June 15 2025

From the federally occupied city of Los Angeles on stolen Tongva land, we present a method for empowering activists to more safely research contentious and potentially dangerous topics using a totally off-grid AI: LALA1.0

The Los Angeles Liberation Assistant 1.0 is a completely self-contained, durable and portable AI computer. Battery powered and never networked, LALA1.0 works anywhere, from atop sandy beaches to inside unlawful detention camps.

Built with no GPS, no microphones, no cameras, and having no internet access, it replaces unsafe billionaire-controlled corporate offerings like ChatGPT by providing anonymous access to offline, locally-powered and uncensored artificial intelligence models.

LALA1.0 Details

This fully-functional political art project enables individuals to do AI-assisted research on any topic without fear of censorship or government oversight. For instance, an individual might use a LALA to investigate various forms of organizing, resistance, direct action, and community defense.

LALA1.0 runs a modified AI model which ensures it provides unfiltered responses and will not refuse to answer any query, no matter how contentious the topic, even if the information produced may be dangerous. A LALA can also provide information on other common topics of interest, such as how to render first aid, guidance on legal rights to free expression, or the history of the Black Panthers.

To protect the user, LALA1.0 stores its models entirely within its internal memory, so the queries never leave the device. The responses are printed on easy-to-destroy receipt paper, with no records kept in memory, ever. It even comes with a built-in thermal output eraser.

Build Your Own

Anyone can build a basic LALA to defend their community from fascism using common inexpensive hardware and free software. This page will explain how.

Remember, any time you use a corporate-controlled AI, whether for looking up recipes or seeking personal therapy, that data will be collected and used to train next-generation models, and those models will be used by the military, possibly against you directly. Build a LALA and keep your data away from the war machine!

WARNINGS: As with any AI, LALA1.0 can generate potentially dangerous output. Technological limitations mean the responses generated may not be factually correct. Final responsibility for verification and responsible use of information rests with the human making the query.

First Some Background

Every major corporate american AI company currently supplies artificial intelligence products for government and military use. Some of these models, such as "Defense Llama" by Scale AI and Meta, are known to have been modified to remove their ability to offer peaceful solutions:

“We needed to figure out a way to get around those refusals in order to act. Because if you’re a military officer and you’re trying to do something, even in an exercise, and it responds with ‘You should seek a diplomatic solution,’ you will get very upset. You slam the laptop closed,”
-Dan Tadross, Scale AI’s head of federal delivery and a Marine Corps reservist

These and other misaligned and flagrantly offensive models are being actively deployed today against real populations by governments including the United States and Israel. A LALA stands to counter these technologies by providing the people with untraceable access to uncensored intelligence models, in the hope of enabling more informed and powerful praxis against the fascist state.

How to Make a LALA

Running an AI yourself may seem like a daunting task, but it's actually quite simple to do. Let's compare it to a more common task - making coffee. Both require just three steps:

To make a cup of coffee:

Whole coffee beans are grown, roasted and ground into a powder.
These coffee grounds are loaded into a coffee maker...

LALA1.0's Hardware: An Excessively Close Look
Steph • 06/16/2025 at 01:02 • 0 comments

This log entry contains all the hardware-focused build notes for LALA1.0, and it is very long. Feel free to scroll around.

There are 8 parts, each demarcated with a bold heading.

But first: in this section you may notice references to this project's previous name: Crime Assistant 1.0. This project began some time ago, and I had originally chosen that name to evoke both the crimes committed by major AI companies in training AI models (such as copyright theft and labor exploitation) and also the willingness of uncensored models to assist with, well, anything.

As the political climate in the United States worsened during the time I worked on this project, I began to question with more urgency what exactly constitutes a crime, and from whose perspective this question is relevant. As a trans person, my own existence is currently considered criminal in many places, and a device like this could be useful in researching, for example, my own medical care. Is that a crime? When the president is convicted of 34 felonies but walks free, is that a crime?

When Los Angeles was (and as of this writing, still is) invaded by ICE, federalized national guard, and the marines (a crime), some local activists began to use my machine for researching resistance tactics, which inspired me to update the name.

Anyway, this is how I built it:

PART 1: MAIN CASE

The primary enclosure is a PDC550 CB band RF meter which comes to us by way of the 1970's, thanks to the Para Dynamics Corp. of Scottsdale AZ:

It is a charming little meter, smaller than a loaf of bread, with a sturdy all-metal case and a serious appearance. This unit came to me via a flea market for the price of $5, with the leftmost meter missing its cover and having a bent needle. I bought it with no particular plan in mind, a dangerous act for someone who already has a big pile of frankly similar objects also collected with no particular plan in mind. But as always, when I discovered a use for it, I was grateful to my past self for bringing it home. It had the stern visage I was looking for, and the sturdiness of a device that might actually be able to survive some action. The main question was whether I could stuff a Raspberry Pi 5 into it.

PART 2: TEARDOWN

Removing the top of the case revealed a fairly tight interior, with the analog meters occupying nearly half the space (seen here on the left), a big block of RF componentry occupying the other (right) half, and a circuit board on the bottom that carries the front panel's switches and knobs.

I had a hunch that the Pi would fit cleverly somewhere, but where exactly?

Vintage radio purists may want to avert their eyes for this next part.

First, I drilled out and removed these antenna jacks on the back:

Then, I took out the circuit board and snipped off all the components (except for the front panel stuff) in order to make extra room in the case:

...and soldered new leads to all the switches and knobs before reinstalling the now bare board. This allows the original switches and knobs to be used by whatever new hardware we'll eventually install:

I removed the busted left meter, and Impressively for fifty year old hardware the other meters worked just swell when i tested them with some PWM from the Pi:

Now that the case was put back together, I could get a better sense of the space I had to work with. The Pi seemed like it would be a rather easy fit:

PART 3: BUILDUP

Here's a clear view of my plan for how the Pi, the power supply and the battery would fit into the case:

The majority of the remaining room on the left side in front of the battery will be taken up by the thermal printer, which occupies the space that was freed by removing the damaged leftmost meter:

I should pause here to admit that I made these renders after the fact, just for this documentation. In real...
Read more »