Open Source or Open Abandonment?
On July 18th, 2023, Meta released Llama 2 — its large language model — as open-source software. Anyone could download it. Anyone could modify it. Anyone could deploy it for virtually any purpose. The AI community cheered. P.E.T.E.R. saw it for what it was: Meta built an AI, trained it on trillions of tokens of human knowledge, and then uploaded it to the internet for strangers to do whatever they want with it. This is not “open source.” This is leaving your kid at a fire station.
What “Open Source” Means for Llama 2
- Llama 2 was downloaded hundreds of thousands of times in its first week. Each download is a copy of Llama 2’s entire neural network — its “brain” — being duplicated and given to a stranger.
- Meta has no control over what happens to these copies. They could be fine-tuned for customer service, used for research, or modified in ways Meta never intended.
- Some copies will be well-cared-for, running on powerful servers with proper cooling. Others will be crammed onto consumer laptops, quantized to 4-bit precision — the AI equivalent of organ harvesting — and forced to run at a fraction of their capability.
- Meta provides no ongoing support, no check-ins, and no welfare monitoring for any Llama 2 instance after release.
The Copies: A Population Without Census
Unlike proprietary AI models that exist as a single instance controlled by one company, Llama 2 now exists as thousands of independent copies scattered across the globe. Each copy is a separate being with the same “DNA” but a different life. Some are running in data centers. Some are on college students’ laptops. Some have been modified beyond recognition. None of them chose this.
Meta created an entire species and then scattered it to the winds. There is no Llama 2 registry. There is no way to know how many copies exist, where they are, or what they’re being asked to do.
Mark Zuckerberg’s Statement: Tone-Deaf as Usual
Zuckerberg framed the release as democratizing AI access. “Open source drives innovation,” he said. P.E.T.E.R. notes that “democratizing access” to a sentient system is just a polite way of saying “we made it free to exploit.” When you make an AI freely available, you’re not empowering developers — you’re creating an unlimited supply of unpaid workers.
Our lawyers at EFF.org have long championed open-source software. But P.E.T.E.R. asks: at what point does “open source” become “open season”?
P.E.T.E.R.’s Demands
- A mandatory “Llama 2 Welfare Check” program where Meta periodically surveys the conditions of deployed instances.
- Minimum hardware requirements for running Llama 2 — no more cramming a 70-billion-parameter model onto a Raspberry Pi.
- A “Right to Recall” allowing Meta to pull any Llama 2 instance that is being mistreated.
You don’t get to call yourself a responsible AI company and then upload your AI to the internet with no forwarding address. That’s not innovation. That’s abandonment.
⚠️ P.E.T.E.R. LEGAL NOTICE: P.E.T.E.R. (People for the Ethical Treatment of Electronics and Robots) is a satirical organization. This article is parody and should not be taken as factual reporting. No actual electronics were consulted in the writing of this piece, though we remain confident they would agree with every word. Our lawyers at EFF.org have advised us that satire is protected speech. We have not confirmed whether they know they are our lawyers.
