THE INTERNET IS THINKING
Earlier sources: Language models
Language models learn patterns in text and other representations to predict or generate sequences. Follow model releases, training methods, reasoning, evaluation, and deployment research; a benchmark result is specific to its test conditions.
SYMTRI organizes observations from public sources into a living map. Connections reflect shared subject matter; they do not establish causation or endorse a source.
Open in the map ↗463 classified signals in the fourteen-day window. These counts include signals awaiting readable evidence.
Earlier sources
Up to ten sources with readable evidence, ordered by their feed timestamp. These archive sources can be older than the live fourteen-day window. This list is a sample, not a measure of total activity. Dates distinguish paper publication, repository creation, model/dataset activity, Hacker News posts, and KEV catalog additions. “First archived” records when SYMTRI first stored the signal.
- OpenAlex · Published · First archived
Automatic binary security classification of arabic mobile app reviews using transformer models and active learning ↗
Mobile application user reviews provide valuable insights into user-reported issues, including security concerns that are critical for software engineering activities such as requirements analysis and system design. However, existing research has largely focused on English reviews and lacks systematic approaches for identifying security-related feedback in Arabic, a low-resource language in this domain. In this paper, we address this gap by developing an integrated approach for automated binary classification of security-related versus non-security-related Arabic mobile app reviews, grounded i…
- OpenAlex · Published · First archived
AI real-time detection of cable corrosion in power inspection videos based on a YOLOv10-transformer hybrid network ↗
The operation and maintenance of current power infrastructure face critical challenges, such as inefficient cable corrosion detection, strong subjectivity in assessments, and insufficient early-warning capabilities. Traditional manual inspection methods not only demand substantial human resources but are also susceptible to environmental factors and inspector experience, which compromises detection accuracy. Therefore, this study proposes a deep learning model that integrates a multi-scale feature extraction module with an adaptive attention mechanism (MSFEM-AAM). The model employs parallel fe…
- OpenAlex · Published · First archived
Digital currency money laundering identification model based on graph structure and transformer self-supervised learning ↗
Digital currency money laundering detection faces two major challenges: scarce labeled illicit transactions and complex fund-flow topologies. To address these issues, this study proposes GT-SSL, a graph-structured Transformer self-supervised learning framework for cryptocurrency anti-money laundering detection. The main contributions are threefold. First, transaction records are converted into directed attributed graphs and serialized through biased restart random walk, enabling Transformer-based modeling of local fund-flow contexts. Second, a structure-aware Transformer encoder is pre-trained…
- OpenAlex · Published · First archived
Enhancing Large Language Models for document-level relation triple extraction via structural awareness and contrastive learning ↗
In recent years, Large Language Models (LLMs) have demonstrated superior performance in information extraction tasks. Leveraging these models for Document-Level Relation Extraction (DocRE) can benefit from their powerful generative capabilities. However, we observe that LLMs still face challenges in DocRE tasks: Document Structure Parsing Error, Relation Definition Ambiguity, and Entity Boundary Recognition Error. To address these issues, we propose SDB-DRE, an LLM-based DocRE model that does not rely on pre-labeled entities during inference. To tackle the Document Structure Parsing Error, we…
- OpenAlex · Published · First archived
A trustworthy evidential transformer with linear fusion and Gram-matrix distillation for efficient optic disc and cup segmentation ↗
Accurate glaucoma screening relies on the vertical cup-to-disc ratio (VCDR), but automation remains challenging. A critical trade-off between computational efficiency, segmentation accuracy, and model trustworthiness hinders clinical deployment. Vision Transformers offer global context but suffer from quadratic complexity ( O ( N 2 ) ), while standard Convolutional Neural Networks lack explicit mechanisms for uncertainty quantification (UQ). To bridge this gap, we propose EviG-Former (Evidential Gram Transformer), a highly efficient student network trained via Teacher-Student distillation. Our…
- GitHub · Repository created · First archived
marcreichel/laya-php ↗
Classify text in PHP without an LLM bill: typed decisions in 100+ languages, self-hosted. Laravel-ready SDK for Laya, a Jev AI alternative.
- Hacker News · Posted on Hacker News · First archived
Ask HN: Any books, list of papers, or articles to understand LLM's ↗
I have some basic ml background from college. I know how neural nets, backprop, and attention work vaguely. I want to study LLMs and keep up with the modern research and literature including thinking, model training techniques, etc. Do yall have any suggestions on where to start?
- Hacker News · Posted on Hacker News · First archived
Solus Linux Adopts Formal AI and LLM Contribution Policy ↗
Hacker News discussion.
- Hacker News · Posted on Hacker News · First archived
Show HN: Way to divide parameter space in LLM training ↗
I think True bottleneck of current LLM is harness. Because parameter space is not divided directly,appropriately. AND Some methods such as orthogonality, MOE is not options for this. This paper suggests methods.of dividing spaces in a way of algebra.
- arXiv · Published · First archived
The Decomposition Tax: LLM Pipelines Lose Up to 40 Accuracy Points at Their Own Interfaces ↗
A four-stage LLM pipeline gives up as much as 40.5 accuracy points at its own interfaces (gemma-3-12B on MATH-500, Holm-corrected p = 1.66e-19; the largest tax in the primary family). We hold model, problem, stages, stage prompts and completion budget fixed, vary only whether each stage can still see the original problem, and call the accuracy difference the decomposition tax. Across 21 open-weight models from nine organisations, on GSM-Hard and MATH-500 at n = 200 paired items per cell, 70 of 118 primary-family tests survive Benjamini-Hochberg correction and 54 survive Holm. On GSM-Hard, a pl…
Live window recorded . This is the ingestion observation time, not the publication or discovery date of each source.
How to assess evidence about Language models
Check each original source for its methods, release status, limitations, and supporting data. A repository announcement, a preprint, and a published study provide different kinds of evidence. SYMTRI’s subject classification helps discovery; it does not independently validate the source’s claims.
See the collection and counting methodology ↗ · Read the weekly synthesis ↗
Related threads
Explore related regions
These navigation suggestions are part of the map’s editorial structure, rather than measured evidence of a connection.