All Updates
Page 140 of 1443
June 16, 2026
Synthetic document finetuning for instilling positive traits
Google DeepMind researchers demonstrate a method to instill specific values in Gemini 3 Flash by combining midtraining on synthetic documents with chat-based SFT. This approach aims to improve model alignment and robustness, even when handling out-of-distribution scenarios.
Predicting AI model behavior via deployment simulation
OpenAI introduces Deployment Simulation, a new methodology that uses real conversation data to forecast model behavior prior to public release. This approach aims to enhance safety protocols and improve the accuracy of model evaluations.
June 15, 2026
Sundar Pichai faces protests over Google's defense AI ties
Sundar Pichai faced protests and walkouts at a Stanford graduation ceremony. The disruption was driven by concerns over Google's AI contracts with defense and government agencies.
Canada introduces new legislation to curb data-driven price discrimination
The Canadian government has introduced Bill C-36 to replace outdated privacy laws. The legislation aims to restrict companies from using personal data to implement dynamic pricing strategies that charge consumers more.
New Neocortex Learning Framework Challenges Backpropagation
A new research paper proposes a biologically-inspired learning framework that approximates human-level intelligence using corticothalamic circuits. Implemented in the Axon simulation framework, this approach aims to surpass backpropagation in efficiency and scalability.
Xiaomi unveils autonomous robotic EV charging arm
Xiaomi has developed a residential robotic arm that automatically connects and disconnects EVs from chargers. The system uses computer vision to detect vehicle positioning and manages the charging process without human intervention.
Rivian to launch supervised point-to-point self-driving
Rivian CEO RJ Scaringe announced that supervised point-to-point self-driving will roll out to second-gen vehicles and the R2 later this year. The system is positioned as a direct competitor to Tesla's FSD.
IMEC unveils 325GHz chip platform for 6G
IMEC has expanded its 300mm RF silicon interposer platform to support frequencies up to 325GHz. This advancement aims to make 6G hardware production commercially viable by reducing signal loss.
Hong Kong faces $274B IPO lock-up expiry wave
Hong Kong's market is bracing for a massive influx of shares as lock-up periods for recent IPOs, including tech firms like MiniMax Group and Knowledge Atlas Technology, expire. Goldman Sachs estimates that up to US$274 billion in shares could hit the market over the next 12 months.
Sakura Internet and TEPCO Discuss AI Power Demand
Sakura Internet CEO Kunihiro Tanaka and TEPCO's Hiroshi Okamoto discussed the critical challenge of meeting the surging power demands of AI data centers. The dialogue focused on the potential of 'Watt-Bit' integration to optimize energy consumption for AI infrastructure.
Xbox studios negotiate independence amid Microsoft restructuring
Compulsion Games, Double Fine Productions, and Ninja Theory are in talks to buy themselves back from Microsoft. These studios are seeking independence to avoid potential closures resulting from broader gaming division restructuring.
AI dating assistants may sabotage your romantic life
A new study indicates that using AI to generate dating app messages creates a significant disconnect between a user's digital persona and their authentic self. This gap often leads to negative outcomes in real-world relationship building.
SMBC Invests 50 Billion Yen in Generative AI Strategy
Sumitomo Mitsui Financial Group is pivoting its organizational culture to prioritize innovation, backed by a 50 billion yen investment in generative AI. The company emphasizes that successful digital transformation requires a willingness to abandon failing projects.
Osaka Gas Cuts 2000 Hours Using Notion AI
Osaka Gas and another company successfully reduced information search time by 2,000 hours per month using Notion and AI. The project focuses on converting buried internal data into accessible knowledge assets to prevent knowledge silos.
Exploring Generative AI in 3D CAD Workflows
This article explores the integration of generative AI into 3D CAD, specifically using Autodesk Assistant. It demonstrates how natural language prompts can generate initial 3D models, highlighting both potential and current limitations.
Google Workspace feature exploited for data exfiltration
A China-linked espionage group has been using a built-in Google Workspace feature to exfiltrate sensitive data from research and military networks. The attackers reconfigured workspace rules to automatically copy and forward internal messages.
US government bans Anthropic models over security concerns
The Trump administration forced Anthropic to pull its latest cybersecurity models. This move signals increasing government interference in the AI industry, moving beyond simple jailbreak concerns.
Android 17: Gemini Intelligence and New Features Preview
Android 17 is expected to integrate deeper AI capabilities through Gemini Intelligence. The update will likely feature UI enhancements like app bubbles and advanced 3D emoji support.
Cleo: Fitting full analyst behavior into a 2B model
Cleo is an open-source, fine-tuned Qwen3.5-2B model designed to perform complex analyst tasks like text-to-SQL. It utilizes a unified harness to handle training, inference, and SQL safety, proving that small models can achieve high-level reasoning.
Meta CTO Admits AI Reorganization Was βAtrociousβ
Meta CTO Andrew Bosworth acknowledged that the company's recent AI-focused reorganization caused significant internal friction. He is now prioritizing stability and improved communication to boost employee morale.