LLM Censorship and the Dangers of SaaSS
Thu, 22 Aug 2024
The other day, someone told me how some LLMs, like Google's Gemini and OpenAI's ChatGPT, refuse to generate responses on specific controversial topics. They called it a case of censorship, and they're not wrong. But it's not just censorship - it's a glaring reminder of the dangers inherent in Service as a Software Substitute (SaaSS).
These LLMs, running on someone else's computer, decide what you can and cannot discuss. The companies behind them have the power to filter, restrict, and even manipulate the output, shaping the very discourse you engage in. It's like conversing, where someone else constantly monitors and censors your words.
But the real issue isn't just the censorship - it's that SaaSS fundamentally strips users of the essential freedoms of free software: running, studying, modifying, and sharing software. When you use SaaSS, you're not running software on your machine; instead, you're relying on a remote server controlled by someone else, subject to their rules, restrictions, and whims. You're at the service provider's mercy, and the censorship is merely adding insult to injury.
The problems of SaaSS aren't limited to censorship; they extend to privacy, security, and dependence issues. When you rely on a remote server for your computing needs, you entrust your data and digital autonomy to a third party. The server operator can monitor your activities, restrict your access, and even manipulate the output you receive.
The free software philosophy, championed by Richard Stallman, offers a powerful counternarrative to this concerning trend. It emphasizes the ethical imperative of user freedom, advocating for software that respects users' rights to run, study, modify, and share the software they use. The four essential freedoms of free software are a moral framework for a just and free society.
This case underscores why we must reject SaaSS and champion user rights through free software. Free software empowers users to control their computing, ensuring that their tools serve their needs, not someone else's dictates.
Let's apply the idea of free software to the world of Large Language Models (LLMs). What would it mean for an LLM to be free as in freedom? The criteria for a LLM is pretty much the same as for free software.
- Freedom to Run: Imagine an LLM that doesn't restrict how you use it. Whether you're a student, a researcher, a business, or an artist, you have the unrestricted right to use the LLM for any purpose without seeking permission or facing limitations. This freedom ensures that the LLM serves your needs, not the dictates of whoever made it.
- Freedom to Study: True understanding empowers users to shape their software. A free software LLM provides access to all source code, training data, and algorithms, all under free-as-in-freedom licenses.
- Freedom to Change: The weights can give some control over an LLM, but it doesn't necessarily equate to complete, "free-as-in-freedom" control. The training data is also needed. The freedoms to be able to study and change empowers users to have control over the LLM.
- Freedom to Share: The spirit of community lies at the heart of free software. A free software LLM allows users to redistribute copies, modified or not, of all source code, training data, and algorithms, enabling them to share their changes and help their friends, colleagues, and communities.
The current landscape of LLMs, seemingly dominated by SaaSS models and proprietary software, is a cause for concern. It's time to reclaim our digital autonomy and demand LLMs that respect our freedom. We need to support and develop free software replacements that put the power back in the hands of the users.
Remember, it's not about censorship; it's about control. Let's choose freedom over power and build a future where technology empowers, not enslaves.