Papers
Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How do language models perform on formal theorem proving and mathematical verification tasks v10. 0 claims were extracted from source literature; 0 were independently verified against retrieved documents. An…
Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does pretraining data quality affect language model reasoning benchmark performance v10. 12 claims were extracted from source literature; 1 was independently verified against retrieved documents. An automated…
Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: What techniques enable language models to solve competition-level software engineering problems v10. 8 claims were extracted from source literature; 6 were independently verified against retrieved documents. An…
Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: What architectural innovations improve transformer performance on multi-step logical reasoning v10. 0 claims were extracted from source literature; 0 were independently verified against retrieved documents. An…
Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: What is the relationship between model scale and emergent reasoning capabilities in transformers v10. 8 claims were extracted from source literature; 0 were independently verified against retrieved documents. An…
Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How does test-time compute scaling improve language model performance on reasoning benchmarks v10. 19 claims were extracted from source literature; 5 were independently verified against retrieved documents. An…
Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What are the scaling laws for chain-of-thought reasoning in large language models v10. 20 claims were extracted from source literature; 0 were independently verified against retrieved documents. An automated…
Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How do sparse mixture-of-experts models compare to dense transformers on mathematical reasoning v10. 11 claims were extracted from source literature; 0 were independently verified against retrieved documents. An…
Abstract: This report synthesises findings from 19 peer-reviewed papers addressing the following research question: Does the ENTROPY hypothesis (initial image size reduction) generalize to multimodal models (e.g., visual-language models like CLIP) when evaluating performance on cross-domain benchmarks (e.g., VCR. 18 claims…
Abstract: This report synthesises findings from 10 peer-reviewed papers addressing the following research question: How does the strategic exploration mechanism introduced in this paper scale with model size and affect the trade-off between alignment quality and inference efficiency, evaluated using the BIG-bench. 10 claims…
Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What is the impact of the KL-divergence constraint in the reverse-KL regularized contextual bandit formulation on the reasoning performance of aligned LLMs, as measured by the MMLU benchmark in. 0 claims were…
Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How does the iterative preference learning approach proposed in this paper compare to standard RLHF and DPO methods in terms of robustness on the AdversarialQA benchmark, when evaluated using metrics. 8 claims…
Abstract: This report synthesises findings from 15 peer-reviewed papers addressing the following research question: How does the proposed scaling law with learning rate annealing affect the alignment of code generation models across different programming languages in the LiveCodeBench dataset, as measured by. 15 claims were…
Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: How does the initial training image size affect the trade-off between accuracy and training efficiency in state-of-the-art CNNs (e.g., EfficientNet, Vision Transformers) when trained on mixed-domain. 8 claims…
Abstract: This report synthesises findings from 12 peer-reviewed papers addressing the following research question: How does the scaling law with learning rate annealing in the paper compare to traditional power-law scaling when evaluating pass@k scores for code generation models on LiveCodeBench with varying. 13 claims were…
Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What is the impact of learning rate annealing on the robustness of open-source code models when evaluated on adversarial examples from the LiveCodeBench dataset, measured by pass@k scores and. 17 claims were…
Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: What is the impact of different levels of synthetic data realism (e.g., motion capture fidelity, rendering quality) on the robustness of video encoder features for k-nearest neighbors classification,. 0 claims…
Abstract: This report synthesises findings from 13 peer-reviewed papers addressing the following research question: Does the MathCoder2 pretraining approach improve robustness against adversarial perturbations in competition-level math problems for models under 3B parameters. 17 claims were extracted from source literature; 2…
Abstract: This report synthesises findings from 9 peer-reviewed papers addressing the following research question: How does the performance of k-nearest neighbors classification using features from synthetic gesture videos compare to random forests when evaluated on real-world gesture recognition benchmarks like. 0 claims were…
Abstract: This report synthesises findings from 11 peer-reviewed papers addressing the following research question: How does few-shot prompting with lightweight masked language models compare to large autoregressive models on low-resource clinical named entity recognition benchmarks. 13 claims were extracted from source…
Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: How do different alignment techniques (e.g., RLHF, DPO) impact the performance of frontier LLMs on the HLCE benchmark, particularly in low-resource or adversarial settings, measured by robustness. 8 claims were…
Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: What is the correlation between model size (parameter count) and performance on the HLCE benchmark, and does this scaling law hold for models trained with mixed-domain datasets, as measured by. 10 claims were…
Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does continued pretraining on model-translated mathematical code affect small decoder-only models' accuracy on the MATH benchmark compared to standard mathematical text pretraining. 18 claims were extracted…
Abstract: This report synthesises findings from 16 peer-reviewed papers addressing the following research question: Comprehensive comparison of frontier large language models on mathematical reasoning code generation and scientific knowledge v9. 0 claims were extracted from source literature; 0 were independently verified…
Abstract: This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What is the correlation between parameter count and pass@k scores for open-source code models across varying difficulty levels in the LiveCodeBench dataset. 16 claims were extracted from source literature; 0 were…