Korean
PulseAugur coverage of Korean — every cluster mentioning Korean across labs, papers, and developer communities, ranked by signal.
15 day(s) with sentiment data
-
New benchmark KoNeoBench tests LLMs on Korean neologisms
Researchers have introduced KoNeoBench, a new benchmark designed to evaluate how well large language models understand Korean neologisms. The dataset comprises 1,785 neologisms found in online news since 2020, each acco…
-
PaGNet combines GBDT and neural networks for corporate tax avoidance forecasting
Researchers have developed PaGNet, a novel hybrid model that combines Gradient Boosting Decision Trees (GBDT) with neural networks to forecast corporate tax avoidance proxies. This model addresses the challenge of extra…
-
Apple's iOS 27.2 beta adds new languages for Siri AI
Apple's latest iOS beta, version 27.2, has expanded Siri's AI capabilities to include several new languages such as French, Japanese, Korean, Portuguese, and Spanish. This update builds upon the existing English-only Si…
-
330 LLMs tested on Korean; many fail script adherence
A recent evaluation of 330 language models tested their performance on Korean language tasks, revealing significant issues with script adherence. A significant portion of models failed to maintain correct script usage, …
-
AI Agent Tool for Korean Business Verification Developed
A developer has created a remote MCP server designed to allow AI agents to verify Korean businesses using official data. This project, detailed on Mastodon, aims to leverage AI for business verification processes.
-
New 'synthetic blips' method analyzes dynamic treatment effects
Researchers have developed a new method called "synthetic blips" to analyze dynamic treatment effects in situations where units receive multiple sequential treatments. This approach generalizes synthetic control methods…
-
New Korean LLM benchmark KoSimpleQA reveals significant factuality gaps
Researchers have introduced KoSimpleQA, a new benchmark designed to evaluate the factuality of large language models (LLMs) specifically for Korean cultural knowledge. The benchmark comprises 938 short, fact-seeking que…
-
Korean LLM Alignment Leads to Unintended Response Changes
Researchers have investigated the unintended consequences of aligning a Korean 27B language model, Qwen3.8-27B, to a specific response style. The study found that while the model was trained for verbosity, list usage, a…
-
New methods improve multilingual video transcription accuracy
Researchers have developed methods to improve speech transcription accuracy from videos across multiple languages, aiming to aid the creation of automated tools for cross-cultural understanding. By leveraging publicly a…
-
New framework improves Korean speech recognition error correction
Researchers have developed a new text-only framework called Detector-Gated Contextual Span Correction (DCSC) to improve Korean speech recognition error correction. This method is designed to address the scarcity of anno…
-
LoGAN framework uses VLM agents for multilingual font localization
Researchers have introduced LoGAN, a novel framework utilizing a vision-language model (VLM) to facilitate multilingual font localization. This approach breaks down the complex task into several components, including a …
-
LLM deliberation struggles to represent population opinion in simulations
A new research paper explores the use of multi-agent LLM deliberation to simulate public discourse, finding that while these simulations can generate reasoned arguments and show significant opinion shifts, they struggle…
-
Apple's Siri AI expands to five new languages in October
Apple is expanding the availability of its redesigned Siri AI to five new languages: French, Japanese, Korean, Portuguese, and Spanish, starting in October. The AI-powered assistant has been in beta testing in English a…
-
Korean AI Foundry Recombines Models, Skips Costly Pretraining
A small Korean team has developed an "AI Foundry" approach, shifting focus from expensive LLM pretraining to model diagnosis, recombination, and optimization for specific domains. This method involves inspecting trained…
-
South Korean startup VIDRAFT announces independent AI foundation model
VIDRAFT, a South Korean AI startup, has announced the development of an independent AI foundation model. The announcement was made through an official government policy briefing channel, indicating the model's national …
-
New benchmark reveals LLMs fail to track customer financial history
A new benchmark called FinLifeBench has been introduced to evaluate the ability of large language models to reconstruct a customer's life events and financial state over time from banking dialogues. The benchmark, which…
-
New benchmark reveals LLMs struggle with precise text structure reconstruction
Researchers have developed OrderProbe, a new benchmark designed to evaluate how well large language models (LLMs) can reconstruct the precise structural order of text. Unlike previous methods that allowed for multiple c…
-
New Korean Jamo-Level Typo Vulnerability Found in LLMs
Researchers have identified a new vulnerability in large language models related to Korean typography, specifically at the jamo (sub-character unit) level. Errors within Korean syllable blocks can lead to corrupted inpu…
-
New KoALa-Bench benchmark evaluates Korean speech understanding in LLMs
Researchers have introduced KoALa-Bench, a new benchmark designed to evaluate the performance of large audio language models (LALMs) specifically on Korean speech understanding and faithfulness. The benchmark includes s…
-
New benchmark READI tests AI's understanding of indirect speech in visual contexts
Researchers have introduced READI, a new multimodal benchmark designed to evaluate how well AI models understand indirect speech acts (ISAs) within visual contexts. Existing benchmarks often fail to capture the pragmati…