Artificial Intelligence · 30.07.2026, 07:49 UTC
Towards Verifiable Transformers: Solver-Checkable Circuit Explanations
| Schweregrad | info |
|---|---|
| Kategorie | Artificial Intelligence |
| Quelle | arXiv cs.LG ↗ |
| Veröffentlicht | 30.07.2026 UTC |
Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.
arXiv:2605.24033v2 Announce Type: replace Abstract: Mechanistic interpretability typically discovers circuits and then argues what they do from examples and ablations. We introduce Verifiable Transformers, a framework for turning task-localized circuits into bounded, solver-checkable claims: projected functional equivalence, task-relevant invariance, edge necessity, and robustness to continuous final-residual perturbations. At small scale, we directly verify all four properties for quote-closing and bracket-type circuits, including program-mediated circuits whose attention selection is entirely symbolic. At GPT-2 scale, we remove LayerNorm from a sparsemax/LeakyReLU model after training with a +0.0087 OpenWebText loss increase, replace retained attention heads with synthesized restricted-DSL programs, and calibrate only program-local readouts while freezing and hashing every other parameter. The resulting three-edge quote circuit (embedding $\to$ MLP 0 $\to$ program head $\to$ logits) verifies all four properties over a hash-pinned 1,280-prompt domain in linear real arithmetic: 1,280/1,280 equivalence and invariance, 640 edge-necessity witnesses per edge, and robustness at $\epsilon = 0.01$ with minimum certified radius 0.01515. For two alphabet variants of the same opener-detection/copy-type task, untouched gates expose a localization frontier: bracket-type extraction is exact only with all 144 heads, while the constructed quote artifact verifies with three edges. Naive discovery-based verification failed for measured reasons; at scale, we find we must build the object …
Maßnahmen
⬇ Als MarkdownVerwandte Beiträge
- info Best GPU Neoclouds 2026: CoreWeave, Nebius, Lambda, Crusoe, and Groq Ranked by Published Pricing and Contracted Power
- info Anthropic brings Mythos 5 to its Claude Security vulnerability scanner
- info How agents can delegate better
- info Why API Test Generation Is a Judgment Problem, Not a Code Generation Problem