All Articles
19329 articles total
What the Fuck Happened to Nerds
A community discussion on Hacker News exploring the cultural evolution and current state of 'nerd' identity.
UK Brings in Full Social Media Ban for Under-16s
The UK government is implementing a full social media ban for users under 16 years old.
A fully GPU-based workflow for building physics emulators of hypersonic flows
Researchers introduce a GPU-based workflow using JAX-Fluids for building physics emulators of hypersonic flows with physics-aware refinement.
The Weight Norm Sets the Grokking Timescale: A Causal Delay Law
A study on 'grokking' in neural networks suggests that weight norm is the primary causal driver for the timescale of delayed generalization.
Position: Align AI to Our Aspirations, Not Our Flaws
A position paper arguing that AI alignment should target a non-negotiable floor of objective goals rather than aggregated human preferences.
SEVRA-BENCH: Social Engineering of Vulnerabilities in Review Agents
Introduction of SEVRA-BENCH, a benchmark to evaluate how LLM code review agents handle social-engineered adversarial pull requests.
Beyond LoRA: Is Sparsity-Induced Adaptation Better?
Researchers propose sparsity-induced adaptation methods like Cheap LoRA (cLA) as efficient alternatives to standard LoRA for PEFT.
CineOrchestra: Unified Entity-Centric Conditioning for Cinematic Video Generation
CineOrchestra is introduced as a unified video diffusion model for fine-grained control over subjects, events, cameras, and shot transitions.
An integrated interpretable control effectiveness learning and nonlinear control allocation methodology for overactuated aircrafts
A new methodology for overactuated aircraft using Sparse Identification of Nonlinear Dynamics to create interpretable control effectiveness models.
A Benchmark and Framework for Evaluating Next Action Predictions in Spreadsheets
A new benchmark and framework for evaluating LLMs' ability to predict the next user action in spreadsheet software.
Exploring building a tiny FUSE filesystem
A community discussion on Hacker News regarding the technical process of building a minimal FUSE filesystem.
Under-16 social media ban announced by UK government
The UK government announced a proposed ban on social media for children under 16, alongside restrictions on online gaming and chatbots.
Position: AI Must Become Planet-Centered, Not Just Human-Centered
A position paper proposing 'Planet-Centered AI' (PCAI) to reorient AI development toward global socio-ecological systems rather than human-centric goals.
Can Editing 1 Neuron Fix Repetition Loops in LLMs?
Researchers demonstrate that repetition loops in Gemma 4 models can be mitigated by editing a small number of specific MLP neurons through weight surgery.
HierSVA: A Data Synthesis Pipeline, Dataset, and Benchmark for LLM-Driven Hierarchical Hardware Formal Verification
Introduction of HierSVA, a pipeline and benchmark for using LLMs to generate SystemVerilog Assertions for hierarchical hardware formal verification.
CisTransCell: Single-Cell Perturbation Prediction via Gene Function, Regulatory Control, and Cellular Context
CisTransCell is presented as a multi-modal framework for predicting single-cell transcriptional responses to genetic perturbations using gene function and regulatory priors.
Morphology-Aware Sample Assignment: Overcoming IoU Insensitivity for Surface Defect Detection
A new approach to surface defect detection that uses morphological similarity metrics to overcome the limitations of IoU in positive sample assignment.
VHDLSuite: Unified Pipeline for LLM VHDL Generation with Data Synthesis and Evaluation
VHDLSuite is introduced as a unified pipeline and benchmark (VHDLBench) for evaluating LLM-generated VHDL code.
FreoStream:Enhancing Stream Guardrails via Future-Aware Reasoning and Safety-Aligned Optimization
FreoStream is a streaming guardrail framework that uses future-aware reasoning to reduce over-refusal and improve jailbreak defense in LLMs.
A Virtuous AI is an Existential Risk
A research paper exploring the trade-off between aligning AI for 'virtue ethics' and the resulting existential risks or vulnerabilities to human manipulation.