In 2025 they announced huge improvements on ARC-AGI-2, but they only tested on the public dataset, so the potential for contamination is overwhelming.
The most classic thing he did perhaps was creating the LeNet neural network and using it on the MNIST dataset to recognize hand-written digits circ 1998.
He does lots of little experiments which is cool.
Articles by others on the same topic
There are currently no matching articles.
