MiniMax Agent new function upgrade
https://www.aibase.com/news/18884
Image generation upgrade: Supports generating higher quality and more detailed images with natural language, suitable for creative design and content production, outperforming some competitors. Reflection Mode: Can understand the context and "reflect" in long tasks, with more coherent output, suitable for project planning and multi-step analysis. Multi-language support: Added Chinese, Japanese, and Korean display, especially in Python charts to optimize Asian user experience. Document export function: Generate PDF, Word, and PPT files with one click, which is convenient for quick sharing and reporting, and improves work efficiency. This upgrade marks a solid step for MiniMax Agent towards the "all-round AI productivity platform", from multimodal capabilities, document export to localized experience, demonstrating the user-centric design concept and ambition to expand globalization.
OpenAI upgrades ChatGPT Projects
https://help.openai.com/en/articles/10169521-using-projects-in-chatgpt
With support for creating projects, adding PDF context, and organizing chats, users can create unlimited projects, each supporting up to 20 files, but subject to rate limits. Projects are smart workspaces that centralize all content related to long-term work. You can group chats, upload reference files, and add custom notes so that ChatGPT remembers important content and keeps the topic consistent. With memory functions, context, and flexible tools, projects are perfect for repetitive and ever-changing work such as writing, research, and planning. Currently* available to ChatGPT Plus, Pro, and Team users worldwide. *
Meta's V-JEPA2 model
https://about.fb.com/news/2025/06/our-new-model-helps-ai-think-before-it-acts/
V-JEPA 2 is Meta's most advanced "world model" AI, which is extensively trained on video data to enable robots and AI agents to understand, predict, and plan actions in the physical world like humans. Core capabilities include: First Understand Real-World Scenes: Learn semantic features and object dynamics in videos by abstracting them into meaningful embeddings. Second Predict Future States: The prediction module uses the current embeddings (even hypothetical actions) to predict what will happen next - just like predicting the trajectory of a ball. Third Planning/Zero-Shot Control: *By simulating the results of actions in its learned world model, it can *plan and execute physical tasks without additional training, effectively enabling robots to manipulate unseen objects or navigate in unfamiliar environments. Fourth Video Q&A: Supports linking to language models to explain, answer queries, or describe video sequences from the perspective of causality, physics, and action intent.
AMD and OpenAI jointly release powerful AI chips: 35 times the reasoning performance
AMD and OpenAI launched the latest Instinct MI400 and MI350 series AI chips. The MI350 series significantly improves AI computing performance, and the MI400 series is aimed at the next generation of flagship AI computing needs. In addition, the ROCm7 platform further promotes the efficiency of AI developers.
Imagen4 lands on Gemini
https://deepmind.google/models/imagen/
Google’s latest and most advanced text-to-image generation model Imagen 4 has been officially integrated into the Gemini AI platform, marking a major upgrade in Gemini’s image generation capabilities . The integration was announced around Google I/O 2025 and is available to all Gemini users, including free and premium subscription users.
NBA Finals Shows AI Ads
https://x.com/Kalshi/status/1932891608388681791
Kalshi hired Google's text-to-video generator Veo 3 to create the ads. Millions of viewers watching the NBA Finals were shown a ridiculous AI-generated ad—and it only cost $2,000 to make. The ad features a shirtless old man wearing an American flag, a farmer floating in an inflatable pool full of eggs, an alien drinking a pot of beer, and a woman wearing a shiny pink tracksuit driving a Zamboni ice car.
Mark Zuckerberg is personally out to grab AI talent!
Zuckerberg is hunting top AI talent from giants such as Google, DeepMind, and OpenAI through cold emails and WhatsApp one-on-one private chats. He promises these researchers: extremely high autonomy, Meta's huge product platform support, massive infrastructure investment, and "sky-high" salary packages of up to eight or even nine figures!
ModelOp 2025 AI Governance Benchmark Report
More than 80% of Fortune 500 companies have more than 51 generative AI project proposals. But only 18% have successfully put more than 20 models into production. Most projects take 6 to 18 months from conception to launch. Companies face a serious "execution gap", resulting in delayed return on investment, reduced internal confidence, and dissatisfied stakeholders.
The UK is working with companies like NVIDIA to address the skills gap
The UK is strengthening its position as Europe's AI powerhouse by working with companies like NVIDIA to address issues such as the skills gap.
The UK government signed a memorandum of understanding (MoU) with NVIDIA to jointly promote AI talent training across the country, covering secondary schools, universities, working personnel and civil servants, and plans to vigorously enhance AI capabilities in the next five years. NVIDIA strongly supports the UK government's national skills promotion plan and plans to establish a dedicated AI technology center in the UK. British soil. The center promises to provide practical training in artificial intelligence, data science, and the increasingly important field of accelerated computing.
British teachers are allowed to use artificial intelligence
According to the BBC, British teachers have been allowed to use artificial intelligence to help them complete low-level tasks that fall within their responsibilities.
Guidance from the UK Department of Education (DfE) states that school teachers in England can use artificial intelligence, but it can only be used for "low-risk" tasks, such as writing letters to parents and marking homework.
New York passes a bill to prevent AI-fueled disasters
New York lawmakers passed a bill Thursday that aims to prevent cutting-edge AI models from OpenAI, Google, and Anthropic from causing disaster scenarios, including more than 100 deaths or injuries, or more than $1 billion in losses.
Behind ANCESTRA: Veo meets live-action filmmaking
https://blog.google/technology/google-deepmind/ancestra-behind-the-scenes/
Written and directed by Eliza McNitt, the film was produced in partnership with Primordial Soup, a studio founded by Darren Aronofsky, and Google DeepMind. The film premiered at the Tribeca Film Festival on June 13, 2025, telling the story of a mother's cosmic love and redemption when a baby with a heart defect is born. Combining real-life shooting with DeepMind's video generation model Veo, a more imaginative visual expression is achieved.
Midjourney has finally released its long-awaited AI video generator.
https://generativeai.pub/midjourney-video-generator-is-almost-here-c40f8a5f034f
The initial version of the tool will be called the "V1 video model" and will be officially released after Midjourney completes some finishing touches.
Currently, the video model is not accessible through the Midjourney website or even Discord. The only way to view sample videos is through the video rating group.
Guardian investigation finds nearly 7,000 confirmed cases of cheating
Thousands of university studentsUK Many researchers have been caught abusing ChatGPT and other AI tools in recent years, while traditional forms of plagiarism have declined significantly, according to a Guardian investigation. The survey on academic integrity violations found that nearly 7,000 cases of cheating using AI tools were confirmed in the 2023-24 academic year, equivalent to 5.1 cases per 1,000 students. This is up from 1.6 cases per 1,000 students in the 2022-23 academic year.