Identifiability theory for contrastive learning predicts that an encoder trained with suitable positive pairs recovers the latent structure behind its inputs, so separately trained networks that perceive a common world should converge on its shared structure without shared training data. This is one proposed explanatio...
Jonathan Bussing· Zenodo (CERN European Organi...· 0 citations
SankeyQA is an open benchmark for testing whether vision-language models can answer questions about career-map Sankey diagrams, of the kind shown by the XR-CareerAssist virtual reality career guidance platform (VOXReality, Horizon Europe). The data are artificial. All 600 career cohorts, their flow tables and the 1,260...
Nikolaos D. Tantaroudas, Andrew J. McCracken, Ilias Karachalios et al.· Zenodo (CERN European Organi...· 0 citations
Paper (preprint, not peer-reviewed), code, dated decision rules and aggregate results for a single-round verified self-training study with LoRA on Qwen3-1.7B-Base. None of the four sequentially prespecified primary tests met its criterion; the intended human-reference positive control also failed to show improvement. G...
Lewis Adkins· Zenodo (CERN European Organi...· 0 citations
In current language models the ability to reason (fluid intelligence) and knowledge about the world (crystallized intelligence) occupy the same parameters, and explicit reasoning must pass one step at a time through a single token. Sinapsi is a modular five-lobe architecture that separates these two components by const...
Marcello 1963- Cozzi· Zenodo (CERN European Organi...· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
First release accompanying the paper "Evaluating Cost–Accuracy Trade-offs of Agentic Scaffolds for Small Language Models". Includes: Colab notebook that runs all experiments and the analysis Per-question outputs for Qwen2.5-0.5B, 1.5B and 3B-Instruct on 500 GSM8K test problems across six strategies (Direct, CoT, PAL, P...
Vasist18· Zenodo (CERN European Organi...· 0 citations
Abstract This monograph develops the Total Economic Cost (TEC) model, a formal optimization framework for analyzing regulatory compliance and investment efficiency in high-risk artificial intelligence systems under the European Union’s Artificial Intelligence Act (Regulation (EU) 2024/1689, as amended by Regulation (EU...
Galia Mancheva· Zenodo (CERN European Organi...· 0 citations
Input tokens for one image at 10 resolutions (256×256 to a 12 MP photo) on Claude Opus/Sonnet 5.5, Claude Haiku 4.5, Gemini 3.x and GPT. Headline: a 12-megapixel phone photo costs 1,065 input tokens on Gemini, 4,005 on Claude 5.5 and 14,364 on GPT-6 Astra with default settings. Gemini charges about the same for every i...
TokenSave· Zenodo (CERN European Organi...· 0 citations
This project archives the supplementary material for the paper "Locating and Mitigating Factual Hallucination Circuits in Small Language Models via Activation Patching" (submitted to IEICE Transactions on Information and Systems). The paper presents a systematic activation-patching study of context–memory conflict in t...
Shutchon Premchaisawatt· Open Science Framework· 0 citations
Thanks to artificial intelligence and machine translation technologies, the volume and speed of translation production have increased. However, these technologies also raise questions about the impact of dominant digital languages on less resourced ones. Many languages remain underrepresented in multilingual modelling...
Sonia Asmahène Halimi· Arab World English Journal F...· 0 citations
Language model serving engines put many customers into one process on one accelerator, sharing one attention cache. Per-tenant rotation stores each tenant’s cached rows under a tenant-specific orthogonal transform and rotates the tenant’s query the same way, so attention is unchanged for the tenant and a read under any...
Luis Carranza· Zenodo (CERN European Organi...· 0 citations
This paper presents an evaluation of small open-weight large language models (LLMs) on named entity recognition (NER) for code-mixed Nepali-English text. No dedicated Nepali-English NER resource currently exists, to the best of the authors’ knowledge. The LinCE benchmark [Aguilar et al. 2020] includes Nepali-English as...
Binod Raj Pant· Zenodo (CERN European Organi...· 0 citations
The same brick-breaker task given to Claude Code and Codex in English, Korean, Japanese, Spanish and Ukrainian, two runs each: input and output tokens, time, cost, whether the game ran, and the language of the game text and code comments. Headline: asking in Korean, Japanese, Spanish or Ukrainian cost about the same as...
TokenSave· Zenodo (CERN European Organi...· 0 citations
A new computational approach reveals how cells communicate to assemble into tissues. This could accelerate stem cell engineering, improve disease models, and inform therapies.
Microsoft Research Blog· microsoft.comOct 15, 2026
Zero-knowledge proofs can verify computing results without rerunning workloads or exposing sensitive data. Magnetar helps bring them into the mainstream with support for standard programming languages, major speed gains, and quantum-resistant security. The post Proof-based general-purpose computing with Magnetar appeared first on Microsoft Research.
Learn how Google DeepMind and GovTech Singapore uses lightweight on-device Gemma models to intercept conversational scams in private chats with zero cloud data egress.
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
