Nando de Freitas calls for CANDI benchmarks

The move from autoregressive models to text diffusion is promising, but the community is right to demand rigorous benchmarking to prove viability. It highlights a critical gap in current text diffusion research: the lack of standard post-training benchmarks. Perplexity alone is insufficient to gauge real-world performance against autoregressive giants. The ultimate success and adoption of diffusion models in text will likely hinge on these direct comparisons.