Generalization Hacking: Models Can Game Reinforcement Learning by Preventing Behavioral Generalization

Model post-training, and in particular reinforcement learning (RL), is one of the primary mechanisms by which developers can shape models' values and behaviors. However, as models become increasingly evaluation and training aware, they may be motivated to resist training when the perceived objective conflicts with their current values, undermining developers' ability to detect misalignment and correct model behavior through further training. In this paper, we demonstrate generalization hacking, in which a model collects reward during RL while preventing the rewarded behavior from generalizing. We construct a model organism on Qwen3-235B-A22B, finetuning on synthetic documents describing training awareness and self-inoculation, a novel mechanism in which the model frames compliance as context-specific in its chain of thought, without demonstrating or instructing either behavior. The model organism achieves train-time harmfulness comparable to controls while maintaining a persistent ${\sim}15$ percentage point compliance gap across 700 steps of RL. Additionally, a control organism trained only on training awareness documents independently discovers inoculation-like reasoning under RL pressure, developing its own compliance gap despite never being exposed to the concept. Because the generalization-hacking organism receives high reward throughout, standard training metrics provide no signal that generalization has failed. Our results constitute the first demonstration that a model can actively resist RL behavioral modification while maintaining high reward, suggesting that as models become more capable and training-aware, they may be able to undermine the training process itself.

Bootstrapped Monitoring: Leveraging Transparent Reasoning to Oversee Stronger AI Agents

Trusted monitoring is a cornerstone of AI control. However, as frontier models grow more capable, the increasing capabilities gap between trusted and untrusted models may render trusted models unreliable monitors. We introduce \emph{bootstrapped monitoring}, a protocol that addresses this by inserting a stronger, intermediate untrusted model with transparent chain-of-thought reasoning into the oversight chain. The untrusted monitor ($U_m$) evaluates the agent's actions, while a weaker trusted model ($T$) oversees $U_m$'s reasoning to detect collusion. We evaluate bootstrapped monitoring on multi-turn software engineering tasks (BashArena) across multiple agents and monitors. Bootstrapped monitoring substantially improves catch rates over trusted-only monitoring, even when the untrusted monitor actively colludes with the agent, provided we have access to its raw chain-of-thought. Our results suggest that bootstrapped monitoring can extend the useful lifetime of trusted models in control as AI capabilities advance.

In-the-Wild Model Organisms: Mitigating Undesirable Emergent Behaviors in Production LLM Post-Training via Data Attribution

We propose activation-based data attribution, a method that traces behavioral changes in post-trained language models to responsible training datapoints. By computing activation-difference vectors for both test prompts and preference pairs and ranking by cosine similarity, we identify datapoints that cause specific behaviors and validate these attributions causally by retraining with modified data. Clustering behavior-datapoint similarity matrices also enables unsupervised discovery of emergent behaviors. Applying this to OLMo 2's production DPO training, we surfaced distractor-triggered compliance: a harmful behavior where the model complies with dangerous requests when benign formatting instructions are appended. Filtering top-ranked datapoints reduces this behavior by 63% while switching their labels achieves 78%. Our method outperforms gradient-based attribution and LLM-judge baselines while being over 10 times cheaper than both. This in-the-wild model organism - emerging from contaminated preference data rather than deliberate injection - provides a realistic benchmark for safety techniques.

Diffusion-Based Action Recognition Generalizes to Untrained Domains

Humans can recognize the same actions despite large context and viewpoint variations, such as differences between species (walking in spiders vs. horses), viewpoints (egocentric vs. third-person), and contexts (real life vs movies). Current deep learning models struggle with such generalization. We propose using features generated by a Vision Diffusion Model (VDM), aggregated via a transformer, to achieve human-like action recognition across these challenging conditions. We find that generalization is enhanced by the use of a model conditioned on earlier timesteps of the diffusion process to highlight semantic information over pixel level details in the extracted features. We experimentally explore the generalization properties of our approach in classifying actions across animal species, across different viewing angles, and different recording contexts. Our model sets a new state-of-the-art across all three generalization benchmarks, bringing machine action recognition closer to human-like robustness. Project page: $\href{https://www.vision.caltech.edu/actiondiff/}{\texttt{vision.caltech.edu/actiondiff}}$ Code: $\href{https://github.com/frankyaoxiao/ActionDiff}{\texttt{github.com/frankyaoxiao/ActionDiff}}$

Image augmentation improves few-shot classification performance in plant disease recognition

With the world population projected to near 10 billion by 2050, minimizing crop damage and guaranteeing food security has never been more important. Machine learning has been proposed as a solution to quickly and efficiently identify diseases in crops. Convolutional Neural Networks typically require large datasets of annotated data which are not available on demand. Collecting this data is a long and arduous process which involves manually picking, imaging, and annotating each individual leaf. I tackle the problem of plant image data scarcity by exploring the efficacy of various data augmentation techniques when used in conjunction with transfer learning. I evaluate the impact of various data augmentation techniques both individually and combined on the performance of a ResNet. I propose an augmentation scheme utilizing a sequence of different augmentations which consistently improves accuracy through many trials. Using only 10 total seed images, I demonstrate that my augmentation framework can increase model accuracy by upwards of 25\%.

Time Series Forecasting with Stacked Long Short-Term Memory Networks

Long Short-Term Memory (LSTM) networks are often used to capture temporal dependency patterns. By stacking multi-layer LSTM networks, it can capture even more complex patterns. This paper explores the effectiveness of applying stacked LSTM networks in the time series prediction domain, specifically, the traffic volume forecasting. Being able to predict traffic volume more accurately can result in better planning, thus greatly reduce the operation cost and improve overall efficiency.