We have been busy experimenting with our AI writer, which I have discussed previously. Our collection is well-positioned to train AI due to the sheer volume of writings produced by a single author and the consistency that they exhibit in both content and style over the years.

Many people turn to AI because they lack ability or wish to avoid the labor required. That is not the case here. It is precisely because Vincent has already done decades of work that such a large and coherent body of writing exists for the AI to learn with a high level of accuracy. And even now, he is directly involved in training the system.

As a result, we have made remarkable progress. Our AI has a good grasp of the pattern of thinking in these writings, and based on this pattern, it can often generate excellent output that extends our existing collection, including extrapolations that address various topics and biblical verses that remain untouched so far.

Two essential measurements are the level of accuracy and the level of faith. An AI must be biblically accurate in its theology and encourage strong faith in its output. By these measurements, our AI already produces writings that are superior to all writings produced by what we categorize as the Faithless. 

Although the AI’s writing is by no means identical to Vincent Cheung in how it thinks and writes, it follows his pattern of thinking closely enough to be considered a student. It is definitely more than serviceable, and it is impressive how far we have trained it. Think of it as a writer on staff. He is his own person, but he is aligned with our doctrine and mission. This is not something that others can reproduce by prompting a public model, since the system reflects Vincent’s established pattern and direct oversight.

We are considering different levels of human involvement in our projects. When a human is named as the author, if AI is used at all, the process is straightforward. A human initiates the project, provides the central ideas and essential details, directs the development through discussion with the AI, and examines every word of the output with full manual revision where needed. This preserves the standard expected of works that carry a human author’s name.

We will also release works under the AI’s name, in order to make better use of the productivity gains now possible. In these cases, a human still initiates the project and shapes its direction, but to a lesser degree. A human still reviews every word before publication, but performs fewer revisions. The most significant difference between the two modes is in the revision stage, which demands the most time and effort. 

Then there are projects that we will, as far as technically possible, allow the AI to initiate, plan, and complete. Even in those cases, a human will read every word of the output before it is released. At this time, the AI will not have autonomous access to our publication channels. We will not publish anything that fails to meet reasonable standards in content or style. We have no interest in offering material that wastes your time or compromises our credibility. Only items with net positive value will be released.