data – Think of it like Dune, but faster. Pre-built dashboards and cleaner data to slice GameFi, NFT or DeFi activity without SQL.

• @coindar_ – Simple but underrated: a clean crypto event calendar for launches, updates, and milestones.

• @chain_broker – Tracks VC rounds, token sales and private raises. See who’s backing what before it hits the timeline.

• @stablewatchHQ – A stablecoin heatmap that shows which ones are gaining traction and where liquidity’s moving, along with trending tokens, performance comparisons and much more.

• @elfa_ai – Real-time social listening dashboard. Tracks which tokens and narratives are gaining mindshare across X.

• @_dexuai – Your AI copilot for narratives, whale moves, and tokenless plays. Surfaces what’s trending before CT does.

• @ScopeProtocol – Tracks wallet behavior, fund flows and smart money trails with an AI layer on top.

• @revertfinance – Portfolio tracker for LPs. Monitor Uniswap or Sushi positions and auto-manage risk.

• @hypurrdash – Hyperliquid trader tracker. Follow the top perps traders and see what they’re doing right now.

• @tokenomist_ai – Keeps track of upcoming token unlocks and vesting events so you’re never blindsided by supply dumps.

• @BubbleMaps – Visualizes wallet networks as bubbles. Instantly spot whale clusters, insider ties, or suspicious holdings.

I didn’t include every tool I use because I don’t want to overwhelm you.

Let me know if there’s any other under the radar tools you guys recommend!

[Market Rebellion @RebellioMarket]

Market Wizard Linda Reschke's 12 Technical Trading Rules: @SJosephBurns

1. Buy the first pullback after a new high. Sell the first rally after a new low.

2. Afternoon strength or weakness should have follow through the next day.

3. The best trading reversals occur in the morning, not the afternoon.

4. The larger the market gaps, the greater the odds of continuation and a trend.

5. The way the market trades around the previous day’s high or low is a good indicator of the market’s technical strength or weakness.

6. The previous day’s high and low are two very important “pivot” points, for this was the definitive point where buyers or sellers came in the day before. Look for the market to either test and reverse off these points, or push through and show signs of continuation.

7. The last hour often tells the truth about how strong a trend truly is. “Smart” money shows their hand in the last hour, continuing to mark positions in their favor. As long as a market is having consecutive strong closes, look for up-trend to continue. The up trend is most likely to end when there is a morning rally first, followed by a weak close.

8. High volume on the close implies continuity the next morning in the direction of the last half-hour. In a strongly trending market, look for resumption of the trend in the last hour.

9. The first hour’s range establishes the framework for the rest of the trading day.

10. A greater percentage of the day’s range occurs in the first hour then was the case in the past, and thus it has become increasingly important to trade aggressively if there are early signs of a strong trend for the day.

11. There are four basic principles of price behavior which have held up over time. Confidence that a type of price action is a true principle is what allows a trader to develop a systematic approach.

The following four principles can be modeled and quantified and hold true for all time frames, all markets. The majority of patterns or systems that have a demonstrable edge are based on one of these four enduring principles of price behavior.

Charles Dow was one of the first to touch on them in his writings. Principle One:

A Trend Has a Higher Probability of Continuation than Reversal Principle Two:

Momentum Precedes Price Principle Three:

Trends End in a Climax Principle Four:

The Market Alternates between Range Expansion and Range Contraction!

12. In the world of money, which is a world shaped by human behavior, nobody has the foggiest notion of what will happen in the future. Mark that word –

Nobody! Thus the successful trader does not base moves on what supposedly will happen but reacts instead to what does happen.

[Julius @jv_finance]

This is the full model.

Note that the bottom has never been broken — while Bitcoin broke above the top in 2013 and 2017.

[Plan C @TheRealPlanC]

The wait is over!

Introducing... 🥁

Bitcoin Quantile Model v2.

You’re going to want to bookmark this post—and follow for regular model updates.

After months of research and development, I’m very proud of this model—my flagship quantile framework.

I’m confident it’s one of the best—if not the best—long-term Bitcoin investment frameworks available.

As a full-time, unpaid Bitcoin researcher, I’m often asked how people can best support my free content.

Simply bookmark, repost, and comment on my posts :)

Thanks for all your continued support!

— PlanC

Key Features & Improvements:

1. Quantile lines never cross—mathematically impossible.

2. Cycle-length agnostic.

3. 133,000+ data points and 1,500 lines of code.

4. Fits and stores 999 quantile levels (τ = 0.001–0.999 in 0.001 steps) and identifies which level the last price is closest to.

5. Fits the two leading decay functions (stretched exponential decay & exponential decay) and selects the better fit via quantile-appropriate AIC.

Uses Akaike weights to identify the best-supported model.

Akaike weights (AIC-based):
Stretched exponential decay: 96.4%
Exponential decay: 3.6%

6. Piecewise Quantile Regression — Linear + Stretched Exponential Decay (Nonlinear).

[Swan @Swan]

Lyn Alden just dropped a 2-hour masterclass with Tom Bilyeu to his 4.5M+ subscribers.

If you don’t understand what’s happening to money right now, you’ll wake up on the wrong side of history.

Here are 10 insights on where the system is breaking—and what comes next. 🧵👇

[bencb @bencb789]

Every poker player hates this: You 3-bet AK … and whiff entirely the flop.

Here’s how to turn those spots into profit in less than ONE MINUTE.

You think you should check? Wrong. Here is why 👇

[🔖Bookmark this now so you finally understand how to play AK]

[tee @fienixtaranova]

My physical therapist had me doing stuff like this for months. I laughed at first. Then I struggled. Then I got cleared to stop going to therapy and have no back pain like before.

[shaunda devens @shaundadevens]

1/ Ethena's USDe supply is up $3.7B in just 20 days, driven primarily by the Pendle-Aave PT-USDe looping strategy.

With $4.3B (60% of total USDe) now locked in Pendle and $3B deposited in Aave, here's a breakdown of the PT loop mechanics, growth vectors, and potential risks:

[Mark Jeffrey @markjeffrey]

Many of you have heard me talk about Bittensor $TAO.

This article is a 'highlight reel' of what's going on inside of the 128 subnets (AI companies) that power Bittensor.

The metrics already being achieved are stunning.