George Drayson

George Drayson

We present Jupiter-N, a hybrid reasoning model post-trained from Nemotron 3 Super, a fully open-source 120 billion parameter LLM. We target three objectives: (1) agentic capability via uncertainty-curated trajectories; (2) UK cultural alignment via synthetic data grounded in cultural norms; and (3) Welsh language support via parallel corpora and LLM-translated Welsh conversations. Our data curation strategy carefully preserves the base model's capabilities: using our Forget-Me-Not framework, we mix on-policy synthetic replay with off-policy task data to mitigate catastrophic forgetting, and include a mixture of reasoning and non-reasoning traces to maintain Nemotron's hybrid reasoning ability. Jupiter-N achieves standout gains over Nemotron in Welsh (+18 on ARC-Easy, +5.25 on MMLU-Lite), terminal-use (+9.1 on Terminal Bench 2) and instruction following (+4.4 on IFBench), while retaining the base model capabilities. We frame this work as a reproducible template for sovereign post-training: substituting cultural knowledge, institutional corpora, and target languages produces an equivalent pipeline for any country. All model weights and all post-training datasets are publicly released under open licences.

George Drayson, Efimia Panagiotaki, Daniel Omeiza et al.

Corner case scenarios are an essential tool for testing and validating the safety of autonomous vehicles (AVs). As these scenarios are often insufficiently present in naturalistic driving datasets, augmenting the data with synthetic corner cases greatly enhances the safe operation of AVs in unique situations. However, the generation of synthetic, yet realistic, corner cases poses a significant challenge. In this work, we introduce a novel approach based on Heterogeneous Graph Neural Networks (HGNNs) to transform regular driving scenarios into corner cases. To achieve this, we first generate concise representations of regular driving scenes as scene graphs, minimally manipulating their structure and properties. Our model then learns to perturb those graphs to generate corner cases using attention and triple embeddings. The input and perturbed graphs are then imported back into the simulation to generate corner case scenarios. Our model successfully learned to produce corner cases from input scene graphs, achieving 89.9% prediction accuracy on our testing dataset. We further validate the generated scenarios on baseline autonomous driving methods, demonstrating our model's ability to effectively create critical situations for the baselines.

George Drayson, Emine Yilmaz, Vasileios Lampos

As Large Language Models (LLMs) become increasingly prevalent, their generated outputs are proliferating across the web, risking a future where machine-generated content dilutes human-authored text. Since online data is the primary resource for LLM pre-training, subsequent models could be trained on an unknown portion of synthetic samples. This could lead to model collapse, a degenerative process whereby LLMs reinforce their own errors, reduce output diversity, and ultimately yield declining performance. In this study, we investigate the impact of decoding strategy on model collapse, analysing the text characteristics at each model generation, the similarity to human references, and the resulting model performance. Using the decoding strategies that lead to the most significant degradation, we evaluate model collapse in a more realistic scenario where the origin of the data (human or synthetic) is unknown. We train a machine-generated text detector and propose an importance resampling approach to prevent model collapse by up-sampling likely human content in the training data. Our method is validated on four LLMs from two model families (GPT-2 and SmolLM2), across a range of model sizes 124M to 1.7B). We demonstrate that it not only prevents model collapse but also improves performance compared to training on purely human data, underscoring the benefit of synthetic samples and the importance of data curation.