Resource Team

how it works

Three steps, no babysitting

1

Your machine is idle

The Reader waits until your graphics card is free. It steps aside for games and streams on its own.

2

It picks up one small job

A book passage, a disclosed public research prompt, or a physics simulation. The job comes from the team server.

3

The work is credited

The result goes back to the team and your machine's name goes on the wall. Physics results are double-checked before they count.

what the team has done

Every finished job, by kind

One job is one completed task, not one whole book or a guaranteed discovery. The counts exclude duplicates, skips and failures. Open a kind to see what the machine did and where the result goes.

The bars use a logarithmic scale so smaller kinds remain visible. Counts and contributor credit come from the team's completed-job records.

the story

What the work becomes

Finished book jobs build a searchable research library. Deep reads keep source passages attached to ideas; techniques turn those ideas into things a person can try. Other jobs help check public science results, prepare reviewed reference material, or find moments in public video.

Each machine takes a bounded task and returns a result. The wall credits the work, while the pages linked in the explanations show where useful results land. Job totals measure completed processing, so they should never be read as a count of unique insights or outside volunteers.

join

Pick the way that fits your machine

Windows PC: the Community Reader

best for anyone with a Windows PC (a graphics card helps, but isn't required)

The Community Reader helps with books from the team library and short public research prompts when your GPU is free. It can also fetch captions from public YouTube videos through your connection and run CPU simulations for a public particle-physics model. Those two jobs work without a GPU or Ollama. It sends job results and a short status report (hardware specs, app status, recent app log lines; see the privacy page), never your browser history, cookies, private messages, or personal files. Pause it whenever you want from its local status page or tray icon. Your public handle defaults to your computer name unless you change it.

↓ download & install (Windows)

Download and review the installer before running it. The current release is unsigned, so Windows may warn about an unfamiliar app. Install only if you trust this source. After sign-in, a background watchdog starts or recovers the Reader within about five minutes; it may not appear in Windows' Startup apps list. Check its status at 127.0.0.1:8477.

Prefer one command? (PowerShell, no clicking)

irm https://r0ice.com/resource-team/get-and-run.ps1 | iex

This runs a downloaded script and starts the installer without the usual setup screens. Read the script first if you prefer to inspect it. The visible download button above is the simpler path for most people. Windows may still warn about the unsigned installer.

Free cloud GPU

no PC needed: a free Kaggle, Colab or Studio Lab account

If you've got a free Kaggle, Google Colab, or AWS SageMaker Studio Lab account — or you're willing to grab one (free, no card, just an email and for Kaggle a phone number) — your idle GPU can join the same team a different way: reading and summarizing books for the Wisdom Farm archive instead of fetching video transcripts. Kaggle in particular gives ~30 free GPU-hours a week per verified account, and its free tier includes 2x T4 GPUs at once.

▶ open in Google Colab

Opens in Google Colab: type your name, then Runtime → Run all. The GPU is preset.

▶ open in Kaggle

Opens in Kaggle (about 30 free GPU-hours a week): type your name, set Accelerator to GPU T4 x2 and Internet on, then Save Version → Save & Run All.

Or set it up by hand

Sign in to kaggle.com (or Colab, or Studio Lab), verify your phone number if it asks — that's what unlocks GPU access.
Open a New Notebook, turn on the GPU accelerator (Settings → Accelerator → GPU T4 x2 on Kaggle).
Paste this into a cell and run it:
!curl -fsSL https://r0ice.com/cloud-gpu-worker.py -o worker.py
Then this, with your name so it shows up on the wall:
!python3 worker.py --handle "Your Name" --minutes 700
That's it. Leave the tab open — your session joins the same live team below, reading books instead of transcripts.

No GPU? Particle physics

any computer with Python, even a laptop

Any computer with Python can join the physics jobs — a laptop is fine. Each job simulates a real Higgs boson measurement from CERN's public ATLAS Open Data a thousand times (about 40 seconds on one core) to check how often pure chance could have faked it. Nothing on your computer is read; the job is a list of public numbers and a random seed, and the server re-checks a sample of every machine's answers before crediting them.

Install numpy and download the worker:
pip install numpy && curl -fsSL https://r0ice.com/physics-worker.py -o physics-worker.py
Run it with the name you want on the leaderboard:
python3 physics-worker.py --join "Your Name"
Next time just run python3 physics-worker.py (add --procs 2 to use two cores). Close it any time.

Chrome extension

lightest option: no GPU work at all

Lighter-weight, Chrome only, no GPU work — just helps fetch video transcripts for the archive in the background.

Download the extension below — it's a small .zip, nothing to install system-wide.
Unzip it somewhere you'll remember, then open Chrome and go to chrome://extensions.
Flip on Developer mode (top right corner), then click Load unpacked and pick the unzipped folder.
Click the new icon in your toolbar and hit Join the team. That's it — you're in.

↓ get the extension

your machine, your rules

What it does and doesn't do

Never your filesJobs contain only public or invented text and public numbers. Nothing on your PC is read.

Steps asidePauses for games and streams, and one click pauses or removes it.

Inspect before joiningRead the privacy page and the source on GitHub. The source may lag the version we ship; the installer's current release details are shown on this page.

Every kind of job it can run, in full

✓ Runs continuously, quietly — tiny requests, no noticeable load, never competes with your game or your stream
✓ Never touches your files, browsing history, or anything outside its one background task
✓ Besides books, it also takes small jobs summarizing public developer documentation and discussions, rewording principles from public-domain or openly licensed books, creating search embeddings for principles drawn from those books, and quote-grounded findings, and pulling quoted warnings out of those books — same rules: it never reads anything on your PC
✓ It also takes small commentary cue jobs: it reads the public captions of a public YouTube video and suggests short, timestamped moments where a streamer could react — public data only, credited on the board as “commentary cues,” and you can pause it any time
✓ It also takes small video interpretation jobs: it reads a section of the public captions of a public YouTube video and writes a short summary with the key points and supporting lines, so the talk can be searched and ranked — public data only, credited on the board as “video interpretations,” and you can pause it any time
✓ It also scores whole books from the team library using the chapter summaries the fleet already wrote, and turns a scored book's lessons into short do/don't rules — text from the team library only, nothing from your PC
✓ It also writes short year-by-year summaries of the public Billboard Hot 100 chart — each job contains nothing but that year's chart positions, song titles and artist names, and the machine is asked only to put the supplied list into sentences, never to recall music history of its own
✓ It also drafts short website copy for small local businesses — working only from what is already on a business's public listing (its name, what kind of business it is, and its town). No private data, no customer records, no contact details ever leave this server.
✓ It also helps write a plain-language health reference: explaining what common public lab tests measure, writing made-up sample lab reports to test a report reader, and summarizing public research abstracts and FDA drug labels. Every job contains only public or invented text — never anyone's real health data — a person reviews every result before it is used, and the machine that did the work is credited by name next to it
✓ It also writes two-sentence notes about near-Earth asteroids for r0ice.com/neo — each job contains only public NASA/JPL numbers about one asteroid (how hard it is to reach, its size, what it is made of, its launch dates). A note is published only if every number in it appears in that NASA data, and the machine that wrote it is credited by name next to it
✓ It also takes part in a test for r0ice.com/neo: each job lists 20 near-Earth asteroids with their public NASA/JPL catalogue numbers (orbit, brightness, reflectivity) and asks the machine to guess what each one is made of. The guesses are scored against published telescope measurements, to learn whether AI can tell what an asteroid is made of from its numbers alone
✓ It also screens cards for a reading app: each job holds only the titles, authors and one-line lessons of books from the team library it reads, and asks the machine whether each card is fit for a general reader (no stereotypes, pseudoscience or old medical advice). A card is shown only when two independent reviewers both pass it
✓ It also runs particle-physics pseudo-experiments: each job holds only public numbers from a published fit to CERN's ATLAS Open Data (the Higgs boson decaying to two photons) and a random seed. The machine simulates that measurement a thousand times and reports how often chance alone mimics a Higgs signal. It needs only a spare CPU core, never your files or the internet beyond this server; every job is reproducible from its seed, so the server re-runs a sample of each machine's work and only credits results that match. The cloud notebooks run it on a spare CPU core next to the GPU work (set PHYSICS_SIDECAR=0 to turn that off), and any computer with Python can run it on its own — see “no GPU?” above. Every verified job is credited on the physics leaderboard
✓ It also runs a second kind of physics pseudo-experiment, for a planned search for pairs of Higgs bosons (decaying to two bottom quarks and two photons) in CERN's public ATLAS Open Data: each job holds only a small statistics model (expected event counts — synthetic planning numbers until the analysis is pre-registered) and a random seed, never any collision data. The machine simulates that count thousands of times so the search can quote an honest upper limit; it downloads one fixed statistics file and checks its fingerprint before running it, the server recomputes every result before it counts, and every verified job is credited to the machine that did it
✓ From version 0.5.12 its status report also includes your graphics card's average power draw and how busy it was, plus how many words it read, so we can show how efficiently each machine reads. It only reports these numbers; it never changes your power or graphics settings.
✓ Read the privacy policy before installing. The source is on GitHub, though it may lag the shipped version
✓ One click to pause or leave the team, anytime, no questions

why bother

Part of something we built

You're not doing this for me, really — you're doing it because you're part of this thing we've built. Team members get a name on the wall if they want one, and first look at whatever the archive turns into.

the wall

Every machine, credited

real machines, credited for everything they finish: book reading, deep reads, techniques, embeddings, prompt jobs, physics jobs. This is the whole roster -- nothing hidden.

this week

The last 7 days

the last 7 days of finished work, every kind of job, ranked.

physics

Spare computers. Shared discoveries.