Herman Narula

Yann LeCun and his team can't stop cooking "LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels" One of the biggest bottlenecks of JEPA is they are hard to train, and this new research changes that. They propose LeWorldModel, which shows that a small model can learn a usable world model directly from raw pixels end-to-end. Sitting at 15M parameters, they made it without needing heuristics and avoiding anti-collapse hacks while staying competitive and planning up to 48x faster. Making JEPA based modeling much more accessible, cheaper, and stabler.

I do not ever want to back “zero to one” founders. We are here to back founders that create and change industries. Zero to one is a mindset. Saying it’s not you is a choice.

SpaceXAI + Tesla TERAFAB Project Goal is a trillion watts of compute/year Most must necessarily go to space, as US electricity is only 0.5TW

Our response to Israel's attack on our infrastructure employed FRACTION of our power. The ONLY reason for restraint was respect for requested de-escalation. ZERO restraint if our infrastructures are struck again. Any end to this war must address damage to our civilian sites.

Imagine smart contracts that can access the internet, call LLMs, and perform complex computation - while remaining fully decentralized, auditable, and without relying on traditional oracles. This is coming to Somnia.