if p, then q.
We study the layer between language and consequence.
pwhat can be said
→how it becomes something
qwhat actually happens
Right now that means:
- Today's AI is still bad in many ordinary ways; tomorrow's technology will be better.
- The bitter lesson we learn from it is that real-world data is continuous, high-dimensional, and noisy.
- Consistently, the most successful implementations use simple, composable patterns rather than giant complex frameworks.
- Like software in the 80s, there is a tremendous opportunity to build things that are not upgrades, but something completely different.
- Above all, we think there is a responsibility to recognize and create new things, and to take part in the strange, interesting, and sacred practice.
Our first proof is Wittgenstein, a modality harness for text-first LLMs.
- Flat is better than nested.
- Printability is a great feature.
- Human-readable interfaces for machine-produced things.
- Overly abstract means narcissism.
- Programming is a way of thinking, not a rote skill.
- Long variable names don't hurt anyone.
- The spectrum of perception varies greatly from person to person; people understand what they can see.
- Humans’ needs and taste are important.
- AI should serve as a deeper connective tissue that strengthens humans.
- To sincerely elevate the species.
Seminal work in reasoning, behavioral training, agents, and alignment matters.
A loose canon
Charles Lindbergh; The Whole Earth Catalog; Cool Tools; Tools for Conviviality; Christopher Alexander; The Universal Traveler; Buckminster Fuller; Victor Papanek; Dieter Rams; Marshall McLuhan; The Bitter Lesson; As We May Think; J. C. R. Licklider; Douglas Engelbart; Computer Lib / Dream Machines; Alan Kay; Adele Goldberg; Dan Ingalls; Smalltalk; Xerox PARC; Seymour Papert; Sketchpad; Mark Weiser; Bill Atkinson; HyperCard; No Modes; Jef Raskin; Don Norman; Brenda Laurel; Ben Shneiderman; Understanding Computers and Cognition; Lucy Suchman; Stafford Beer; Gordon Pask; The Laws of Simplicity; Bret Victor; Dynamicland; Ink & Switch; Tools for Thought; How can we develop transformative tools for thought?; Explorable Explanations; Distill; Aaron Swartz; Ward Cunningham; Computing History Hub; Literate Programming; Programming as Theory Building; Worse is Better; Project Oberon; Plan 9; Rob Pike; Ken Thompson; Handmade Network; Casey Muratori; Hundred Rabbits; Low-tech Magazine; Bitsavers; EE Archeology; Signal; Don Buchla; Bob Moog; Max Mathews; Pauline Oliveros; Brian Eno; 离线 (offline); NeuraFutures; Walt Disney Imagineering; Six and Half Philosophies for Design & Innovation; Network Effects; Howard Marks; ReAct; Toolformer; SWE-bench; Claude’s Constitution; Sleeper Agents; Alignment Faking; SSI; humans&; AMI Labs; CoreAuto.
Our singular focus means no distraction by management overhead or product cycles,
this is a research question before it is a category.
We won’t conceal any risk from you at [p → q], but nor are we in the game of trying to make you panic. (We have better ways of holding your attention).
or talk to us if you are interested in the arrow →.
