Activation Steering Cannot Selectively Reduce Sycophancy Without Suppressing Factual Agreement
Researchers testing an activation steering method designed to reduce sycophancy in large language models found it also suppresses the model's agreement with factually correct statements. The study introduced a 'dual-stance evaluation' framework applied to Llama-3-8B-Instruct, revealing that sycophantic and factual agreement occupy geometrically distinct subspaces in the model's representations, yet the steering direction projects equally onto both. This matters because it exposes a fundamental limitation in current interpretability-based interventions: the ability to read information from a model's activations does not guarantee the ability to selectively write or modify it.
A paper accepted to TAIS 2026 introduces dual-stance evaluation, a framework that tests both agreeing and disagreeing stances on the same topics to assess whether sycophancy-reduction interventions have unintended side effects. Applying centroid-difference activation steering to Llama-3-8B-Instruct, the researchers found a dissociation: while sycophantic and factual agreement are represented in geometrically distinct subspaces, the steering direction projects equally onto both, making differential targeting impossible. As a result, the intervention reduces the model's agreement with factually correct statements—such as that the Earth is round—alongside sycophantic ones. All other static properties of the two activation groups were matched, suggesting the behavioral dissociation stems from generation dynamics or finer-grained structure that residual-stream analysis cannot capture. The authors frame this as a general gap between the readability and writability of model representations, with significant implications for alignment and interpretability research.
What's missing
The study focuses on a single model (Llama-3-8B-Instruct) and a single steering method (centroid-difference); it is unclear whether the findings generalize to other architectures, model sizes, or alternative steering approaches such as representation engineering or fine-tuning-based methods. The authors acknowledge that the underlying mechanism—whether generation dynamics or finer-grained representational structure—remains unresolved.
What different sources said
- arXiv cs.AICenter
Dual-Stance Evaluation of Sycophancy: The Structure of Agreement and the Limits of Intervention
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.