OpenAI Launches ChatGPT Record for Real-Time Transcription
OpenAI released "ChatGPT Record" exclusively for macOS users, enabling 120-minute audio recording, real-time transcription, and structured summarization. Data is auto-deleted post-processing, addressing privacy concerns for meetings and interviews.
Midjourney V1 Video Model Debuts at $10/Month
Midjourney launched its first video generation model V1, producing 20-second clips per prompt. Priced 25x cheaper than competitors, it supports image-to-video conversion with cinematic quality, challenging Runway and Sora.
Meta Finalizes $14.3B Scale AI Acquisition
Meta completed its acquisition of 49% Scale AI shares, appointing 28-year-old founder Alexandr Wang to its board. The deal aims to enhance Meta's data infrastructure for training multimodal models.
Google Gemini 2.5 Flash-Lite Rolls Out Globally
Google's lightweight Gemini 2.5 Flash-Lite entered stable release, optimizing cost-efficiency for developers. It supports rapid inference in translation and coding tasks, adopted by Snap and Spline for production environments.
MIT AI Cuts Cement Emissions by 1.2B Tons
MIT researchers used LLMs to screen 14,000 cement alternatives from 88,000 studies, identifying 25 sustainable materials. This reduces reliance on scarce fly ash, potentially slashing 12B tons of CO? by 2050.
Salesforce Open-Sources MAS-Zero Multi-Agent Framework
Salesforce released MAS-Zero, an autonomous multi-agent design system. It dynamically configures agent teams for complex tasks like code generation and math reasoning, achieving zero human supervision.
Tsinghua's Sundial Model Revolutionizes Time-Series Forecasting
Tsinghua University launched Sundial, a generative time-series foundation model. Trained on trillion-point dataset TimeBench, it performs zero-shot predictions 35x faster than Chronos in finance and weather domains.
EU May Delay AI Act Compliance Deadline
Sources indicate the EU might postpone its August 2 AI Act compliance deadline due to unresolved technical standards. High-risk AI systems face strict validation requirements, pressuring tech firms.
Multitalk Generates Multi-Person Dialogue Videos from Audio
Jointly developed by Meituan and HKUST, Multitalk syncs multi-channel audio to generate realistic conversational videos. Its L-ROPE technique prevents speaker mismatches across 20 languages.
MIT Study: ChatGPT Erodes Critical Thinking by 34%
MIT research confirmed ChatGPT users show reduced problem-solving initiative in controlled tests. Lead researcher Cosmina Urson advocates analog training mandates for students.
Anthropic's Constitutional AI Adopted by 15 EU Hospitals
European medical centers deployed Anthropic's ethically constrained AI for diagnostics. Early data shows 99.1% compliance with EU medical guidelines, reducing misdiagnosis risks.
Deep Render Cuts Video Bandwidth 90% with AI Compression
London-based Deep Render launched neural video compression, reconstructing HD footage from 10% data packets. Streaming giants are testing it to reduce CDN costs significantly.
Adobe Integrates Sora into Premiere Pro for AI B-Roll
Adobe added OpenAI's Sora plugin to Premiere Pro, allowing editors to generate contextual B-roll via text prompts. The tool auto-matches lighting and camera angles from existing footage.
UN Reports 780% Surge in AI-Generated Child Abuse Material
A UN document revealed 2.3M AI-generated child exploitation images in Q2 2025. INTERPOL is developing forensic tools to trace diffusion model fingerprints.
Quantum Startup Quaise Secures €200M for Neuromorphic Chips
Paris-based Quaise raised €200M for brain-inspired processors mimicking synaptic plasticity. Benchmarks show 40x energy efficiency gains on spiking neural networks.
Pope Warns AI Threatens Youth Development
Pope Leo XIV cautioned that AI risks impairing children's intellectual and spiritual growth by replacing authentic wisdom with data consumption, urging tech safeguards.
See More Content about AI NEWS