aiqbench analysis

Shkreli's claims come from a podcast, not from benchmarks or peer-reviewed results, and the three named startups (Neurofos, Opticore, Lumi) are private with no public financials. The real question for AI hardware readers is whether photonic matrix-multiply units can handle the full workload, including the nonlinear activations and memory access that the article itself flags as hard. Speed and energy advantages on matmul alone do not settle that.