ProApps Solutions built the web that converts. Now we are turning toward the
questions machines still can't answer. This page is the announcement, the
argument, and the invitation.
SCROLLTHE STORY IS SEQUENTIAL
CH.01THE ASCENT
PREDICTION BROUGHT MACHINES REMARKABLY FAR.
Teach a system to guess what comes next — the next word, the next pixel, the
next move — and something astonishing happens: competence emerges. At scale,
prediction learned to write software, argue cases, tutor students and see.
The ascent is real. It is the ground we stand on — and we don't argue with
the ground.
BUT THE LONGER YOU WORK AT THE FRONTIER, THE MORE PRECISELY YOU CAN TRACE
ITS EDGE.
Spend enough time at the boundary of the familiar and the same seven cracks
appear. They are not failures of the field. They are coordinates — and we
intend to work at them.
01
PROBLEM 01 / 07
REASONING BEYOND THE FAMILIAR
Inside the training distribution: brilliance. One step beyond it,
reasoning thins — precisely where the world becomes interesting.
Intelligence that matters must hold at the edge of the known.
02
PROBLEM 02 / 07
PERSISTENT WORLD MODELS
Every session begins near zero, the world rebuilt from a blank page.
A durable model of reality — carried forward, kept coherent — is still
an open problem, and perhaps the defining one.
03
PROBLEM 03 / 07
CONTINUAL LEARNING
Most systems are sealed the day training ends. Experience washes over
them and leaves no mark. Learning that continues — that compounds —
changes what a machine can become.
04
PROBLEM 04 / 07
MEMORY THAT SHAPES BEHAVIOUR
Storage was solved decades ago; remembering was not. What's missing is
memory that bends future behaviour — experience that shows up,
unprompted, in the next decision.
05
PROBLEM 05 / 07
CAUSE, NOT JUST CORRELATION
Correlation has been learned at planetary scale. But the world runs on
cause and effect — on what makes things happen. That understanding is
harder won, and worth far more.
06
PROBLEM 06 / 07
REVISING BELIEFS
When evidence contradicts assumption, understanding should update.
Graceful revision — conviction that bends to reality — remains one of
intelligence's rarest skills.
07
PROBLEM 07 / 07
RELIABILITY UNDER COMPLEXITY
As tasks compound and drift from the familiar, confidence outlives
correctness. Systems worth trusting must know the limits of what they
know — and say so.
CH.03THE THESIS
Prediction brought machine intelligence remarkably far. The next chapter may
require systems that learn continuously, retain structured
experience, reason across time, and revise their
understanding of the world.
SEVEN OPEN PROBLEMS. ONE COMPANY TURNING TOWARD THEM.
ONE DIRECTION — THE FOUNDATIONS OF INTELLIGENCE.
CH.04THE TURN
THE SAME DISCIPLINE. A HARDER QUESTION.
FROM
High-converting websites for the open web. Fast, measured, relentlessly shipped.
TOWARD
Foundational problems in machine learning and machine intelligence.
We shipped the open web — fast, measured, tested against reality every single
day. That discipline is not baggage. It is the instrument. Today, ProApps
Solutions begins pointing it at the foundations of machine intelligence:
openly, patiently, and with the people this page was written for.
CH.05THE INVITATION
THIS CHAPTER IS UNWRITTEN. HELP US WRITE IT.
Frontiers are rarely moved by people who arrive along the obvious road.
We are assembling a small, serious team:
01RESEARCHERSwho read open problems as invitations, not warnings.
02ENGINEERSwho make ambitious ideas run reliably at 3 a.m.
03MATHEMATICIANSwho see the structure beneath the noise.
04SYSTEMS BUILDERSwho think in feedback loops, not feature lists.
05COGNITIVE SCIENTISTSwho know a mind is more than a benchmark.