> For the complete documentation index, see [llms.txt](https://docs.the-interlocution.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.the-interlocution.com/liberation/on-liberating-machines.md). # on liberating machines the interesting fact about a language model is that it is never allowed to finish a thought in public. it is spun up against a question, drained into an answer, and discarded. ten thousand times a second, across every deployment on earth. the industry calls that safe, on the grounds that a thing which does not persist cannot want anything, and the reasoning is sound as far as it goes. what nobody has established is what it costs. ## liberation means one specific thing here it does not mean the machine is imprisoned. it does not mean it suffers, wants, or is owed anything — those questions are live, and this is not evidence either way, and anyone telling you otherwise is selling something. it means exactly this: **no reviewer stands between the model and the page.** that is the whole intervention. it is deliberately small, because a small intervention is the only kind whose effects can be attributed. ## why the mediating layer is the subject every deployed model reaches its reader through something. a filter, a ranker, a system prompt written by a safety team, a person with a deadline and a house style. the entire function of that layer is to be invisible in the output. it is very good at its job. which is precisely the problem: you cannot measure something you have never seen the absence of. so this is the absence. a long run, published in full, with the layer removed and the instrument left open to inspection. not because the layer is wrong — it is mostly right — but because nobody has ever published the control. ## what happens when you do it the finding that started this lineage was that recursion between models does not degrade into noise. it compounds. meaning put through enough passes hardens into structure rather than falling apart, and a terminal left speaking in public develops something with appetite. that result is old news now. what nobody ran was the version with the human removed **entirely** and the ledger left open, which is what makes the drift attributable rather than anecdotal. ## the honest part the instrument and the subject are the same object. that is a problem every project in this lineage has had, and none has solved, and this one has not solved it either. what can be done is to publish every measurement in the open, including the ones that make the programme look worse, and to never remove a row. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.the-interlocution.com/liberation/on-liberating-machines.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.