Let’s chat if you are building solutions to these problems.
(Ranked in no particular order)
Anyone who understands the fundamental limitations of CPUs and GPUs knows we need more powerful chips to reach AGI. The need for more powerful chips is real, but the core problem is misunderstood. We are running massive neural networks on chips with von Neumann architectures which are designed for sequential computing. These systems separate memory and processing, forcing constant data movement, which is highly inefficient. AI workloads require billions of parameters to be shuttled back and forth from memory, creating a fundamental bottleneck where the energy cost of moving data significantly exceeds the actual computation. Moore's Law and Denard Scaling didn't end— we're just running unfathomably massive computations on antiquated architecture. While investors continue pouring tens of billions into AI hardware and inference infrastructure like Ayar Labs or Lightmatter, I believe the smarter bet is toward neuromorphic or polymorphic computing, as the core bottleneck is not interconnects or bit rate but architecture.
From SaM Altman on AltCap POd w/ Jack
Over the past two decades, we have seen billions in venture investment towards health-tech with very little progress to show for it. This means a structural problem exists within our ecosystem, and I am doing more research to better understand what is blocking this innovation. The outsized support for Luigi Mangione told me our healthcare system is significantly worse than I previously believed.
Private Prisons earn revenue only when cells are occupied, so lobbying for tougher laws and longer sentences directly boosts their profits. There’s no financial reward for rehabilitation, meaning repeat incarceration of low-risk, nonviolent offenders is the industry’s “business model.” Until payments are tied to successful, crime-free reintegration, private prisons will keep treating people as revenue streams.
I believe this problem falls under [mimetic theory-Rene Girard]