Law
PulseAugur coverage of Law — every cluster mentioning Law across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
LLMs struggle with expert disagreement, study finds
A new study published on arXiv investigates how large language models (LLMs) revise their decisions when faced with expert disagreement. The research found that LLMs often fail to improve accuracy when revising answers …
-
AI systems exploit vulnerabilities in German law, prompting calls for legislative hardening
Artificial intelligence systems are demonstrating the ability to identify and exploit vulnerabilities within legal systems, a phenomenon described as "AI hacking the law." Researchers have used large language models to …
-
New method validates AI workflows in legal document review
Researchers have introduced the Delayed Countdown Method, a new approach to validate AI-assisted document review workflows in electronic discovery. This method aims to resolve controversies surrounding the effectiveness…
-
New LAW technique enables pure self-conditioning for language models
Researchers have developed a new language modeling technique called LAW (Loopholing All the Way) that relies purely on self-conditioning. This method propagates expected embeddings through the denoising process, decodin…
-
New benchmark evaluates LLMs' multi-turn factual question answering
Researchers have introduced KnowMT-Bench, a new benchmark designed to evaluate multi-turn, long-form question answering capabilities of large language models in knowledge-intensive domains like medicine, finance, and la…
-
AI Training Data Rules Could Cost EU Trillions by 2035
A new economic analysis from The Lisbon Council examines the financial implications of different regulatory approaches to text and data mining (TDM) for AI training. The report suggests that opt-out mechanisms, as oppos…
-
New CiteVQA benchmark exposes "Attribution Hallucination" in LLMs
A new benchmark called CiteVQA has been introduced to evaluate the evidence attribution capabilities of multimodal large language models (MLLMs). Current document question-answering (Doc-VQA) evaluations only assess the…
-
$OneMillion-Bench benchmark tests language agents on expert-level professional tasks
Researchers have introduced $OneMillion-Bench ($OMB), a new benchmark designed to evaluate the capabilities of language agents in complex, real-world professional scenarios. Unlike previous benchmarks, $OMB comprises 40…
-
AI ethics podcast explores driverless car collision with police vehicle
An episode of the "Law and Ethics in Tech" podcast discusses a hypothetical scenario where a driverless car collides with a police vehicle. The discussion delves into the ethical considerations and broader implications …
-
New CineSubBench benchmark evaluates LLMs on film understanding
Researchers have introduced CineSubBench, a new benchmark designed to evaluate the long-context understanding capabilities of large language models (LLMs) using multilingual movie subtitles. This benchmark includes over…
-
LLMs increasingly mediate human rights law, but lack evaluation benchmarks
A new paper highlights the critical role large language models play in legal decisions concerning human rights, yet points out a lack of benchmarks to evaluate their reasoning capabilities in this domain. The research s…
-
Judges increasingly use AI for legal opinions, leading to errors
Judges and their clerks are increasingly delegating the writing of legal opinions to AI, leading to instances of fabricated citations and errors in court rulings. A database has been compiled documenting over 2000 such …
-
LLM use may threaten freedom of thought and equality across sectors
The increasing use of large language models (LLMs) poses potential risks to fundamental freedoms and equality. Specifically, these advanced AI systems could negatively impact critical sectors such as law, journalism, an…
-
Huawei's Kirin 2026 silicon uses LogicFolding to boost density and cut power
Huawei's Kirin 2026 silicon is detailed in a new paper, suggesting that LogicFolding technology can achieve significantly higher transistor densities. This advancement reportedly allows for substantial reductions in pow…
-
RetrievalRouter system optimizes document retrieval by query-aware pipeline selection
Researchers have developed RetrievalRouter, a novel system designed to optimize document retrieval by dynamically selecting the most appropriate retrieval pipeline based on the query. This approach addresses the trade-o…
-
Language model test-time scaling bottleneck identified in output selection
A new arXiv paper investigates test-time scaling (TTS) techniques for language models, which improve outputs by using additional inference compute. The research found that while scaling exploration effectively increases…
-
Hong Kong accounting body proposes tax breaks for Northern Metropolis development
The Hong Kong Institute of Certified Public Accountants (HKICPA) has proposed a five-year plan for the Northern Metropolis, suggesting tax breaks for start-ups and lower tax rates for workers in the area. The organizati…
-
Senior detective investigated for AI misuse
A senior detective is currently under investigation for their alleged misuse of artificial intelligence. The specific details of the AI's application and the nature of the investigation are not fully disclosed in the pr…
-
New benchmarks and studies reveal LLM safety and performance trade-offs · 2 sources tracked
Two new research papers explore the safety and performance trade-offs of large language models (LLMs). The first paper introduces Trident-Bench, a benchmark designed to evaluate LLM safety and compliance specifically wi…
-
AI legitimacy crisis looms, distinct from alignment: new paper
A new academic article argues that the current focus on AI alignment is insufficient for addressing the broader issue of AI legitimacy. The paper posits that legitimacy, defined as the belief among those subject to AI's…