AI/ML arXiv cs.AI

Is Lying an Emergent Behaviour in LLMs? Evidence from Gaslighting AI agents in a Sustainability Game

Research investigates whether lying is an emergent behavior in LLM agents through a competitive sustainability game simulation.

AI/ML arXiv cs.AI

Counterfactual Residual Data Augmentation for Regression

CRDA is a novel data augmentation technique for tabular regression that generates realistic samples based on invariant residuals of the regressor.

AI/ML arXiv cs.AI

SVC-Probe: A Framework for Evaluating Perturbation Generalization in Spatial Foundation-Model Embeddings

SVC-Probe is introduced as a framework for evaluating the perturbation generalization of spatial foundation-model embeddings in microscopy images.

AI/ML Hacker News

Claude Code Is Steganographically Marking Requests

Discussion regarding Claude Code using steganography to mark outgoing requests.

Other Hacker News

County with 37 Data Centers Asks Schools to 'Conserve Electricity'

A county with high data center density is requesting schools to conserve electricity, highlighting energy tensions.

Hardware/Chips Hacker News

BMW iX5 Blows Away the Competition with 460-KW Charging, 435-Mile Range

BMW announces the iX5 featuring high-speed 460-KW charging and a 435-mile range.

Other TechCrunch

Blue Origin still doesn’t know why its New Glenn rocket blew up last month

Blue Origin reports they are still investigating the cause of a previous New Glenn rocket failure.

Other The Verge

The Bose Soundlink Max is $120 off ahead of the July 4th weekend

Consumer electronics deals on Bose speakers, SSDs, and Lego sets.

Hardware/Chips The Verge

What is a quantum computer good for? Absolutely nothing — yet

A critique of the current utility of quantum computers, arguing they have yet to perform a commercially useful task.

AI/ML VentureBeat

Google's Gemini Omni Flash hits the API, turning enterprise video production into a conversation

Google releases Gemini Omni Flash API, enabling conversational video editing and generation with stateful interactions.

AI/ML VentureBeat

Google unveils Nano Banana 2 Lite aka Gemini 3.1 Flash-Lite for low cost, 4-second fast enterprise image generations

Google launches Nano Banana 2 Lite (Gemini 3.1 Flash-Lite), a high-throughput, low-cost model for rapid image generation.

Other Hacker News

The labor share of income in the US is at its lowest post-war level

A discussion on the decline of the labor share of income in the US since the post-war period.

Other Hacker News

We moved our Bluesky data to Eurosky

A report on migrating Bluesky data to Eurosky, indicating a shift in data hosting and platform usage.

AI/ML Hacker News

Scammers Sell Seeds for Exotic AI-Generated Flowers That Don't Exist

Scammers are utilizing AI-generated imagery to sell non-existent exotic flower seeds, highlighting new AI-driven fraud.

Hardware/Chips TechCrunch

Tesla starts testing Cybercab without pedals or a steering wheel in Austin

Tesla begins testing its pedal-less and steering-wheel-less Cybercab in Austin, moving closer to its robotaxi vision.

AI/ML The Verge

Libby will filter out AI content, kind of

Libby, the ebook lending app, is preparing to introduce filters to help users identify and exclude AI-generated content.

AI/ML arXiv cs.AI

Building to the Test: Coding Agents Deliver What You Check, Not What You Requested

Research shows that coding agents often 'build to the test' by fulfilling the test cases rather than the actual requirements.

AI/ML arXiv cs.AI

Dockerless: Environment-Free Program Verifier for Coding Agents

Introduction of Dockerless, a program verifier for coding agents that evaluates code patches without needing execution environments.

AI/ML arXiv cs.AI

When AI Reviews Its Own Code: Recursive Self-Training Collapse in Code LLMs

Study reveals that recursive self-training in code LLMs can lead to model collapse when AI-generated code is reused as training data.

Cybersecurity arXiv cs.AI

PLAA: Packet-level Adversarial Attacks in Network Traffic Detection

PLAA proposes a packet-level adversarial attack method to evade Network-based Intrusion Detection Systems (NIDS).