Sparse Autoencoders (SAEs) have emerged as a promising solution for decomposing large language model representations into interpretable features. However, Paulo & Belrose (2025) have highlighted instability across different initialization seeds, and Heap et al. (2025) have pointed out that SAEs may not capture model-internal features. These problems likely stem from training SAEs on external datasets—either collected from the Web or generated by another model—which may contain out-of-distribution (OOD) data beyond the model’s generalisation capabilities. This can result in hallucinated SAE features, which we term ”Fake Features”, that misrepresent the model’s internal activations. To address these issues, we propose FaithfulSAE, a method that trains SAEs on the model’s own synthetic dataset. Using FaithfulSAEs, we demonstrate that training SAEs on less-OOD instruction datasets results in SAEs being more stable across seeds. Notably, FaithfulSAEs outperform SAEs trained on webbased datasets in the SAE probing task and exhibit a lower Fake Feature Ratio in 5 out of 7 models. Overall, our approach eliminates the dependency on external datasets, advancing interpretability by better capturing model-internal features while highlighting the often neglected importance of SAE training datasets.

Simultaneous Translation (ST) involves translating with only partial source inputs instead of the entire source inputs, a process that can potentially result in translation quality degradation. Previous approaches to balancing translation quality and latency have demonstrated that it is more efficient and effective to leverage an offline model with a reasonable policy. However, using an offline model also leads to a distribution shift since it is not trained with partial source inputs, and it can be improved by training an additional module that informs us when to translate. In this paper, we propose an Information Quantifier (IQ) that models source and target information to determine whether the offline model has sufficient information for translation, trained with oracle action sequences generated from the offline model. IQ, by quantifying information, helps in formulating a suitable policy for Simultaneous Translation that better generalizes and also allows us to control the trade-off between quality and latency naturally. Experiments on various language pairs show that our proposed model outperforms baselines.

This paper describes our submission to the TL;DR challenge. Neural abstractive summarization models have been successful in generating fluent and consistent summaries with advancements like the copy (Pointer-generator) and coverage mechanisms. However, these models suffer from their extractive nature as they learn to copy words from the source text. In this paper, we propose a novel abstractive model based on Variational Autoencoder (VAE) to address this issue. We also propose a Unified Summarization Framework for the generation of summaries. Our model eliminates non-critical information at a sentence-level with an extractive summarization module and generates the summary word by word using an abstractive summarization module. To implement our framework, we combine submodules with state-of-the-art techniques including Pointer-Generator Network (PGN) and BERT while also using our new VAE-PGN abstractive model. We evaluate our model on the benchmark Reddit corpus as part of the TL;DR challenge and show that our model outperforms the baseline in ROUGE score while generating diverse summaries.