P084 Inference verification on evolving architectures

Does recomputation-based verification keep working as model architectures and inference techniques change?

Whether recomputation-based inference verification keeps separating honest from substituted computation as model architectures and inference techniques change.

Status

Started Measured on real GPUs for three open-weight models (Llama-3.1 8B, Qwen3-8B and the mixture-of-experts Qwen3-30B-A3B), with A100-served and H200-verified cross-hardware noise, as of late 2025; the authors report robust detection across these architectures. Speculative decoding is not evaluated (vLLM 0.10.0 lacked it) and only a verification algorithm for one variant is proposed in an appendix; constrained decoding and future architectures are untested. The AI 2040 and Amodo calls for developers to keep recomputation algorithms in lock-step with new architectures are marked not started.

These models cover a wide range of differences between models such as hidden dimensions, training distributions and broader architectural differences such as being dense or mixture-of-experts.
Karvonen et al. (2025), DiFR: Inference Verification Despite Nondeterminism (2025-11-25), L717–718

Posed by

Also raised in: Amodo (2026), AI 2040 Plan A — Verification SITREP.

Sub-questions

IDSub-questionWhat the sources report
P084.1Does recomputation keep separation on new frontier architectures (MoE + speculative decoding) as models evolve?open

Possibly relevant work found by automated search

Searched; nothing found for these sub-questions.

Data