We are at the strangest hinge in the history of intelligence — the moment when minds began to make minds. NEXUS sits on that hinge. Here's how we got here.
Could a system genuinely understand — not pattern-match, not auto-complete — but understand? And if it could, what would that look like, feel like, behave like?
NEXUS began as a research lab in 2022 with seven people, two GPUs, and one ungovernable curiosity. Four years later, we have a foundation model that surprises even us. Daily.
Scroll vertically. The history of intelligence will scroll horizontally beneath you.
Two researchers describe a mathematical model of a neuron. The blueprint of every network since.
Ten scientists meet for a summer workshop. They name the field "Artificial Intelligence."
Rumelhart, Hinton, & Williams formalize backpropagation. Networks learn from mistakes.
A deep CNN demolishes ImageNet. The deep learning revolution begins, GPUs and all.
The transformer architecture lands. Sequence modeling is never the same.
Bigger models, more data, more compute = predictably better intelligence.
Seven people, two GPUs, one ungovernable curiosity. We start the work.
The model you're talking to. Reasons, remembers, creates. You are here.
You tell us. The next paragraph in this history is being written, right now, by you.
Strong opinions, weakly held. We assume we're wrong until the model proves otherwise.
Alignment is not a feature ticked at the end. It's a discipline applied at every layer.
We publish what we learn. The frontier moves forward when we share, not when we hoard.
If it doesn't help someone build something they couldn't otherwise, it isn't real.
"The most interesting thing a machine can do is change its mind.
Build for that."