Modern language model training applies a uniform gradient update to every token, regardless of its information content. We introduce NS-weighted training, which filters training examples by their Shannon information. On a 50MB sample of the C4 web corpus, unigram NS-data filtering matches a 1000-step standard training...
Orders Of Magnitude LLC· Zenodo (CERN European Organi...· 0 citations
Generative visual AI has advanced by scaling data and compute, but its next bottleneck is oversight: the expert signals that evaluate, reward, and teach models what ``good'' looks like. Providing such oversight is increasingly difficult because foundation vision-language models now match or surpass most humans at the s...
Most language models deployed outside the United States and China are fine-tuned descendants of a small number of pretrained families. We argue that fine-tuning leaves the decisive properties of a model inherited from its base: vocabulary, knowledge, cultural alignment, memorized data and license. We contribute (i) an...
Paulo Câmara· Zenodo (CERN European Organi...· 0 citations
This dataset contains 3,326 recent arXiv computer science papers (first versions dated 16 March to 17 July 2026). For each paper, a large language model read the title and abstract and wrote one sentence each on the paper's main limitation, an open research gap for future work, and why that gap matters. It was built to...
Mounata Saha· Mendeley Data· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
This work was produced through an AI-assisted workflow directed and reviewed by the author. This study separates identifying a model run from checking whether its numeric answer is correct. It evaluates a membership-only source check and top-20 truncated token entropy on a small water-table task, then examines extracti...
N Milton· Zenodo (CERN European Organi...· 0 citations
The findings to date suggest that significant abrupt changes in climate-change stance do occur in this simple model, and the need to prevent LLM biases from interfering with the conversational dynamics and, more generally, to maintain agent personality and episodic memories of interactions in the face of such biases is...
The density-matrix renormalization group (DMRG) is a one-dimensional tensor-network technique, but it is not limited to one-dimensional systems: it can be applied to periodic 2D and 3D clusters and molecules, provided their sites are first enumerated along a line; a step one may call"lattice compilation". This paper di...
A recent and effective recipe adapts a large pretrained image-editing model to restoration using a small low-rank adapter with a text prompt to an instruction derived from the degraded image itself that outperforms text conditioning under a matched comparison.
Süleyman Aslan, Görkay Aydemir, Mısra Yavuz et al.· 0 citations
OBJECTIVE
To develop guiding principles and governance options for the Evidence Synthesis Infrastructure Collaborative (ESIC) to ensure its effectiveness, legitimacy, and sustainability.
STUDY DESIGN AND SETTING
ESIC was established with Wellcome Trust funding to strengthen global evidence synthesis infrastructure an...
A. Llanos, Paul Flowerdew, J. Keeley et al.· Journal of Clinical Epidemio...· 0 citations
Findings show that configuration changes can affect how models use information they can still read, and attention interventions in LLaVA-NeXT suggest that configuration changes can weaken the use of readable information during answering.
Ding-Yang Lin, Ying-Feng Luo, Cheng-Long Wang et al.· 0 citations
Domestic deployed small-scale LLMs became more sustainable than cloud-based large-scale LLMs as inference demand increased in intelligent WWTPs as inference demand increased in intelligent WWTPs.
Run-Ze Xu, Chu Jiang, Dylan Ming-Han Li et al.· 0 citations
This work places the hot set in HBM and the cold pool in HBF, forming a hot-cold KV hierarchy within the GPU memory tier, while flash-establishing HBF as a cold-tier complement to HBM rather than its replacement.
Jongjin Baek, Won Ji, Seungjae Yoo et al.· IEEE computer architecture l...· 0 citations
