A paradox of AI fluency

How much does a user's skill with AI shape what AI actually delivers for them? This question is critical for users, AI product builders, and society at large, but it remains underexplored. Using a richly annotated sample of 27K transcripts from WildChat-4.8M, we show that fluent users take on more complex tasks than novices and adopt a fundamentally different interactional mode: they iterate collaboratively with the AI, refining goals and critically assessing outputs, whereas novices take a passive stance. These differences lead to a paradox of AI fluency: fluent users experience more failures than novices -- but their failures tend to be visible (a direct consequence of their engagement), they are more likely to lead to partial recovery, and they occur alongside greater success on complex tasks. Novices, by contrast, more often experience invisible failures: conversations that appear to end successfully but in fact miss the mark. Taken together, these results reframe what success with AI depends on. Individuals should adopt a stance of active engagement rather than passive acceptance. AI product builders should recognize that they are designing not just model behavior but user behavior; encouraging deep engagement, rather than friction-free experiences, will lead to more success overall. Our code and data are available at https://github.com/bigspinai/bigspin-fluency-outcomes

Synthetic Data for any Differentiable Target

What are the limits of controlling language models via synthetic training data? We develop a reinforcement learning (RL) primitive, the Dataset Policy Gradient (DPG), which can precisely optimize synthetic data generators to produce a dataset of targeted examples. When used for supervised fine-tuning (SFT) of a target model, these examples cause the target model to do well on a differentiable metric of our choice. Our approach achieves this by taking exact data attribution via higher-order gradients and using those scores as policy gradient rewards. We prove that this procedure closely approximates the true, intractable gradient for the synthetic data generator. To illustrate the potential of DPG, we show that, using only SFT on generated examples, we can cause the target model's LM head weights to (1) embed a QR code, (2) embed the pattern $\texttt{67}$, and (3) have lower $\ell^2$ norm. We additionally show that we can cause the generator to (4) rephrase inputs in a new language and (5) produce a specific UUID, even though neither of these objectives is conveyed in the generator's input prompts. These findings suggest that DPG is a powerful and flexible technique for shaping model properties using only synthetic training examples.

Document Optimization for Black-Box Retrieval via Reinforcement Learning

Document expansion is a classical technique for improving retrieval quality, and is attractive since it shifts computation offline, avoiding additional query-time processing. However, when applied to modern retrievers, it has been shown to degrade performance, often introducing noise that obfuscates the discriminative signal. We recast document expansion as a document optimization problem: a language model or a vision language model is fine-tuned to transform documents into representations that better align with the expected query distribution under a target retriever, using GRPO with the retriever's ranking improvements as rewards. This approach requires only black-box access to retrieval ranks, and is applicable across single-vector, multi-vector and lexical retrievers. We evaluate our approach on code retrieval and visual document retrieval (VDR) tasks. We find that learned document transformations yield retrieval gains and in many settings enable smaller, more efficient retrievers to outperform larger ones. For example, applying document optimization to OpenAI text-embedding-3-small model improves nDCG5 on code (58.7 to 66.8) and VDR (53.3 to 57.6), even slightly surpassing the 6.5X more expensive OpenAI text-embedding-3-large model (66.3 on code; 57.0 on VDR). When retriever weights are accessible, document optimization is often competitive with fine-tuning, and in most settings their combination performs best, improving Jina-ColBERT-V2 from 55.8 to 63.3 on VDR and from 48.6 to 61.8 on code retrieval.

Invisible failures in human-AI interactions

AI systems fail silently far more often than they fail visibly. In a large-scale quantitative analysis of human-AI interactions from the WildChat dataset, we find that 78% of AI failures are invisible: something went wrong but the user gave no overt indication that there was a problem. These invisible failures cluster into eight archetypes that help us characterize where and how AI systems are failing to meet users' needs. In addition, the archetypes show systematic co-occurrence patterns indicating higher-level failure types. To address the question of whether these archetypes will remain relevant as AI systems become more capable, we also assess failures for whether they are primarily interactional or capability-driven, finding that 91% involve interactional dynamics, and we estimate that 94% of such failures would persist even with a more capable model. Finally, we illustrate how the archetypes help us to identify systematic and variable AI limitations across different usage domains. Overall, we argue that our invisible failure taxonomy can be a key component in reliable failure monitoring for product developers, scientists, and policy makers. Our code and data are available at https://github.com/bigspinai/bigspin-invisible-failure-archetypes

Counterfactual Simulation Training for Chain-of-Thought Faithfulness

Inspecting Chain-of-Thought reasoning is among the most common means of understanding why an LLM produced its output. But well-known problems with CoT faithfulness severely limit what insights can be gained from this practice. In this paper, we introduce a training method called Counterfactual Simulation Training (CST), which aims to improve CoT faithfulness by rewarding CoTs that enable a simulator to accurately predict a model's outputs over counterfactual inputs. We apply CST in two settings: (1) CoT monitoring with cue-based counterfactuals, to detect when models rely on spurious features, reward hack, or are sycophantic, and (2) counterfactual simulation over generic model-based counterfactuals, to encourage models to produce more faithful, generalizable reasoning in the CoT. Experiments with models up to 235B parameters show that CST can substantially improve monitor accuracy on cue-based counterfactuals (by 35 accuracy points) as well as simulatability over generic counterfactuals (by 2 points). We further show that: (1) CST outperforms prompting baselines, (2) rewriting unfaithful CoTs with an LLM is 5x more efficient than RL alone, (3) faithfulness improvements do not generalize to dissuading cues (as opposed to persuading cues), and (4) larger models do not show more faithful CoT out of the box, but they do benefit more from CST. These results suggest that CST can improve CoT faithfulness in general, with promising applications for CoT monitoring. Code for experiments in this paper is available at https://github.com/peterbhase/counterfactual-simulation-training

Transcoder Adapters for Reasoning-Model Diffing

While reasoning models are increasingly ubiquitous, the effects of reasoning training on a model's internal mechanisms remain poorly understood. In this work, we introduce transcoder adapters, a technique for learning an interpretable approximation of the difference in MLP computation before and after fine-tuning. We apply transcoder adapters to characterize the differences between Qwen2.5-Math-7B and its reasoning-distilled variant, DeepSeek-R1-Distill-Qwen-7B. Learned adapters are faithful to the target model's internal computation and next-token predictions. When evaluated on reasoning benchmarks, adapters match the reasoning model's response lengths and typically recover 50-90% of the accuracy gains from reasoning fine-tuning. Adapter features are sparsely activating and interpretable. When examining adapter features, we find that only ~8% have activating examples directly related to reasoning behaviors. We deeply study one such behavior -- the production of hesitation tokens (e.g., "wait"). Using attribution graphs, we trace hesitation to only ~2.4% of adapter features (5.6k total) performing one of two functions. These features are necessary and sufficient for producing hesitation tokens; removing them reduces response length, often without affecting accuracy. Overall, our results provide insight into reasoning training and suggest transcoder adapters may be useful for studying fine-tuning more broadly.

Language models as tools for investigating the distinction between possible and impossible natural languages

We argue that language models (LMs) have strong potential as investigative tools for probing the distinction between possible and impossible natural languages and thus uncovering the inductive biases that support human language learning. We outline a phased research program in which LM architectures are iteratively refined to better discriminate between possible and impossible languages, supporting linking hypotheses to human cognition.

Addressing divergent representations from causal interventions on neural networks

A common approach to mechanistic interpretability is to causally manipulate model representations via targeted interventions in order to understand what those representations encode. Here we ask whether such interventions create out-of-distribution (divergent) representations, and whether this raises concerns about how faithful their resulting explanations are to the target model in its natural state. First, we demonstrate empirically that common causal intervention techniques often do shift internal representations away from the natural distribution of the target model. Then, we provide a theoretical analysis of two classes of such divergences: `harmless' divergences that occur in the null-space of the weights and from covariance within behavioral decision boundaries, and `pernicious' divergences that activate hidden network pathways and cause dormant behavioral changes. Finally, in an effort to mitigate the pernicious cases, we modify the Counterfactual Latent (CL) loss from Grant (2025) that regularizes interventions to remain closer to the natural distributions, reducing the likelihood of harmful divergences while preserving the interpretive power of interventions. Together, these results highlight a path towards more reliable interpretability methods.

Multi-module GRPO: Composing Policy Gradients and Prompt Optimization for Language Model Programs

Group Relative Policy Optimization (GRPO) has proven to be an effective tool for post-training language models (LMs). However, AI systems are increasingly expressed as modular programs that mix together multiple LM calls with distinct prompt templates and other tools, and it is not clear how best to leverage GRPO to improve these systems. We begin to address this challenge by defining mmGRPO, a simple multi-module generalization of GRPO that groups LM calls by module across rollouts and handles variable-length and interrupted trajectories. We find that mmGRPO, composed with automatic prompt optimization, improves accuracy by 11% on average across classification, many-hop search, and privacy-preserving delegation tasks against the post-trained LM, and by 5% against prompt optimization on its own. We open-source mmGRPO in DSPy as the dspy.GRPO optimizer.

Improved Representation Steering for Language Models

Steering methods for language models (LMs) seek to provide fine-grained and interpretable control over model generations by variously changing model inputs, weights, or representations to adjust behavior. Recent work has shown that adjusting weights or representations is often less effective than steering by prompting, for instance when wanting to introduce or suppress a particular concept. We demonstrate how to improve representation steering via our new Reference-free Preference Steering (RePS), a bidirectional preference-optimization objective that jointly does concept steering and suppression. We train three parameterizations of RePS and evaluate them on AxBench, a large-scale model steering benchmark. On Gemma models with sizes ranging from 2B to 27B, RePS outperforms all existing steering methods trained with a language modeling objective and substantially narrows the gap with prompting -- while promoting interpretability and minimizing parameter count. In suppression, RePS matches the language-modeling objective on Gemma-2 and outperforms it on the larger Gemma-3 variants while remaining resilient to prompt-based jailbreaking attacks that defeat prompting. Overall, our results suggest that RePS provides an interpretable and robust alternative to prompting for both steering and suppression.