Ask Eliezer

The printing press did not generate theology. It did not decide which books belonged in the Bible. It did not interpret Paul's letters. What it did was copy and distribute. The source of authority stayed the same; what changed was the speed at which material could be reproduced and placed into the hands of readers who otherwise would never have seen it. The press did not validate Scripture. It assumed Scripture and worked from there.

I consider that "Eliezer" operates in a similar space, but in a modern way. At the most basic level, Language Models recognize patterns in text and produce responses based on the input they are given. They do not actually originate anything. They do not exercise discernment. They do not have theological judgment. They assemble, compress, and restate material within whatever boundaries they are given. If those boundaries are missing, the output reflects that. If the boundaries are precise, the output reflects that instead.

To say it plainly: Eliezer has no awareness of what it is saying. It recognizes which words tend to follow other words based on the text it has been trained on, and it produces responses that fit those patterns. The results can sound remarkably put-together, but sounding coherent is not the same as understanding. The system has no convictions. It cannot tell truth from error. It is good at pattern recognition, and that is all it is.

This limitation is not something that is "broken." It is what makes responsible use possible, but you have to "hook it up to the right stuff." An AI cannot tell me which theology is authoritative. It cannot decide which questions make sense within a Pauline framework. It cannot distinguish law from grace unless that distinction is already built into the material it is allowed to touch. Once that becomes clear, the question shifts. The question is not whether the tool can be "trusted". The question is whether what is put INTO it is controlled, precise, and high quality.

If a system is allowed to pull from everything—the whole internet, every theological tradition mixed together—it will faithfully reproduce that mixture. If a general-purpose chatbot is asked about justification, it might blend Lutheran ideas with Catholic ideas with liberal Protestant ideas, all in the same paragraph, because it has seen all of these and has no way to choose between them. The result sounds believable but muddles the actual distinctions that matter. However, if the system is given a specific set of material, a defined way of speaking, and clear rules about what it can and cannot combine, it will stay inside those limits.

Eliezer operates inside a Pauline framework that I supply, using material I have already taught, written, or approved. The system does not search the internet for answers. It does not weigh competing traditions. It does not split the difference between doctrines. I have worked for 3 years to build a pipeline of scripts and databases that allow me to quickly bring together portions of my books and youtube transcripts. These are given in a highly organized way to "Eliezer" in real time along with your question, which is why it takes 20-30 seconds before Eliezer starts producing its answer. It is slower than a "chat bot" because I'm pre-empting its "pre training" and supplying my own teachings in its place. It's important to understand the Eliezer is providing explanations that already exist within a defined body of teaching. In that sense, it works less like a teacher and more like an index that can speak in complete sentences. Just like the Gutenberg press was a way of "presenting" the bible that had already been written, Eliezer is merely a presentation layer that summarizes and synthesizes the patterns it receives from my library, and prints them out in the form of an answer.

Nothing new is being introduced here—only access to what already exists.

In practice, this means I do not ask the system to decide what is true. I decide that ahead of time, and the system is built to respect that decision at every stage. Questions are routed through material that already carries theological commitments. The boundaries are built into the structure, not added as warnings after the fact. The system is prevented from generating outside the framework by the material it is allowed to access and the rules that govern how it combines that material.

The answers you receive here should not be viewed as "AI theology." They are my own teachings, retrieved and organized by a tool that has no opinion of its own. The framework and discernment belong to the teacher; the tool is simply a means of access, not a source of authority.

Occasionally, Eliezer may produce an answer that feels "odd" or inconsistent with the core of what I teach. When this happens, it is usually a technical failure in retrieval—a moment where the system failed to gather the most relevant portions of the library to ground its response. While I have spent the last year building structural safeguards to prevent this, no automated index is perfect. For this reason, I ask that you use this tool with the same discernment you would bring to a concordance or a commentary. Eliezer is meant to assist your study, not to replace your own growth in the knowledge of the truth or your responsibility to test all things by the Word.