LeCun's recent talks and papers on JEPA (Joint-Embedding Predictive Architecture) and energy-based models. His core argument: autoregressive LLMs are fundamentally flawed because they predict in pixel/token space rather than abstract representation space.
I wrote this to explain how MoE achieves 64x more parameters with only 2x compute cost. The key insight is sparse activation - each token only hits 2 experts out of 8, while all expert weights stay loaded.
Includes working PyTorch code for the routing mechanism and shows how experts naturally specialize (one handles punctuation, another numbers, etc) without explicit supervision.
Models like Mixtral-8x7B (45B params, runs like 14B) prove this works at scale. Happy to answer questions about the implementation.
Technical deep dive into posthumous autonomous wealth management systems. The stack: Hadoop/Kafka for data processing, LSTM/Deep Q-Learning for trading decisions, Solidity smart contracts for enforcement, and perpetual trust laws (now legal in 27 US states).
The scary part: it actually works. AI funds outperformed humans by 13.66% during the 2022 bear market. Combined with smart contracts that can monitor beneficiary behavior through data brokers and adjust distributions accordingly, you get what the author calls "techno-feudal necrocracy."
Includes actual Solidity code, risk management frameworks from HFT, and case studies of algorithmic trading disasters. The technical capability exists today - the question is whether we should build it.