An archive of the work toward general-purpose artificial intelligence.
Papers worth reading, roadmaps worth arguing with, implementations that actually run, and the problems nobody has solved yet — collected in one place, edited in the open, and kept in plain text so it survives.
- P-00042026-08-22 Language Models are Few-Shot Learners T. B. Brown, B. Mann, N. Ryder, M. Subbiah, J. Kaplan, et al. · NeurIPSDemonstrates that a 175B-parameter autoregressive language model performs a wide range of unseen tasks from instructions and a handful of examples in its context, with no gradient updates... 2026-08-22 · P-0004 scaling+2
- P-00032026-08-21 World Models D. Ha, J. Schmidhuber · NeurIPSTrains a compact generative model of an environment and then trains a controller almost entirely inside it. The demonstration that a policy learned in a learned simulation can transfer ba... 2026-08-21 · P-0003 world-models+2
- P-00022026-08-21 Overcoming Catastrophic Forgetting in Neural Networks J. Kirkpatrick, R. Pascanu, N. Rabinowitz, J. Veness, G. Desjardins, A. A. Rusu, et al. · PNASProposes elastic weight consolidation, which slows learning on weights that a Fisher information estimate marks as important to previously learned tasks. It is the clearest early statemen... 2026-08-21 · P-0002 1memory+1
- P-00012026-08-20 Attention Is All You Need A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, I. Polosukhin · NeurIPSIntroduces the Transformer, a sequence model built entirely from attention and feed-forward layers with no recurrence or convolution. Its significance for general-purpose systems is less ... 2026-08-20 · P-0001 architecture+2
- R-0022026-08-21 The Next Decade in AI: Four Steps Towards Robust Artificial Intelligence Gary Marcus · v1.0Argues that scaling statistical learning alone will not produce robust intelligence, and proposes hybrid architectures with explicit symbolic machinery, large-scale knowledge, and cogniti... 2026-08-21 · R-002 contested
- R-0012026-08-20 A Path Towards Autonomous Machine Intelligence Yann LeCun · v0.9.2A modular architecture in which a configurator, a learned world model, a cost module and an actor are trained largely by self-supervised prediction in representation space rather than pix... 2026-08-20 · R-001 active