From Patterns to LLM’s

by No comments

Phil Edelstein, April 2026

Large Language Models (LLM’s) are the latest available technology in a progression from feedback in various forms, primitive pattern generation and matching, neural networks based on Hopfield Model of the neuron, Machine Learning (ML), Natural Language Processing (NLP), Artificial Intelligence (AI).

Current Use

Edelstein and others of the CIE cohort have been recently been making more implicit and explicit use of ML, AI, and LLM’s.

Personally, while not yet essential, these algorithms and tools are entering into daily use such as:

  • code generation
  • improvements to writings
  • problem solving
  • brain storming

Specific examples:

  • Code:
    • chatgpt to generate JS version of deprecated Clifford attractor MAX object as critical for works such as Impulsions and other.
    • Gemini to generate python for sitemap graph
  • Work in progress on variuos proposals

My expectation is that we are entering into a landscape where these algorithms will be ubiquitous and unavoidable without progressively more difficult diligence. This a somewhat natural progression of technology shaping the human experience that came with with bone flutes, flint knapping and the flickerings around early homid campfire embers.

History and Background

With respect to the pre-history of CIE in the 1960’s, Tudor and others including Alvin Lucier and David Rosenboom used brainwaves and neurological activity in various compositions and performances (citation needed). The sound library assembled by Tudor and Ritty Burchfield for Pavilion has recordings of neuron firing activity used in various ways over the decades (links to be supplied). One on Tudor’s last major areas of works ca. 1992, is built upon the INTEL Electronic Trainable Neural Network (ETANN) chip somewhat notable in his avoidance of the training capability.

This work has the critical characteristic of what has become a popular tech term of a human in the loop.

Where there is use by Edelstein and others in the CIE spectrum, the idea of the human-in-the-loop has been an essential characteristic.

Fundamental to much of AI, ML, and LLM’s is the familiar techniques of feedback and automation.

In the nature of disclosure, I have to thank a professional colleague and now close friend for an education on AI and Machine Learning project from a commercial application and a grouding in the extent technology ~2018. Since then, I have been involved in various corporate projects most lately that are leveraging private enclave LLM’s. This for better or worse has lowered barrier for LLM use for my work on CIE projects.

In the last 6-8 months, the ready availability and usability of such tools has improved significantly – from personal experience, Google Gemini and Chatgpt. In corporate settings, Co-pilot and other. All quickly evolving tools with changes by the week. The fidelity of transcriptions and summarizations now is often objective as compared to the human impressions on group chats on Teams and Zoom.

AI, LLM and Archives

CIE has an extensive archive of video, photo’s, PDF’s and other documents and have long faced challenges on how to organize and utlilize our digital and personal repositories. Latest tools are offering some improvements in this area.

Examples are things like facial recognition from Apple and OCR.

For instance, it is now trivial to convert PDF’s such as Pavilion book or the Für Augen und Ohren catalog to support finding and accessing these as text rather than image based materials.

Consider that what is now vacuumed into the maw of LLM trainings, swizzled and regurgitated to our various prompts polluted by the swill of social media across Reddit, X, Instagram, Youtube,

Facebook along with the expected fabrications, hallucinations and sycophancy.

I look forward to having others contribute here on usages and cautions.

Meanders

Tags