aiqbench analysis
This research demonstrates that transformer hidden state dimensions function as independent binary registers where sign patterns encode semantic content and magnitudes represent confidence levels without requiring training. The study validates this architecture-general framework across multiple model families, showing that pure Hamming scoring of these signs achieves substantial next-token prediction accuracy while revealing distinct semantic categories through unsupervised discovery in both attention projections and feed-forward networks. These findings suggest a significant reduction in computational overhead for interpretability tasks by leveraging the standard basis directly rather than relying on trained probes or optimization-heavy methods.
Bag of Dims: Training-Free Mechanistic Interpretability via Dimension-Level Sign Patterns
arXiv:2606.12629v1 Announce Type: cross Abstract: We show that the standard basis of transformer hidden states already provides a training-free, architecture-general feature basis. Individual dimensions encode semantic content via their signs and confidence via their magnitudes, functioning as independent binary registers. We validate this Bag of Dims framework across three model families (Qwen 3.5-4B, Gemma 3-4B, Mistral 7B) through four progressive experiments. Sign patterns alone carry predictive content: replacing all magnitudes with unity achieves 72-93% top-5 next-token accuracy through the LM head, and pure Hamming scoring without any decoder reaches 80-90% top-4096. These sign patterns organize into semantic features: using a single-token type cache (one forward pass per vocabulary token, no context), we discover 175 categories via per-dimension sign consistency (mean AUC 0.80) from 50 anchors with zero training. A trained probe adds only +0.018 AUC and converges to axis-aligned weights, confirming negligible cross-dimension structure. This structure extends to attention: all 175 categories remain discoverable in K and V projections. On the write side, static FFN weight inspection links 20% of features to individual writer neurons (>0.70 agreement; random controls: 0%), with top-200 neuron coalitions achieving >0.70 agreement on 99.9% of prototypes via majority vote. Fully unsupervised discovery (random seeds, no labels) scales to 1500 features at 100% yield and 99% sparsity across all three models, with pairwise MI of 0.0014 bits confirming low inter-dimension coupling. These results establish that the standard basis already suffices for feature reading throughout the transformer compute pathway, requiring no training, no optimization, and no GPU-days beyond a single forward pass per vocabulary token.
Abstract:We show that the standard basis of transformer hidden states already provides a training-free, architecture-general feature basis. Individual dimensions encode semantic content via their signs and confidence via their magnitudes, functioning as independent binary registers. We validate this Bag of Dims framework across three model families (Qwen 3.5-4B, Gemma 3-4B, Mistral 7B) through four progressive experiments. Sign patterns alone carry predictive content: replacing all magnitudes with unity achieves 72-93% top-5 next-token accuracy through the LM head, and pure Hamming scoring without any decoder reaches 80-90% top-4096. These sign patterns organize into semantic features: using a single-token type cache (one forward pass per vocabulary token, no context), we discover 175 categories via per-dimension sign consistency (mean AUC 0.80) from 50 anchors with zero training. A trained probe adds only +0.018 AUC and converges to axis-aligned weights, confirming negligible cross-dimension structure. This structure extends to attention: all 175 categories remain discoverable in K and V projections. On the write side, static FFN weight inspection links 20% of features to individual writer neurons (>0.70 agreement; random controls: 0%), with top-200 neuron coalitions achieving >0.70 agreement on 99.9% of prototypes via majority vote. Fully unsupervised discovery (random seeds, no labels) scales to 1500 features at 100% yield and 99% sparsity across all three models, with pairwise MI of 0.0014 bits confirming low inter-dimension coupling. These results establish that the standard basis already suffices for feature reading throughout the transformer compute pathway, requiring no training, no optimization, and no GPU-days beyond a single forward pass per vocabulary token.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2606.12629 [cs.LG]
(or arXiv:2606.12629v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2606.12629 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 350 words
Related coverage
[AINews] High Taste Testers Ep. 1
Latent Space · Oct 17, 2026
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026
Links #5: 2026/07
LessWrong · Jul 31, 2026
Angles of attack for continual learning safety
LessWrong · Jun 16, 2026