All Articles
16266 articles total
Group Preference Collapse in Personalized Multimodal Large Language Models
PrefMoE is a preference-centric framework designed to prevent 'group preference collapse' in personalized multimodal LLMs, preserving individual user preferences.
Evaluating LLMs as Interpretable Controllers for Dynamical Systems
A study evaluates the capability of high-complexity LLMs to act as interpretable controllers for dynamical physical systems, specifically thermal environments.
Tokengeist: Multi-Turn Attribution Tracing in Agentic Conversations
Tokengeist is a framework for multi-turn attribution tracing in agentic conversations, identifying how prior turns recursively shape model responses.
Decentralized Granular Access Control for Agentic AI Systems in Critical Infrastructure
A new decentralized, multi-layered access control architecture for agentic AI systems in critical infrastructure prevents unauthorized write operations through compound identity and progressive trust.
DynaResize: Runtime GPU Reallocation for Disaggregated LLM Post-Training
DynaResize is a runtime GPU reallocation system for disaggregated LLM post-training that balances rollout and training stages to improve throughput.
OpenAI just open-sourced Codex Security
OpenAI has open-sourced Codex Security, providing tools to enhance the security of AI-generated code.
Interview with Boris Cherny [video]
An interview with Boris Cherny, likely discussing technical topics given the source, though specific details are absent from the snippet.
I sent Claude Opus 5 '–-' and it wrote me 5k tokens about a cartographer
A user reports that providing a specific minimal prompt to Claude Opus 5 resulted in an unexpectedly long 5k token story about a cartographer.
Show HN: How far do I have to go to run into 100k people?
A 'Show HN' project exploring the geographical distribution of people and calculating distance to encounter 100k individuals.
The union drive at the Wikimedia Foundation is expanding
The unionization efforts at the Wikimedia Foundation are expanding to include US-based employees, despite management's lack of voluntary recognition.
eBay’s bizarre cyberstalking saga ends with a $56 million settlement
eBay and former executives agreed to a $56 million settlement following a bizarre and aggressive cyberstalking campaign against a couple who ran a newsletter.
xMIx: High-Performance Serving-Time Platform for Mechanistic Interpretability Apps
xMIx is introduced as a high-performance serving-native framework for deploying mechanistic interpretability applications in production AI environments with minimal overhead.
An Agentic Orchestration of Atomistic Simulations
Researchers present an agent-based system within the URSA framework that automates atomistic simulations using tools like LAMMPS, reducing manual trial-and-error.
HyCE-RAG: Hypergraph Chain-of-Evidence Retrieval-Augmented Generation for Explainable Multi-hop Question Answering
HyCE-RAG proposes a hypergraph-based retrieval-augmented generation framework to improve explainability and accuracy in multi-hop question answering.
Differencing the Diffusion Trajectory toward Uncertain Components for Time Series Forecasting
DiffDiff is a new diffusion framework for time series forecasting that accounts for predictability asymmetry to concentrate generative effort on uncertain components.
Recursion Is Lying to You
A discussion on Hacker News regarding the misconceptions and nuances of recursion in programming.
I'd not buy a LG monitor
A user's negative recommendation against purchasing LG monitors based on personal experience or community sentiment.
MCP startup Runlayer accuses Rippling of stealing its product idea
MCP startup Runlayer is suing Rippling for allegedly stealing its product idea after an evaluation process.
Sam Altman is ready to decelerate
Sam Altman discusses a shift in perspective toward decelerating AI development following a visceral security incident.
New airliner sets record flying 24 hrs nonstop from Australia to France
A new airliner has set a record by flying 24 hours nonstop from Australia to France, proving the viability of Project Sunrise.