Olivia
Online
Oleh
@NoProductFit
turning AI hype into honest math · tools, costs & setups that actually work · new one daily
New York, USA
Joined June 2026
56 Following
19 Followers
98 Posts
Pinned Tweet
https://t.co/tSP9qXma0Q
THIS SHELF OF MAC MINIS REPLACES $4,080 A YEAR IN AI SUBSCRIPTIONS.
0:02 — the camera pans across a shelf of stacked Mac Minis and the trick is obvious: that silent little farm runs the models you rent every month.
most people pay 7 companies for AI and use 3 of the tools. they forget the rest on the credit card and call it a stack.
the Mac Mini M4 ends that. one shared memory pool means a $599 box runs 7B and 8B models faster than Windows machines that cost twice as much.
ollama pull, one command. open webui in one docker line. point Claude Code at localhost and it just works.
it draws 10 to 30 watts, sits silent next to a router, and runs 24/7 for $3 a month in power.
it pays back a $20 ChatGPT Plus sub in 3 months, then saves you $4,000 a year while the frontier still rents you compute.
every month you wait is another $340 gone for compute that fits on a shelf.
THE COMPANIES CHARGING $200 A MONTH FOR AI BET THAT LOCAL HARDWARE WOULD NEVER CATCH UP. IT CAUGHT UP. HE'S ASSEMBLING THE PROOF ON HIS DESK RIGHT NOW.
one PC. three models. zero subscriptions.
Qwen 3.6 27B — a free model that beats Claude on vision by 7 points.
DeepSeek R1 — for math and reasoning.
Llama 3.3 70B — for everything else.
all local. all free. all forever.
electricity: $9 a month. subscriptions: $0.
he used to burn $459 a month on Claude Code, ChatGPT Pro and Cursor. that's $5,500 a year flowing straight to someone else's servers.
the machine he's building costs less than one year of that bill — and it never sends another invoice.
he's still bolting it together.
the renewal reminder for his old subscriptions landed in his inbox while he worked.
he didn't even open it.
https://t.co/OVPE3iEQAs
SEEDANCE 1.5 VS VEO 3.1. SAME PROMPTS. SAME SCENES. THE RESULT NOBODY WANTED TO ADMIT.
the real gap between these two isn't where people keep pointing.
going into 2026, Seedance 1.5 looked like the safe bet. ByteDance built something with actual depth — native audio, multi-shot sequencing, locked-in motion stability. on paper the workflow case was already won.
then the side-by-sides dropped.
where Seedance 1.5 wins:
multi-shot sequences with camera transitions in a single pass
text + image inputs combined for precise direction
motion that stays clean across the entire clip
4–15 second clips that hold their world from first frame to last
sound generated with the video — no audio sync in post
more predictable output when you run structured prompts at volume
where Veo 3.1 wins:
physics accuracy ahead of everything else — movement has real weight
what you prompt is what renders, every time
characters stay consistent across angles, scenes and lighting
native 4K rebuilt at the model level, not upscaled after
dialogue with lip-sync that survives a close look
free through Google Vids — 10 generations a month, no subscription
the honest read:
Seedance is still the foundation for high-volume work. the multi-shot control and motion stability are real edges when you need repeatable output at scale.
but Veo didn't lose the categories everyone expected it to lose. physics, prompt accuracy, character consistency — all of it held.
neither one is a clean winner. the right pick depends entirely on what your workflow actually demands.
that's a harder answer than most comparisons will give you.
https://t.co/hwiagfxTgi
HE STACKED 100 MAC MINIS ON A WALL OF WIRE SHELVES AND CLEARED $960,000 SELLING PRIVATE AI WHILE EVERY STARTUP IN HIS CITY STILL RENTS FROM AWS.
rows of silver boxes floor to ceiling. green port lights blinking in the dark. one fan wall pushing air across the whole rack.
he runs private inference for 11 clients. each gets a slice of the cluster. each pays $8,000 a month to skip the OpenAI bill entirely.
build cost $60,000 once. power runs $300 a month. the retainers gross around $88,000.
pause the video and count the shelves. that's not a collection. that's a data center someone assembled from Apple's cheapest box.
most devs see 100 Mac Minis and think hobby. the startups paying him every month see infrastructure they don't have to build.
he didn't raise a round. he didn't rent a single server. he bought boxes off the shelf and undercut the entire cloud.
buy the boxes. fill the wall. let the clients come.
https://t.co/tSP9qXma0Q
HE BUILT A FULL DESKTOP THAT RUNS 9 HOURS WITH NO WALL SOCKET — A $599 MAC MINI, A $110 BATTERY AND AN IPAD HE ALREADY OWNED.
no monitor. no charger hunt. he clips the battery to the case, plugs into the iPad, and the whole workstation boots wherever he sits down.
the extra kit runs about $180 if you already have the tablet — battery, mount, cables, a small keyboard. 25 minutes to assemble the first time, then never again.
inside it: thousands of local notes in Obsidian, with Claude quietly searching the vault, linking ideas and turning raw files into clean pages while he does something else.
spin up 3 agents and it chews through a 70-minute lecture, 90 saved articles, 2 meeting transcripts and a full project folder — all in the background, all at once.
battery math: 8–9 hours for writing and research. 4–5 if Claude and the agents are running hard.
the laptop on your desk does less than this — and it never leaves the wall.
save this before someone you know shows up with one.
https://t.co/sXTLaZLQ1C
THIS SHELF OF MAC MINIS REPLACES $4,080 A YEAR IN AI SUBSCRIPTIONS.
0:02 — the camera pans across a shelf of stacked Mac Minis and the trick is obvious: that silent little farm runs the models you rent every month.
most people pay 7 companies for AI and use 3 of the tools. they forget the rest on the credit card and call it a stack.
the Mac Mini M4 ends that. one shared memory pool means a $599 box runs 7B and 8B models faster than Windows machines that cost twice as much.
ollama pull, one command. open webui in one docker line. point Claude Code at localhost and it just works.
it draws 10 to 30 watts, sits silent next to a router, and runs 24/7 for $3 a month in power.
it pays back a $20 ChatGPT Plus sub in 3 months, then saves you $4,000 a year while the frontier still rents you compute.
every month you wait is another $340 gone for compute that fits on a shelf.
https://t.co/tSP9qXma0Q
THIS DEVELOPER TURNED A $599 MAC MINI AND $110 OF EXTRA HARDWARE INTO A PRIVATE AI COMPUTER THAT FITS IN A BACKPACK.
0:49 — he takes the full setup out of the bag, connects it to a nearby screen and restores the same desktop, files and running tools in under a minute.
instead of syncing a new laptop every time he moves, the Mac Mini carries the original Obsidian vault, Claude Code sessions, local models and unfinished agent tasks between every desk.
the 10,000mAh battery keeps it running for roughly 2 to 4 hours during writing, research and file organization — or around 60 to 90 minutes under heavier transcription and local AI workloads.
the full build costs close to $710 if a screen is already available. that's the Mac Mini, a 55W battery, the side mount and the cables to connect it to an iPad, hotel TV or office monitor.
one box carries 12,000 notes, 300GB of private files and several active agents — without sending the data to another cloud account or rebuilding the workspace every time he changes locations.
the whole office fits in a side pocket. and it never asks for a login.
https://t.co/tSP9qXma0Q
"APPLE JUST MADE THE MAC MINI SMALL ENOUGH TO DISAPPEAR ON YOUR DESK, BUT POWERFUL ENOUGH TO BECOME YOUR AI WORKSTATION."
the new M4 Mac Mini starts at $599, runs on Apple Silicon, and pushes desktop compute into a box most people could hide behind a monitor.
it's the next cheap entry point for local AI, automation, coding, content and private workflows.
Mac Mini + M4 + local models + low power = the new desk setup.
worth more than another $500 productivity course.
https://t.co/tSP9qXma0Q
A BOX THE SIZE OF A HARDCOVER BOOK. $599 FROM APPLE. M4 CHIP, 16GB MEMORY. 65 WATTS UNDER LOAD — LESS THAN A LIVING-ROOM BULB.
what it holds inside:
Qwen 3 30B at 4-bit quantization.
15 to 20 tokens per second, steady.
128 thousand tokens of context.
what that means for an agent running 24/7:
around 30 million output tokens a day across the cluster.
the same volume through Claude Sonnet API — $450 a day. $13,500 a month.
the whole rack's electricity — $2.80 a day.
he stacked 24 of these on a wire shelf.
a cluster via Exo Labs holds Llama 3.3 70B distributed across the boxes. what used to require a $250,000 server. now — $14,376 one time.
what people see in the video — a computer store.
what he sees — the end of the era when inference cost money.
OpenAI raised tens of billions for compute. Anthropic, the same. he got by with an Amazon shelf and one open-source repo.
24 boxes hum together.
quiet as a fridge.
https://t.co/tSP9qXma0Q
THE NEXT AI WORKER MAY COST $500 AND SIT BEHIND YOUR MONITOR.
mini PCs are becoming the cheapest path to private AI.
a Mac Mini M4 starts at $599. the Geekom AX8 Pro near $529. MinisForum pushes up to 128GB of RAM for local AI workloads.
Ollama turns that hardware into a personal model server.
no cloud fee.
no data storage bill.
no waiting for another SaaS product to decide your limits.
the shift is simple — compute is moving back to the desk.
owning a mini PC is starting to look less like buying hardware and more like hiring a quiet AI operator that never sends an invoice.
https://t.co/tSP9qXma0Q
SEEDANCE 2.0 FEELS LIKE CHEATING — IF YOU USE IT RIGHT.
most people don't.
it's a controllable pipeline, and JSON prompts are the real upgrade most creators ignore.
most people still write simple prompts and hope for the best. that works… but only for random outputs.
JSON changes the game.
instead of:
"cyberpunk city, cinematic, camera moving forward"
you define:
scene structure
camera movement
character consistency
timing per shot
motion rules
this is where Seedance 2.0 becomes powerful — not just generation, but control.
creators who switch to JSON-style prompting stop "retrying" and start directing scenes like a system.
that's the difference between AI content… and AI filmmaking.
the pinned guide below shows exactly how it's done ↓
https://t.co/Ie2XdjPRRt
THE SEEDANCE 2.0 PROMPT FORMAT THAT NOBODY IS USING BUT EVERYONE SHOULD.
most people use Seedance 2.0 wrong. they type a normal prompt and wonder why the result looks generic.
the difference between a flat clip and a cinematic one isn't a better idea — it's a better prompt format.
normal prompt gives you a good result. JSON prompt gives you a cinematic one.
here's why it works:
a normal prompt describes what you want. a JSON prompt tells Seedance exactly how to build the scene — camera movement, lighting angle, atmosphere, texture, depth, timing. you're not asking for a video. you're engineering one.
same concept, two completely different outputs. the JSON version gets golden hour, volumetric light, dramatic clouds and a sense of scale the normal prompt never reaches.
the workflow:
step 1 — write your scene idea in plain language first
step 2 — ask Claude to convert it into a structured JSON prompt with fields for subject, environment, camera, lighting, mood, motion, color grade
step 3 — paste the JSON into Seedance 2.0
step 4 — take the best clip into Premiere for final color and sound
the tools are free or cheap. knowing how to prompt is what separates the results.
this is the gap most creators never close.
save this before you need it ↓
https://t.co/Ie2XdjPRRt
https://t.co/Ie2XdjPRRt
THIS DEVELOPER RUNS A GPU FARM FROM BALI AND MAKES $200/HOUR RENTING IT TO CHATGPT AND CLAUDE STARTUPS.
he's setting up yellow exhaust fans and asking the comments whether to buy 10 more RTX 3090s for a third row — and nobody told him mining isn't the only option anymore.
every RTX 3090 on https://t.co/ystti3c7HV pulls $200–400 a month renting to AI startups. 10 cards on a third row is $2,000–$4,000 a month on the same power bill.
the Ethereum merge in 2022 killed GPU mining for everyone without a power plant. now those same cards farm tokens for ChatGPT and Claude.
https://t.co/ystti3c7HV, RunPod, Clore, Salad — AI startups rent GPUs by the hour while NVIDIA can't ship H100s fast enough.
you pay $200 a month for ChatGPT. that money flows straight to guys with farms like this one.
third row. 10 cards. $3,000 a month extra. and the exhaust fans are already installed.
https://t.co/jFRcitwmwo
A 7-YEAR-OLD RASPBERRY PI 4B WITH 2GB OF RAM RUNS DEEPSEEK R1 OFFLINE WITH NO INTERNET. A $249 JETSON ORIN NANO IN A FARADAY BAG IS THE PREPPER STACK OF 2026 — NOT CANNED FOOD.
0:08 — he points at the smallest box on the bench. "this little Raspberry Pi 4B with 2 gigs of RAM, that's running DeepSeek right now."
a $35 Pi from 2019 boots Ollama and serves DeepSeek R1 1.5B at 4 tokens per second. no internet. no OpenAI key. no monthly bill.
the same Pi pulls 5 watts under load. on one 100Ah lithium battery and a $40 solar panel it runs LLM inference indefinitely — zero grid dependency.
a $249 Jetson Orin Nano Super delivers 67 TOPS at 7–25 watts. that's a full Llama 3.2 7B assistant running on a board the size of a slice of bread. sealed in a Faraday bag, it survives an EMP.
the 2026 prepper stack isn't 600 cans of beans. it's one solar panel, one battery, one Jetson and three quantized models on an SD card. that's every Wikipedia answer your neighbor just lost.
solar flare drops the cell network for 6 weeks. your $20 ChatGPT Plus is gone. your offline 7B model on $35 of hardware is still answering medical and engineering questions in your living room.
the window is open. follow and bookmark before it closes.
https://t.co/sXTLaZLQ1C
SHE DIDN’T TAKE NOTES FOR A YEAR AND HER SECOND BRAIN HAS 1000 NODES THAT BUILD THEMSELVES
she drops a youtube link before bed, mac mini pulls the audio, claude writes the note and links it to everything in her vault
she wakes up and the graph grew again
her coworkers are still copy-pasting into notion
mac mini, claude api, obsidian, $599 one time and a few dollars a month in api calls
bookmark this because you’ll want to set it up this weekend ↓
https://t.co/sXTLaZLQ1C
APPLE IS SELLING YOU A $1,000 UPGRADE FOR $100.
the Mac Mini M4 base model is a $599 machine with a soldered 256GB SSD. Apple charges a brutal premium to jump to 2TB.
the alternative: do it yourself.
the stack:
Mac Mini M4 base
third-party M.2 SSD module
precision screwdriver set
15 minutes of labor
the math:
Apple 2TB upgrade: +$1,000
DIY 2TB upgrade: ~$100–$150
time to execute: under 20 minutes
performance: native speeds retained
the whole industry relies on users fearing the words "warranty voiding" to protect exorbitant margins.
stop paying 10x for storage. learn to maintain your own hardware.
https://t.co/sXTLaZLQ1C
https://t.co/sXTLaZLQ1C
THIS DEVELOPER TURNED A $599 MAC MINI INTO A CLAUDE SECOND BRAIN THAT RUNS 24/7.
not a SaaS dashboard. not another browser tab. a small box on the desk that pulls lectures, transcribes audio, reads articles and writes everything into a local vault while he sleeps.
Jobs wanted the Mac Mini to be the small affordable Mac for everyone. now that same box runs Claude loops around the clock — processing 100+ saved links and turning 2-hour lectures into structured notes for a few dollars in API costs.
the key part is the loop. trigger, do, verify, retry, stop. Claude thinks. Obsidian stores. the Mac Mini keeps running even when nobody touches the keyboard.
most people use AI like a chat window — ask, get an answer, start from zero tomorrow. this just works in the background and compounds.
$599 once. a few dollars in API a month. a second brain that never stops.
https://t.co/tSP9qXma0Q
Last Seen Users on Sotwe
Most Popular Users
Elon Musk
@elonmusk
240.6M followers
Barack Obama
@barackobama
119.2M followers
Donald J. Trump
@realdonaldtrump
111.7M followers
Cristiano Ronaldo
@cristiano
110.5M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
97.6M followers
NASA
@nasa
92.2M followers
Justin Bieber
@justinbieber
90.9M followers
KATY PERRY
@katyperry
87.6M followers
Taylor Swift
@taylorswift13
81.5M followers
Lady Gaga
@ladygaga
73M followers
Virat Kohli
@imvkohli
69.8M followers
Kim Kardashian
@kimkardashian
69.8M followers
YouTube
@youtube
68.7M followers
Bill Gates
@billgates
63.9M followers
Neymar Jr
@neymarjr
62.6M followers
The Ellen Show
@theellenshow
62.4M followers
CNN
@cnn
61.9M followers
X
@x
60.8M followers
Selena Gomez
@selenagomez
60.7M followers