The Information · Google · Alignment Forum
Investigating this risk, Engels et al. surprisingly surfaced that DiffusionGemma scores similar monitorability to Gemma
Compiled by KHAO Editorial — aggregated from 1 source. See llms.txt for citation guidance.
◌ Single Source
Thus somehow the information in seemed to have been essential, conflicting with the results of high monitorability.
Key facts
- The team use 56 binary concept datasets from the SAE-Probes benchmark ( Kantamneni et al
- Specifically, they use more diffusion steps ( = 48 → 96), a wider temperature range (0.8–0.4 → 1.0–0.5), and a lower entropy bound (0.1 → 0.02), all of which contribute to stabilization
- Note however that there is a natural mechanism to implement this computation: embedding letters as points on a circle and rotating by an angle of 3/26
- Example of one intervention. a subleading injection on H (the leader G keeps rank 1) switches the answer slot from J to K = H+3 one step later
Summary
Google DeepMind's recent model DiffusionGemma (DG) generates text via diffusion, meaning many diffusion steps happen before generating the final output. Still, they also find some rare case studies where the distribution vector is load-bearing computationally, i.e. where top-1 projection would be detrimental. Apart from model behavior, they also examined how interpretability techniques carry over to DiffusionGemma, including probes, steering, and J-lens. Overall, this supports the paper's conclusion that DiffusionGemma remains highly monitorable, while nevertheless showing that there are cases where models can learn to use vector-valued information. Large language models arrive at answers to complex questions through chains-of-thought. DiffusionGemma is a particular model whose architecture allows latent reasoning, by passing a vector encoding a probability distribution between steps.