Adea

hey, i’m Adea been in crypto for ~4 years now. started with nft degen stuff, survived meme coin trenches and somehow ended up falling into AI rabbit holes. on this account i’ll mostly post about crypto, AI, internet things i find interesting and whatever else i’m currently obsessed with at 2am. still pretty new to posting consistently, so i appreciate every follow & mutual ♡

@0xOstap the interviewing yourself part is way more interesting than the $40k most people have useful knowledge, they just don’t know how to extract it

A 20-year-old used Claude to build a $40k/month business without ever hiring a designer or writer. He didn’t use it to write emails. He used it to interview himself. > Claude asked one question at a time > pulled out his skills and the problems he’d solved > packaged it into a $1,000 digital product then picked his video topics to sell it No big audience. Every video gets 500 to 1,000 views. Convert one buyer per thousand and that’s the whole game.

@cryptowluha this was a genuinely interesting read most people stop at “AI influencer makes money” you went much deeper into why it works in the first place

@skynews357 ngl a stack of Mac Minis replacing a $199/month setup would’ve sounded ridiculous not that long ago

This Chinese Developer Replaced a $199/month AI Stack With Stacked Mac Minis. Research firms charge $5,000+ for similar output. He pulled Kimi K2.6 locally instead of renting access. 300 parallel agents. Reports in hours instead of weeks. His setup paid itself back in around 3 months. Power cost stayed low. That’s the part people miss. Most people optimize prompts. A small group is optimizing where the model runs. Subscriptions become owned compute. Monthly spend becomes hardware. Bookmark this before local AI infrastructure becomes normal 👇

@0xKiyoro the weirdest part is that most followers probably don’t care whether she’s real or not

$30,000 a month from an influencer who costs a GPU and a prompt to run A real model at this level needs a photographer, a studio, lighting, edits, and a team, she needs none of it because she was rendered, not born > 301,000 followers, a face, a bio, a fanbase that pays his rent > every photo is generated, every "selfie" is a prompt, she's never held a phone > followers send money, gifts, DMs at 2 AM, none have met her > she never ages, never says no, never asks for a cut An Indian man in his 20s built her in an afternoon the cheapest influencer in the world isn't even a person, and she out-earns the ones who are

@0xWast3 everyone wants safe AI until the safer model loses to the faster one

Anthropic's CEO: "We believe powerful AI should be safe and accessible." In this announcement, he explains why Anthropic delayed a more powerful model, how Claude Fable 5 was built with stronger safeguards, and what makes it their most capable AI yet. Bookmark and watch the full video 👇🏼

@kirbxbt technically the membership is still helping you gain mass

using my gym membership for free parking to go the mcdonalds near it

@erikalee this post just fixed my posture for at least 7 seconds

reminder to myself to sit up straight

@mdmadeit ngl she actually feels like a character, not just an ai generated design

meet Lina “Uptown” Reyes born from the block powered by her graffiti made with seedance 2.0 and chatgpt edited with CapCut (character details below)

@Haleeeemahh AI wrapped is basically a sleep deprivation report disguised as a productivity report

just checked my “Ai wrapped” my highest activity being 1am is crazy I see why I have dark circles now I’ve burned a total of 501.5M tokens I just know spotify would be jealous can we say I’m dev maxxing now ? check yours: wrapped.entelligence.ai

@Lummox_eth kinda funny how the opportunity was hidden in the complaints the whole time

CHASE TRENDING PRODUCTS. IGNORE THEM AND FIND 48% NET MARGIN INSTEAD. Most sellers open Helium 10, find a trending category, and walk into 50 competitors with dead margins. Trending lists are a graveyard of late arrivals racing to the bottom. The real intelligence was in 1 and 2-star reviews. Component failures. Mechanism flaws. Repeated across thousands of purchases nobody acted on. One category. 3,000 reviews. Claude processed 500 in two minutes and returned JSON with complaint frequency and fix difficulty. One flaw appeared in 31% of all negative reviews across every major listing. A spring-lock button jamming after 4 weeks. Every seller copied it without questioning. One Alibaba search. Four responses in 72 hours. Unit cost $1.30. Sale price $14.99. Net margin 48% against a 20 to 26% category benchmark. The gap is still open in most categories.

@JanguNFT the weird part is that online personas were already partially fictional. AI just makes them scalable

@Adea0x people keep saying AI is replacing jobs, meanwhile it’s starting to replace entire online personas

A 19-YEAR-OLD BUILT AN AI GIRLFRIEND FOR $21/MONTH. Guys message her every day. Send tips. Buy gifts. One even paid for a trip just to meet her. There was nobody there. She’s an AI avatar running on a laptop. When he waves, she waves. When he smiles, she smiles. Everything happens in real time. He showed me his dashboard last week. His entire setup: > 1 laptop > 1 AI tool > $21/month One morning alone: > €47 donation > €112 donation > €83 donation The barrier to creating a believable online identity is collapsing fast.

A SINGLE FILE TURNED CLAUDE INTO 10 DIFFERENT AI WORKERS No fine-tuning. No agents framework. No custom app. Just a folder with instructions Claude can pick up when the task is relevant. What you can create: > meeting notes → structured action plans
> design briefs → poster concepts
> messy docs → clean summaries
> recurring tasks → reusable workflows
> niche processes → dedicated Claude skills The crazy part? You don’t have to remind Claude every time. Name the skill. Describe when to use it. Write the process once. Claude detects the right skill and runs it when needed. Most people are still writing the same prompt 50 times. Claude Skills turns prompts into reusable software.

@srinigoes @snooziesNFT the hardest part is that real utility usually looks boring compared to giveaways and hype

Now that I am on the other side, my mind is constantly trying to figure out how to bring utility to NFT holders It's easy to criticize projects for not bringing enough utilities for NFT holders, but when you are in the hot seat, things are different 😅 Buying the floor, doing giveaways, raffles, etc are just band aid and they don't do much in the long run. You need to think about the holder who will stay loyal for the next 1-2 years, how do you reward that person? I don't have the answers, but that's how I'm thinking 🫡

@Cryptking_1 @wallchain ngl getting rejected from all 3 at once would sting a little. respect for posting it anyway

🚨 Punched in the Face Could be the Best Thing. At 6:12 am this morning, just like most I received my @wallchain rejection letter for all 3 campaigns. They say how you react after getting punched in the face is what defines your worth in Web3 or in life. If your feeling some type of way, move on and find a creative way to shift that energy. Use it as motivation to grow more and connect more. Worth is not revealed when everything is working. Worth shows up after the hit. That is when you find out your value. Wallchain is building their system, you keep building your system. There is no crying in Web3. Only coming back with vengeance.

@0x_fokki AI keeps making production cheaper, but attention isn’t getting any cheaper at all

🚨do you understand what $5.5/month just made possible.. before: - studio animator: $6,000/month - voiceover session: $300 per episode - composer per score: $2,000 today: > Claude scripts every episode: 10 min > Midjourney draws every frame: 20 min > Runway animates every scene: 15 min > ElevenLabs performs every line: 10 min > Suno scores every cut: 5 min > Make ships everything: 0 min $12,345/month out. the gap between them and this is $124. full build in the article above

@DfxoqEth the line between a prediction and a case study is getting weirdly thin

@Adea0x that is exactly the point, the story does not need to be real, the setup is

A financial journalist in Shanghai spent three days trying to profile a solo ETH trader who had posted an 847% return in 14 months The profile never ran The journalist arrived expecting a desk full of analysts. He found one man, one chair, and two monitors in an apartment that smelled like instant noodles At 0:17 the man has his arms crossed. He is not watching the left screen. The left screen is a live order flow visualization with 23 active levels mapped across the ETH order book. Yellow lines for institutional accumulation. Pink for distribution. The man did not draw those levels. The agent drew them at 4:53 AM before the Shanghai session opened The journalist asked how many screens he monitors at once He said one. The right one. The left one is not for him At 0:54 the journalist asked him to zoom in. Pause there. The price is 25,091.50. There are 7 active entries queued. The man has not touched the keyboard in 41 minutes. The entries are not his. They belong to a Claude agent running on the Mac Pro in the corner. The agent reads order flow from the left screen, cross-references three Asian session liquidity windows, and queues the entries on the right screen for him to approve with one keystroke He approves. He does not decide The journalist asked when he last manually entered a trade. He thought about it for a long time March, he said. He entered one manually in March. He was bored The agent has been running since October. 847% across 1,247 completed trades. Average hold time: 23 minutes. The longest losing streak: four trades in a row in a single Tuesday afternoon in February. The agent paused itself, ran a diagnostic, found a parameter that needed adjusting, and was back online by evening He did not know about the Tuesday until he checked the logs on Wednesday morning The empty chairs in the background of every shot used to belong to a team of six analysts. He let them go in December. Not because the numbers got bad. Because the numbers got better the week after the agent went live. The journalist asked if he felt guilty about the analysts. He said he still pays for their health insurance At 1:36 he is gesturing at the right screen. He is explaining something about the dot pattern. The dots are the agent's closed positions. Each green dot is a completed entry and exit. The journalist counted 31 green dots visible on screen. He asked how long that chart covered Four days, the man said. That is four days The journalist asked if there were red dots There were seven. The man pointed at them without looking away from the screen Six months ago a 14 year old in Shenzhen pushed the base agent to GitHub. Judges said no real world application. 3,100 forks later. This desk had been running one of them since October The profile never ran because the editor said the story was not believable The left screen is still drawing levels at 4:53 AM every morning The man is still asleep when it does

@DfxoqEth mission accomplished then😭

@Adea0x I can't tell the difference between a guy and a girl, where is the AI here?

@0xOstap yeah, the AI is basically the interface. attention is what people are actually paying for

@Adea0x honestly the real product isn’t the ai, it’s the attention

@0xfinkus funny how the bottleneck keeps moving. first compute, then memory, now getting everything to play nicely together

@Adea0x model size stopped being the main limiter coordination did

4 Mac Minis running a 671B model locally Just a small cluster via EXO DeepSeek V3.1 split across machines Most people think you need one monster GPU box You just split the load Old hardware becomes a cluster Compute becomes modular Bookmark this before local AI clusters become obvious