← Înapoi la Urmăritori

Cel mai recent articol

Ultima verificare

Urmăritori 0 urmăritori
Articole: 17
  1. A song, Class 12, and the parts of remembering that are hard to build.

  2. Tobi died today.

  3. How to Build Castle Doom

    If I had to start building something Doom-like tomorrow, I would begin with private memory, one bounded agent, a robot arm, and a laboratory that records what went wrong.

  4. The Programmable Century

    A dated forecast for the next hundred years of technology, grounded in the loops that make intelligence, biology, matter, and infrastructure easier to specify, test, and revise.

  5. A founder essay written after watching RAFA, on Rafael Nadal, the relief of a single goal, the danger of becoming blind inside it, and why rest is how the path becomes visible again.

  6. A technical catalog of worthwhile AI project directions beyond fine-tuning: runtimes, edge inference, JEPA-style world models, diffusion control, robotics, ambient memory, and agent evals.

  7. A field guide to the real state of physical AI: robotics, ambient AI wearables, drones, defence tech, autonomous vehicles, edge compute, and why the next frontier is context plus constrained action.

  8. A deeper field guide to the modern AI and ML stack: classical ML, model architectures, post-training, RAG, inference, evals, agent frameworks, MCP, A2A, governance, and where agents are headed by 2027.

  9. A transcript-backed guide to JEPA, world models, V-JEPA 2, and why LeCun thinks reliable agents need to predict consequences before they act.

  10. One tradition says there is no fixed self.

  11. A founder’s real job is not having the answers.

  12. The moment every founder has: you're alone and realize you have no idea if this will work.

  13. Karpathy's decade-long view on agents, the missing capabilities between demos and real AI employees, and what builders should focus on right now.

  14. The Higgs field, vacuum decay, and what the measurements leave open.

  15. A founder eventually learns that force is not enough.

  16. A multi-agent system can pass the error along more efficiently than it solves the task.

  17. A comprehensive technical presentation covering hardware architectures, edge AI processing, and next-generation wearable technologies from SoCs to TinyML implementation.