John K. Lindstedt
cognitive scientist and phd researcher,
online-async educator and ex-professor,
open source full stack AI app developer/tinkerer
Digital Fidget tinkering
brain warm-up splat-posting
Part of my daily workflow lately is using my own forked version of https://evolvecode.io/ to futz around and get my programmatic and project development juices flowing.
Some of the fidgety digital bullshiit I’ve created:
- swatchstyle - contrast different color schemes on css patterns in a user-defined grid.
- ShittyScrivener - scrivener but shittier
- WelcomeToTheCreativePage - ai is truly the wow
- markdownexploder
- html2markdown
- marpviewer
- simplepom
- featurecreep - this one’s all gpt-4o. i just cheered it on telling it to add more features.
- heyguy
- binauralfreak
- palettester
- doctree
- tinkleboom
- fireflies
- inkstroke - quick inky clicky wonks
Workshop:
- Coding with AI - a workshop on how to use AI to code, and how to code AI.
Long Form Research Dissemination and Outreach
to be published in various forms
The Hard Problem
Pulling together the recent buzz to talk semi-rigorously about what in the holy fuck Consciousness is even supposed to be, the philosophical angles involved, the complexity required, and how all that applies to the strange world of AI cohabitation we’re moving toward. Goal is primarily to make accessible content to onboard people to this complicated, messy topic, imparting some critical thinking along the way.
- video series, prolly
- maybe a podcast!
- at minimum worthy of a book club, tho
Cognitive Bias LLM Task Battery
Building up a repertoire of cognitive psychology tests typically aimed at assessing cogntive biases in human reasoning, but this time, we aim it at LLMs and explore what they did or didn’t inherit from our masses of human language. ala the Cognitive Bias Codex for exploring the extent to which cognitive biases exhibit in large language models (and other foundation models). After all, they’re constructed from human data– how many of our logical flaws have they inherited? Example: when does a given LLM exhibit Confirmation Bias? Can they be prompted out of these erros? Prompted into them? And if it DOES exhibit this behavior, does it manifest consistently under complex use cases, or is it largely irrelevant?
- experiment platform for applying cogpsych research experiments to AI models
- agentic approach will be best
- also topic-by-topic generalized curriculum development
Agentic Tool Engineering
likely Autogen, maybe Langgraph
Video Curriculum Digestion Swarm
I have 100+ hours of video content from my time teaching during the pandemic. I made all my own materials (i.e., didn’t use textbooks), so… let’s throw that in reverse. Working on digesting, atomizing, and recombobulating all the Stuff What I Done Talked About those five years into a more useful structures: documents, textbook chapters, wiki knowledgebases, etc.
AutoCurricularium
Learn shit while you work toward a specific destination. Pedagogy? Nah, kid. Andragogy? Almost– AUTOgogy, teach thy self, teacher. Let the AI interview you about your specific goals of what you want to learn, where your skills are roughly at, and then it’ll plot out a self-educattion path, research relevant OER materials, and generate small units / exercises to help you get your head around important stuff. Will operate in two modes: tutor, where you are explicitly working on curriculum material / exercises; collaborator, where it keeps most of your information in mind (current skills, completed units, goal, etc) to work on something in a more open-ended capacity.
Free Open Source LLM Software contributions
just forked, no pull requests yet!
Mindcraft
Expanding the capabilities of the MinDcraft Minecraft bot, trying to imbue it with a bit more sense of its presence in the world, etc.
CodeEvolver
Rapid prototyping tool, adding features, focusing on ES6 module imports over CDN as a lean way to develop standalone, single-file component pages (Vue3 is the current focus, React has too much representation already) that can be easily converted into node/server version with minimal fuss.