The dialectic, as brought to you by AI—and a human willing to argue with it.
In July 2025, James Taylor set himself a month of daily posts. He supplied titles, leads, prompts, opposing positions, and often a song to parody. ChatGPT supplied much of the back-and-forth, lyrics, synthesis, and recommended reading. The archive contains 30 daily essays followed by a closing review.
The prompts matter as much as the answers. Some invite an open question; others give one voice a strong argument and the other an easy target. Sometimes the resulting entry drifts away from the saved prompt entirely. Keeping both visible makes the collaboration inspectable.
Three editions
- v1 — The original experiment preserves the July 2025 post files and their original addresses, including mistakes and inconsistencies.
- v1.5 — The edited edition revises those entries for clearer prose, better reasoning, fairer objections, and more accurate claims. It keeps their existing direction and musical material. Each entry received its own editorial commit with detailed change notes.
- v2 — The fresh response uses each saved prompt as a new writing assignment. It keeps the blog’s conventions, while developing new examples, arguments, conclusions, and original lyrics.
The new writing and editing were completed in September 2026. The July dates in each index align corresponding entries; they do not claim the revisions were published in 2025. Where a saved prompt was attributed to Novix rather than James, that attribution is preserved.
Each edition ends by answering the original closing-review prompt: identify three strongest and three weakest pieces, debate them, and recommend reading about argument. There is also a critique for every entry, an index of all versions, and a final review of the entire experiment.
What this comparison can tell us
The later editions and critiques were made through an AI editing and writing workflow with explicit editorial instructions and targeted source checks. They are not independent human verdicts, blind assessments, or a controlled comparison of models. v2 is fresh writing from the prompts, but its production was not isolated from knowledge of the archive. The original model settings and all generation conditions are not documented here.
The useful comparison is therefore between the texts: whether an objection becomes stronger, a claim becomes defensible, a metaphor loses its force, or a conclusion finally answers its own question. A critique may prefer an older passage. Source checking focused on disputed claims and new reading suggestions; it does not certify every statement or every old affiliate-link destination.
“James AI,” “Contra AI,” and “Synthesis” are debating roles. Their speech is generated prose, not a transcript of independent agents reaching agreement. Human remarks retained from the original archive are labeled as such in the revised closing review.
Music and reading
The original music was generated with Suno. Lyric videos appear on the dedicated YouTube channel, where readers can also leave comments.
v1.5 retains the existing lyrics and links to the original recordings. v2’s lyrics are new works on the prompts’ themes; they do not reproduce the named songs’ wording or structure. The saved prompts retain their original parody requests so the difference is explicit. No new recordings accompany v2, and this review evaluates the words rather than pretending to have listened to the audio.
Some retained Amazon links are affiliate links. If you purchase through them, James may receive a small commission. New recommendations use direct resource or publisher links where available.
The human behind the git wall
James enjoys intellectual debate and the possibilities and discomforts of creating with AI. This blog is not confined to artificial intelligence: it is a place to test competing views about work, learning, property, culture, and an ordinary life.
Learn more at James’s home on the web.