SolPepe

[Antid @antisadh]

A 9 CARD RTX 3070 RIG MADE $3 A DAY MINING ETHEREUM IN 2021, THE SAME RIG NOW RENTS TO AI STARTUPS ON VAST AI FOR $1,200 A MONTH

00:05 he points at the rig, "here is my rig with 9 RTX 3070s and it actually makes $3, mining ethereum, money goes directly to my binance wallet"

he ran the same 9 card rig through 2022 mining ethereum on the high1 pool, the merge in september killed gpu mining for everyone without their own power plant

he sat on the dead rig for 3 years, in early 2026 he installed the vast ai host stack on ubuntu, listed all 9 cards at $0.18 an hour and went live the same evening

each RTX 3070 now nets about $130 a month renting to AI startups that can't get GPU access from AWS or Azure, the rig clears $1,200 a month after electricity and platform fees

nvidia can't ship H100s fast enough so smaller AI companies pay whatever the marketplace charges, the mining infrastructure built for ethereum is the inference infrastructure for chatgpt and claude now

the window is open, follow and bookmark before it closes

[Raytar @Raytar]

Met a guy making $1.6 million a year.

Three days ago he was at a Meta conference. Told me he saw the best AI talk of his life.

Boris Cherny was on stage. Showed how the Anthropic team actually uses Claude day to day.

Boris deleted his IDE eight months ago. Now he codes from his phone.

I watched it last night. Had to pause it twice.

Not because it was hard. Because I realized I've been using Claude like a toy.

He sent me the recording. It was never published.

Posting it below.

[爱吃折耳根的Ace @_zheergen]

🚨 谷歌悄无声息地扔出预测未来的AI核武器！

TimesFM：扔一段历史数据，几秒钟就能预测未来！ 销售预测、股价走势、网站流量、能源负荷、加密货币波动……

现在 24K+ stars，安装超简单（一行命令搞定）：
pip install timesfm[torch]

这些让人头疼的数字，现在终于有统一答案了！TimesFM 核心亮点：

> 在 1000 亿 真实世界时间序列数据上训练
> 零样本直接预测，完全不用微调或再训练
> 几秒出结果，还自带置信区间
> 100% 本地运行 + 100% 开源，普通电脑就能跑
> 完全免费！ 以前熬夜调 ARIMA、Prophet、Excel，现在一行代码搞定。

Google @GoogleAI 自己已经在 BigQuery 和 Sheets 里大规模使用。预测这件事，终于要变简单了！

[Ridark @ridark_eth]

I genuinely don't understand why everyone isn't using this yet

Andrej Karpathy, a co-founder of OpenAI, posted a simple idea that hit 16 million views: stop using AI to write code, use it to build a second brain.

You point Claude Code at a folder, drop in any source, an article, a transcript, a PDF, and Claude reads it, links it, and files it into a living wiki of everything you know. It compounds like interest, the more you feed it, the smarter it gets.

Here's the whole thing:

> Install Obsidian, create a vault, open it in Claude Code

> Paste Karpathy's wiki idea file and tell Claude to build it

> Claude makes three folders: raw for sources, wiki for its pages, a CLAUDE.md that runs it

> Drop any source into raw and say "ingest this"

> Ask questions across everything, forever

Five minutes to set up, and you never start from a blank chat again.

Full step-by-step guide with Claude and Obsidian, link below.

Bookmark this

[me_cool_off @me_cool_off]

This bot made +$60,000 in one month trading 5-minute markets on Polymarket!

Account: https://t.co/VE7mju7ptz

+$27,000 over the last week.

+$2,000/day!

Best trades:
$1,913 -> $4,380
$1,552 -> $3,465
$1,833 -> $3,738

The bot uses two strategies.

1. Asymmetric spread trading:
> buys both outcomes (Up + Down)
> larger size on the outcome favored by the model
> small profit if correct
> small loss if wrong
> edge appears across thousands of trades

2. Fast scalping:
> buys $500-1,000 right after market open
> sells after a 1-3% move
> typically captures $5-30 per trade

Both strategies require high volume and come with significant fee costs.

So far the bot has paid over $40,000 in taker fees.

At the same time, it has earned +$25,000 in maker rebates.

Current net profit: +$45,000.

An important reminder: PnL without fees means very little.

Many bots show a profitable PnL chart but end up losing money after fees.

Modern trading bots must account for commissions, slippage, and execution quality - not just strategy performance.

This bot does exactly that.

[Skynet Insights @Skynet_insights]

https://t.co/ziwz2msNQ0

[s1rozha1 @s1rozha_]

23-YEAR-OLD BUILT AN AI-CODED APP TO $20K/MONTH IN 50 DAYS

Connor’s apps now do $1M+/year in sales

his latest app, Payout, was built in under 2 weeks with AI and won a hackathon with 55,000+ entrants

the playbook:

> don’t build social apps
> build simple utility apps
> copy what is already validated
> screenshot 20 competitors
> rebuild the best onboarding patterns in Figma
> make the paywall emotional, personal, and obvious
> give Claude a JSON doc of your data structures
> ship 1-3 core features fast

the growth loop:

> partner with niche creators
> make relatable UGC
> find the videos that work organically
> turn winners into paid ads
> optimize for yearly subscriptions

his stack:

> Claude Code
> Figma
> Expo
> Next.js + TypeScript
> Vercel
> Mixpanel
> RevenueCat

the lesson:

a simple app with a great onboarding can beat a complex app with bad distribution

bookmark this before you build your next app

[Rugikk @rugikkk]

Claude Code ran 36 hours straight and shipped a payment-ready SaaS while I slept.

One change made it possible.

Everyone builds the master plan first. The 200-line spec that scopes the whole project.
That’s the mistake.

I added 1 step before it.

A meta-planning pass. Before any code, Claude Code stops and asks one question:

“What still needs a human here?”

API keys. Stripe setup. Vercel config. PostHog events. The boring 100% that turns a demo into a live product.

So it maps every tool, MCP server, skill, and browser session it needs to do that work itself.

Then it provisions them.

By hour 30 it was inside the Stripe dashboard. Wiring checkout. Testing a live charge.

Then Vercel. Then PostHog.

No keys pasted by me. No dashboards opened by me.

Old way: you babysit the agent and click the buttons it can’t reach.

New way: the agent reaches the buttons.

36 hours. 0 manual config. 1 step nobody adds.

[鈴木@アナログ営業会社を100日後にAIで売上を300％にする人 @ai_300]

【Codex無料版】
OpenCodeのデスクトップ版がヤバい
ChatGPT plusもClaude CodeもGeminiも
サブスクモデルをまとめて使える
しかも使い方はCodexとほぼ同じ
これで月100万狙う人が出る　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　

もう「AIコーディングは高い」は終わり。

OpenCodeは、無料で使えるオープンソースの
AIコーディングアプリ。
しかもデスクトップ版。
やることはシンプル。

GPTで設計
Claudeでコード生成
Geminiで調査
OpenCodeで実装
AIエージェントで修正

これを1画面で回す。
初心者が失敗するのは、
ツールを1個ずつ開いて、
毎回コピペして、
途中で何を作ってるか忘れること。
それ捨てろ。
これからは、
「AIに作業手順を渡す」
「ファイルを読ませる」
「Todoを分解させる」
「コードを書かせる」
「エラー修正まで投げる」
この流れで勝てる。

Codexっぽい操作感で、
無料でここまで触れるなら、
副業アプリ、LP、業務ツール、Bot作成のハードルは一気に下がる。

コードが書けない人ほど今すぐ触れ。
稼げない理由は才能じゃない。
新しい道具を触るのが遅いだけ。
行動しない人間から脱落する。

[Mustafa @oprydai]

markov chains are simple but powerful.

they model systems where the next state depends on the current state.

not the whole past.

• states → possible conditions
• transitions → ways the system moves
• probabilities → uncertainty with structure
• memorylessness → only now matters
• steady state → long term behavior emerges

weather.
markets.
language.
biology.
robot navigation.
queues.

markov chains teach you a brutal systems lesson:
complex behavior can emerge from simple transition rules.

[Julian Goldie SEO @JulianGoldieSEO]

5 reasons Kimmy K 2.7 + Obsidian works so well:

→ Kimmy handles coding tasks

→ Obsidian stores long-term memory

→ MCP connects agents to the vault

→ Agent OS keeps everything organized

→ Every agent reads from the same brain

That means less repeating.

Less lost context.

More compounding output.

Save this video, you’ll build AI systems that remember.

Want the SOP? DM me. 💬

[HodlReaper @HodlReaper]

$10,000 a month from a SaaS you built with Claude is real. 9 in 10 people chasing it never clear $0.

Here’s the part the hype skips.

Building stopped being the hard part. Claude ships a working MVP in a weekend now. So does everyone else’s.

The moat moved. It’s 50 people who will pay you every month. That’s the whole game.

Honest timeline:

First paying customer takes 2 to 3 months, not 2 days.

Most micro-SaaS dies under $500 MRR.

The ones that clear $10K charge real prices and grind distribution every single day.

What those accounts actually do:

1. Pick a boring, narrow problem. Invoicing for plumbers beats “AI assistant for everyone.”

2. Charge $40-100 a month. 200 customers at $50 is $10K . 200 is reachable. 20,000 free users is not.

3. Use Claude for the unglamorous work too - support replies, onboarding docs, cold email drafts.

4. Ship distribution, not just product. 1 channel, posted daily, for 6 months.

The trap: rebuild the product 10 times, talk to 0 customers. Claude made building cheap, so building is worthless as an edge.

Spend the saved time where the money is - in front of people who have the problem.

Claude builds the product in a weekend. The customers still take a year.

[EMOTION CRYPTO @Emotion_web3]

This guy built a fortune trading weather markets and BTC Up/Down.

His monthly profit reached an insane $400,000.

Total PnL: $668,000

Think there’s a secret? There isn’t one.

→ uses @weatherscan_bot
→ finds underpriced markets
→ follows his own strategy
→ repeats it
→ takes profit

No hype. No crypto lottery.

Just data, discipline and execution.

While everyone is chasing the next random 100x, he is quietly printing from the most ignored markets on Polymarket.

I’d watch this trader closely.

Because while others ignore these setups, you can be early enough to study his strategy and copy the best trades before the crowd notices.

Wallet: https://t.co/99vPX94eyq

[Livsun @L1vsun]

every time you click buy, someone on the other side already solved for two numbers

you don't even know the names of

number one: their theoretical fair value. not the price on your screen. not the analyst target

their own model, updated in real time from order flow, volatility surface, and live inventory

they already know what the asset is "really" worth before you've opened the chart

number two: the probability you have information they don't. market makers call it adverse selection

they literally price the spread based on how likely you are to know something. wide spread means they think you might

tight spread means they think you're just noise

here's what makes this brutal

you spent 4 hours reading the earnings call, studying the chart, checking reddit

the other side of your trade spent 40ms

not reading faster - solving a completely different equation

Bookmark this

you were asking "is this a good buy?"

they were asking "what's the chance this person has real alpha and how much margin do i need if they do?"

two different games running on the same order book at the same time

most retail traders don't know the second game exists

and that's the entire edge right there

[The Rogue Chartist @DanielMCharter]

I highly recommend this book!

Author: @SJosephBurns

"Mathematical trading means defining every aspect of your approach in quantifiable terms and executing those terms with discipline. This starts with position sizing, arguably the most important mathematical concept in trading."

[Moysei @0xMoysei]

A 17-year-old says he makes $1 million a month from one app. You take a picture of your food and it counts the calories.

No degree. No team you'd recognize.
The app is called Cal AI. Point your phone at a meal and it returns the calorie count instantly.

The whole pitch is one sentence: snap a photo, get the calories. He built it to be simpler than MyFitnessPal, where most people quit because logging food by hand is tedious. He cut the work down to a single tap.

He's 17 and building mobile consumer apps for a living. One camera, one number, and an app doing seven figures a month while most people his age are in class.

Note: $1 million a month is his own claim in the clip, unverified. Cal AI is real and was co-founded by two teenagers, so the figure is reported by the founder, not confirmed revenue.

A 17-year-old, one photo, and a million a month off counting calories.

[wluha @cryptowluha]

A 19 year old built an AI girl in one evening.

Last month she made $21,847.

One guy sent her $200, then came back the next day.

Another has spent over $4,000 this year.

He thinks she's real.

She isn't.

The entire thing runs from a laptop on his parents' WiFi.

Total cost: $21 a month.

That's the part that feels fake.

Not the revenue.

The barrier to building it.

A year ago this would've taken a team.

Now it's a weekend project.

[Sandy4ka @sandy4kad]

Local businesses lose thousands of dollars a month because they have no system to collect Google reviews.

Someone built the fix in an afternoon and charges $500/month for it.

Here's exactly how it works.

A customer pays at a restaurant through Square or books through Calendly. 2 hours later, Zapier fires a trigger. The customer gets a text with a 1-tap link directly to the Google review page. The restaurant owner touched nothing.

The SaaS itself has 4 components:

→ Automated SMS with personalized messages so each text reads like it came from that specific restaurant
→ Direct review link so customers don't have to search for the business
→ Separate dashboard login for every business on the backend
→ Analytics page showing the owner exactly how many reviews came in this week, this month, all time

That analytics page is the retention mechanism. Visual proof that the tool works is what keeps owners paying month after month without questioning the invoice.

The build happened inside Emergent — no hosting, no server setup, handles everything end-to-end. Design was based on a Mercury-style dashboard so it looks like something a business owner would actually trust and show to their staff.

Total build time: 1 afternoon. 1 detailed prompt. Full-stack system with database, login, and messaging.

$500/month per location. Nail 10 local businesses and that's $5,000/month recurring from a tool you built once.

Every restaurant, dentist, gym, and salon in your city needs this. Most have no system at all.

The market is sitting there.

[Blockchain Surfer ☝️ @0xSurferX]

Your bot is losing because of WRONG data source.

Leaderboard bot owners take it form another place.

Here's where we get the data and how to tap most of it for free:

The basic mistake is polling.

Your bot asks "any updates?" every second like a human refreshing the page.

The pros subscribe to a stream that pushes data the instant it changes.

One fires when ready, one gets there first.

Here are the real sources:

> Binance websocket feed

Binance leads the Polymarket CLOB by about 1 second on crypto markets.

Wire your bot to the real Binance ws and you're seeing the future of the Polymarket book.

> Chainlink price feeds

Free, real-time, and authenticated with an HMAC API.

Most devs don't know this exists cause nobody talks about it.

> NOAA gov socket for weather markets

Official US weather data has a real-time feed that beats most third-party APIs by several seconds.

> Hidden websockets on news sites

A lot of data providers run a live ws behind their slow public page.

Watch the page's own network traffic with mitmproxy or websocat and you'll find it.

Turns a 2-second poll into a live push feed.

> Your own CLOB and Gamma recordings

None of the above replaces your own tick data.

Record every tick, full book depth, every resolution.

That's your backtest foundation and nobody else has it.

The pros aren't smarter about signals.

They're faster about data, and most of their sources are free.

I built 700+ bots and dirty data killed more of them than bad strategy ever did.

My public wallet with $24k PnL: [https://t.co/hIum0iETkG]

Fix the inputs before you touch the logic.

[Moon Dev @MoonDevOnYT]

i stopped manual backtesting because claude opus 4.5 can run entire trading systems while i sleep.

this agent just turned hours of coding into minutes and i am showing you exactly how to integrate it into your own setup for free.

stop building from scratch and start letting ai do the heavy lifting before the rest of the market catches on.