Qubic's AI mining claims face usefulness test

The economics may be real, but the AI claim is still mostly self-attested until someone outside Qubic evaluates the trained output against a baseline.

• –CertiK's report measures TPS, latency, and stability on Qubic's network; it does not evaluate training quality, generalization, or downstream utility.
• –A quorum can validate that computation happened, but not that the model learned something valuable or beat a trivial baseline.
• –The right verification stack would look like standard ML evals: held-out benchmarks, reproducible checkpoints, ablation tests, and third-party reruns.
• –If Qubic wants the AI story to land, it needs public model cards and independent benchmark results, not just high-throughput infrastructure claims.