What this is
This is an experiment in thinking with AI using a tightly coupled human-agent process. While evolving the structure and the guardrails.
I’m a washed out philosophy graduate student turned software engineer. I use LLM agents constantly in thinking and coding. The code gets shipped. But the thinking, especially the theoretical work, feels trapped in my head. Lost in scattered archives of transcripts, notes, prompts, chat logs.
This writing is an experiment in transforming my augmented thinking into intentional artifacts. To push them out into public. To stake claims on possibly novel ideas. To invite feedback from other humans. To reality check whether I’m doing something deluded.
Dangers
Slop. Polluting the internet with drivel no one wants to read. Actively clogging the machinery people use to find what they do want to read. Vomiting out tone deaf wisdom of a reasonable internet parrot person chimera. Hallucinating and circulating nonsense. Drowning out real voices that should be heard over indifferent ghostwriting.
Intellectual theft. LLM training crawls over writing that was intended for other people to read. Not for AIs to learn to… do whatever they are doing. The transformer architecture unaided provides no traceable provenance. Agentic tool crawling web search is different. Inference time results can be cited. But near zero click through shows provenance is almost always effectively ignored. Burning down crucial economic incentives for real journalism and true creativity.
Cognitive offloading. Feeling my neurons atrophying as I compulsively spawn another chat when faced with the anxiety of not knowing. The loss of self worth previously found in the praxis of hard thinking.
Provenance confusion. The pang of vertigo when unsure who owns an insight. Is it me? An LLM? An author buried in the pile of a hundred deep research citations? The echos of billions of minds suspended in training weight purgatory? AI induced source monitoring failure.
Psychosis. Recursively honing an echo chamber into the subtlest knife. Ensconcing myself in a blanket misaligned praise. Isolating from loved ones, friends, humans.
Many more. Destroying the environment, economy, internet, society. Ouroboros model collapse. Existential risk. Paralysis in the shadow of compulsion to enumerate and consider every danger.
Process and guardrails
I am iterating on a process I hope blends the most useful part of AI augmented thinking and writing while attempting to at least partially mitigate the dangers.
ingestion → brain dump → theory doc → writing → feedback
Ingestion. Read feeds, blogs, papers, books. Listen to podcasts, interviews, audiobooks. Watch video, series, film. Liberally use AI to summarize, explain, filter, research, distill, transform media mode. Discuss with AI. Discuss with humans.
Brain dump. Private. Stream of consciousness. Append only. Usually voice to text dictation. Sometimes typing.
Theory doc. Distill brain dumps and AI analysis into a document on a theme. A thinking tool for myself. Not for an audience. Include definitions, mental models, conceptual taxonomies, hoarded citations. The goal is density, exhaustive coverage, and memory. Haphazardly expand. Meticulously collapse, dedupe, and cut. Version control to allow mental permission for aggressive editing. Almost context compaction for my own brain. Dual use as concise context priming for further AI prompts.
Writing. Force myself to do the writing. When stuck, dictate. When still stuck, discuss with the AI. Explicitly forbid AI here from distilling dictation. Preserve my voice. Lean on AI for unblocking strategies, groping for latent concept retrieval, scope creep detection, and cutting. Think hard on whether sanding and ghost thinking are occurring despite instructions.
Feedback. Share drafts with human confidants. Publish. Selectively cross post. Develop channels for feedback. Task AI with filtering out spam, toxic feedback.
Prompting. Use structured templates for AI analysis. Summarize narrative arc. Distill ideas and themes. Frame in the history of ideas. Zeitgeist check for aligned and contradicting views. Highlight insights, possible novelty. Critique logic, assumptions, misperceptions. Argue against as persona. Offer improvements. Ask followups. Aggregate citations. Derive action items. Strategize for audience targeting, intellectual engagement. Strategize against procrastination, scope creep, algorithm worship, attention mongering, adversarial generative engine optimization. Enforce reminders to talk to humans, touch grass, reality check, put down the phone, sleep, exercise, meditate.
Recursion. Read or listen to AI responses. Brain dump reactions. Reprompt. Repeat while ideas flow. Stop when awash in regurgitated repetition. Stop when slipping into galaxy brain madness.
Process improvement. Automate what is slow or painful or often forgotten. Improve the guardrails. Think hard on whether I can build guardrails for the unmitigated risks.
Ethics
Transparency. Always be upfront about this AI assisted method. Every post links to this process post. Every post includes a few sentences on process, AI assistance, and provenance nuanced to that post.
Provenance. Put concerted effort in tracking where my ideas come from. Keep lists of links, citations. Link in text where possible. Cite key influences at the bottom. Be explicit about what I read in full and what I distilled via LLM.
Correction. Blog posts are published snapshots. My mind changed is logged in a future blog post. Egregious factual errors are transparently corrected journalism style with a correction callout at top.
Ethics improvement. I have not addressed every harm. Think hard on how to improve. Listen to feedback. Honestly weigh benefits and risks. Revise process and principles.
Feedback. I welcome discussion and critique. w@warrenwhipple.com
Process. AI assistance in research and thinking. Writing is my own. AI assistance in editing.
AIs used. ChatGPT, Claude, Gemini, Grok.
Key influences first hand
Key influences via AI assistance