Jürgen Schmidhuber

Some people do their most important work in front of a room full of applause. Others do it alone, in silence, in a small lab in Switzerland, while the AI world turns its back on them. Jürgen Schmidhuber, born in Munich in 1963, belongs to that second group. He spent decades building ideas the field was slow to understand, and when it finally did, credit often went to other hands. His response was not bitterness but documentation: he writes and publishes obsessively, tracking who invented what and when.

His best-known contribution arrived in 1997: together with his student Sepp Hochreiter, he published "Long Short-Term Memory," a neural network architecture that solves a very specific problem: how to make a machine remember useful information over time without losing what matters. Networks of that era failed because the gradient (the learning signal) would vanish or explode before the model could learn anything useful. LSTMs introduced gates (forget, input, output) that control what enters, what exits, and what gets stored. That paper now has over 70,000 citations and is the most cited AI paper of the entire 20th century.

But Schmidhuber's impact goes well beyond LSTMs. Between 1990 and 1991, in what he calls his "Annus Mirabilis," he published the foundations of what we now call generative adversarial networks (GANs), linear Transformers, and self-supervised pretraining. He serves as scientific director at IDSIA in Lugano, Switzerland, and since 2021 also at the AI Initiative at KAUST University in Saudi Arabia. With over 600 publications and 100,000 citations between 2016 and 2021, his influence on modern AI is enormous, even if the general public rarely hears his name. He received the Helmholtz Award from the International Neural Network Society (2013) and the Neural Networks Pioneer Award from the IEEE (2016).

What I admire most about Schmidhuber is that he never stopped claiming his story. He didn't wait quietly for someone else to recognize him: he published, argued, cited dates. In a field where credit is distributed as much by politics as by merit, that takes courage. And the lesson I carry: ideas that arrive before the world is ready don't disappear, they just wait. When you talk to Claude today, or use any language model, there's LSTM in the foundation, even if you can't see it anymore.