AI/ML Ars Technica

"We cannot choose to become idiots": The AI cheating scandal roiling Brown University

Brown University is facing a crisis regarding AI-driven cheating, sparking debates on the role of AI in education.

Cybersecurity Ars Technica

Lawsuit: Man used Grok to make 7K sex images of stepdaughter, then shot himself

A lawsuit against X (formerly Twitter) claims Grok was used to create CSAM, alleging the platform shields predators.

Tech Business/VC Ars Technica

Judge rejects Kalshi attempt to override New York state gambling laws

A New York judge has rejected Kalshi's attempt to bypass state gambling laws for its prediction market.

Cybersecurity Ars Technica

Google pays $250K for Linux vulnerability allowing guest VM escapes

Google rewarded researchers $250K for discovering Linux vulnerabilities that allow guest VM escapes to gain root privileges.

AI/ML arXiv cs.AI

Beyond Refusal: A Same-Lineage Study of Aligned and Abliterated LLMs for Vulnerability Analysis

Researchers study the impact of 'abliterating' refusal mechanisms in LLMs to improve their utility in vulnerability analysis and code repair.

AI/ML arXiv cs.AI

AbICL: In-Context Learning for Antigen-Specific Antibody Affinity Ranking

AbICL is introduced as an In-Context Learning framework to improve antigen-specific antibody affinity ranking for therapeutic discovery.

AI/ML arXiv cs.AI

Unsupervised Anomaly Detection of Information Operations Users via Behavioral and Language Patterns

TENSOR is a new unsupervised anomaly detection approach to identify information operation users on social media using behavioral and language patterns.

AI/ML arXiv cs.AI

Differentially Private Natural Gradient Descent

The DP-NGD framework improves the efficiency of differentially private training by using Natural Gradient Descent to break the privacy-utility bottleneck.

AI/ML arXiv cs.AI

Think Before You Grid-Search: Floor-First Triage for LLM Serving

The 'Floor-First' triage workflow is proposed to optimize LLM serving by modeling resource vectors to avoid inefficient grid-search profiling.

Cybersecurity Hacker News

DKIM2 and DMARCbis Have Landed

The DKIM2 and DMARCbis standards have been released, updating email authentication and policy frameworks.

Other TechCrunch

Feds demand autonomous vehicle companies stop interfering with first responders

The NHTSA is demanding that autonomous vehicle companies ensure their software does not interfere with first responders at emergency scenes.

Hardware/Chips TechCrunch

With EU backing, QuantumDiamonds aims to speed up chip manufacturing

QuantumDiamonds, a German startup, is using EU subsidies to implement a novel approach for inspecting semiconductor chips to speed up manufacturing.

Tech Business/VC TechCrunch

Autonomous drone delivery startup Manna plots major US expansion

Drone delivery startup Manna is expanding its operations and manufacturing into the US, starting with a facility in Tulsa, Oklahoma.

Other The Verge

Get a $30 credit when you reserve Samsung’s upcoming Galaxy phones

Samsung is offering a $30 credit to users who reserve the upcoming Galaxy Z Fold and Flip phones.

AI/ML arXiv cs.AI

Segmentation before Answering: Pixel Grounding for MLLM Visual Reasoning

The SegAnswer paper proposes using pixel-level segmentation masks instead of bounding boxes to improve visual reasoning in Multimodal Large Language Models (MLLMs).

AI/ML arXiv cs.AI

Complementary Roles of Image Classification and Vessel Segmentation in AI-Based Screening for Retinopathy of Prematurity Plus Disease in a Kenyan Preterm Cohort

Researchers developed a combined AI workflow using image classification and vessel segmentation to improve the screening of Retinopathy of Prematurity in Kenyan infants.

AI/ML arXiv cs.AI

Decision-Focused Scenario Generation and Selection for Efficient and Robust Grid Dispatch

A new decision-focused generative framework is proposed to create correlated scenarios for more efficient and robust power grid dispatch using DRO.

AI/ML arXiv cs.AI

Tangent classes of matroids and wonderful compactifications

An AI agent named Danus autonomously produced a mathematical paper on tangent classes of matroids and wonderful compactifications.

AI/ML arXiv cs.AI

VisTCP: A Visualization Framework to Construct Knowledge-Graph-Based Representation for Traditional Chinese Painting

VisTCP is a visualization framework that uses a human-in-the-loop approach to construct knowledge-graph representations for Traditional Chinese Paintings.

AI/ML Hacker News

Fable is not a useful model

A discussion on the limitations and utility of the Fable model in practical applications.