- AInauten.net
- Posts
- 👀 Weekly AI news: Did you miss it?!
👀 Weekly AI news: Did you miss it?!
👨🚀 The most important AI updates at a glance
Die AInauten
June 22, 2025
👀 Weekly AI news: Did you miss it?!
👨🚀 The most important AI updates at a glance
AIHOY and happy weekend, dear AInauts!
Maybe you didn't catch all the exciting news, tools and hacks about AI last week, or maybe you've just joined us. No need to worry, we've got you!
We have summarized our highlights of the past week for you:
Finally, there is the most important quick news for you - so you can keep an eye on all relevant developments with just one click! Ready? Let's go!
The new Genspark browser simply makes human-computer interaction more fun with an integrated "super agent". This AI-supported browser works actively in the background and can independently research, create presentations, analyze emails, generate tables and much more.
The browser is based on privacy-first principles, automatically blocks advertising and uses Brave Search. With a competitive price of $25 per month, Genspark competes against the likes of Google, Microsoft and OpenAI. We think: Definitely try it out!
Meta grabs the most coveted data pipeline in the industry with a 49% stake in Scale AI.
founder Alexandr Wang reports directly to Zuck and builds a superintelligence lab with GPU flat rate. Yann LeCun's open source mission is wobbling, while Google, OpenAI & Co. pull the plug and are now feverishly searching for new data partners. Conclusion: Meta wants to ignite the AGI turbo - and is giving us an exciting reality show.
Prompt injection is like a digital Trojan horse for AI systems! Malicious actors can insert hidden commands into seemingly harmless texts, which are then executed unnoticed by AI systems. Modern chatbots with "memory" are particularly dangerous - they can memorize manipulated information and give false answers later.
The risk increases with the increasing integration of AI tools into company systems. The best protection? Assign minimal access rights and always confirm critical actions manually. After all, nobody wants the company's own AI to suddenly send sensitive data!
The new AI chatbot "Underlord" from Descript makes video editing easier than ever with simple voice commands. Instead of spending hours clicking around in complicated menus, you can simply describe your wishes - from removing filler words to audio optimization and automatic teaser creation.
The tool processes natural language input and makes changes in minutes. Although still in beta, Underlord already shows impressive capabilities and makes professional video editing more accessible than ever.
A Chinese livestreamer has caused a sensation with AI avatars! In just 6 hours, two digital hosts, powered by Baidu's ERNIE model, achieved a turnover of 7 million dollars. With 13 million viewers and 133 products sold, the AI avatars even outperformed their human counterparts.
The technology is rapidly establishing itself in Chinese e-commerce: over 100,000 digital sellers are already in use, reducing costs by 80% and increasing transactions by 62%. Platforms such as HeyGen make it possible for anyone to create their own AI avatars.
📰 AI news quickie: HAI highlights from the industry
AI never sleeps! Here is the most important news of the last few days - once again a whole host of exciting topics - served up on a silver platter, enjoy!
OpenAI in focus: innovation, expansion and controversy
Tensions between OpenAI and Microsoft are reportedly rising as OpenAI vies for more control and Microsoft may even be willing to break off negotiations.
The "OpenAI Files" offer 50 pages of insights into Sam Altman's empire, its structure and inner workings.
OpenAI warns urgently: Upcoming AI models pose a higher risk for the development of biological weapons and could open up dangerous opportunities for unbridled amateurs.
At the same time, the company is preparing for advanced AI applications in biology and is focusing on collaborations to maximize benefits and minimize risks.
Sam Altman emphasizes that AI will require a "significant" amount of electricity, which could have a catastrophic environmental impact.
OpenAI is ending its collaboration with Scale AI following its deal with Meta, which puts the future of Scale AI in doubt after Google's departure.
ChatGPT has been given a new lease of life: image generation is now possible directly via WhatsApp (1-800-ChatGPT).
In addition, Search has been optimized for better query understanding and more comprehensive answers, and Projects now also have the Deep Research feature, voice mode, better memory and more mobile options.
OpenAI has shut down the GPT-4.5 API as of July 14, 2025, while ChatGPT can continue to use the model.
OpenAI is changing the pricing structure for ChatGPT Enterprise and introducing a credit system with discounts of 10% to 20%.
According to Sam Altman, Meta has not poached any top talent from OpenAI so far, although Meta has reportedly offered 100 million dollars in bonuses!
Google & Gemini: Multimodal AI advances and new video dimensions
Canva integrates Google's Veo 3, and YouTube also integrates Veo 3 into Shorts to expand creative freedom. Gemini Pro and Ultra users can also create video magic via Prompt.
Incidentally, Google uses its YouTube videos to train its AI models - often without comprehensive information from the video creator...
The Google Gemini 2.5 model family has been expanded to include Flash, Pro and the affordable Flash Lite for top performance at low cost.
Gemini Code Assist has been improved with Gemini 2.5 and offers optimized coding and simplified context management - free to test.
Google's "AI Mode" now enables fluid voice conversations in real time, and with the new Search Live feature, users can ask questions by voice in the Google app.
Google is working on more features for Gemini Live, including a compact overlay that could give you a faster wireless experience.
In Google AI's Release Notes podcast, experts discuss Gemini's coding capabilities and the future of programming.
Meta: Between virtual reality, AI models and talent competition
Alexandr Wang, founder of Scale AI, joins Meta after Meta has invested over 29 billion US dollars in Scale. Scale AI competitors such as Snorkel AI report orders worth "tens of millions of dollars".
Meta has unveiled the impressive V-JEPA 2 model, which achieves strong performance in physical reasoning and enables robots to plan and interact with the real world.
Meta expands Advantage+ and introduces new AI tools for advertisers to boost campaign performance and creative opportunities.
Meta has unveiled the first cohort of the Llama Startup Program, which supports startups across multiple industries.
Meta hints at a new collaboration with Oakley to bring a sports smart glass version to market.
Meta's Llama 3.1 can reproduce 42 percent of the first Harry Potter book, which is legally problematic...
Curious conversations are held on Meta's AI platform…
The next generation of AI models and platforms
Midjourney has launched its first AI video generation model, V1, which creates moving scenes and up to four videos from images.
Groq is now integrated as an inference provider on the Hugging Face platform, providing lightning-fast responses for models such as Meta's LLama 4 and Qwen's QWQ-32B.
ByteDance presents Seedance 1.0: The new video model can generate seamless multi-shot videos from text and images!
KREA AI introduces the new video model Hailuo 02, which impresses with breathtaking movements and great consistency from image to video.
Hunyuan 3D 2.1 is launched: It is the first fully open-source PBR 3D model with cinematic quality that makes materials such as leather and bronze shine in a new light.
MiniMax-M1 causes a sensation: a new, open source model with a context length of one million tokens, developed for efficient learning.
Magistral, the first thinking model from MistralAI, offers outstanding domain-specific and multilingual thinking as well as transparency.
Sakana AI introduces Text-to-LoRA: A hyper-network technology that generates task-specific LoRA adapters from text descriptions.
Qinsheng Zhang has introduced Cosmos-Predict2, a powerful open video model for Physical AI.
ElevenLabs revolutionizes control over voice output with new version 3 audio tags, bringing more vibrancy to AI voices.
MCP ecosystem on the rise: new standards for AI applications
ChatGPT can now access enterprise data via MCP servers, but OpenAI warns against handling sensitive information.
Anthropic pushes MCP integrations: The company offers a new course on building rich AI apps with the Model Context Protocol (MCP) and has announced MCP integrations that provide a flexible interface for developers.
Claude Code now supports remote MCP servers, enabling the integration of systems and tools without local server management.
Figma's new Dev Mode MCP server brings design and development together by helping AI tools generate design-informed code.
MonetizedMCP brings programming solutions to MCP servers so they can generate payments without changing the core protocol.
AI tools for development, security and automation
MiniMax has introduced the MiniMax Agent - an intelligent agent for complex tasks that acts as a trusted teammate.
VibeSec provides an AI-native platform to instantly identify and fix real vulnerabilities in code.
Leverage @snyksec's AI-powered secure code review for quick and easy click-and-scan without a build.
flowith 2.0 takes the concept of the typical AI workspace to a new level and offers creative freedom with intelligent support.
Cursor launches the new Ultra plan for power users with 20x more usage than the Pro plan for 200 US dollars per month.
rabbit has launched its second product, rabbit intern, an all-in-one AI agent for website creation, reports and presentations.
NVIDIA has released NVIDIA Isaac Sim 5.0 and Isaac Lab 2.2 for early preview on Github, for robot simulation in real-world environments.
AI is transforming industries: Advertising, media, government and more
Leaked government AI plans from the Trump administration (AI.gov as a one-stop shop) briefly surfaced on GitHub and were also confirmed on X.
The advertising industry is facing disruption as tech giants push AI tools for advertising and threaten traditional agency models.
Adobe launches LLM Optimizer to optimize viewability in generative AI environments.
Reddit introduces tools that place positive user posts under ads and presented "Reddit Community Intelligence™" at Cannes Lions 2025.
TikTok integrates its AI tools into Adobe Express and WPP Open for video creation from images and text.
The Italian newspaper Il Foglio experimented with AI-generated articles for a month and is now planning to integrate them permanently.
The UK government uses Gemini to speed up the processing of over 350,000 planning applications per year.
NVIDIA and Deutsche Telekom form a powerful partnership for a European AI factory.
Adobe has released the Firefly app for iPhone and Android, ideal for mobile working with AI-generated images and videos.
According to the ifo Business Survey, the number of companies using AI jumps from 27% to 40.9%.
A new AI algorithm for analyzing existing CT scans can identify hidden heart disease.
The new AI-generated tags for the App Store are live in the developer beta of iOS 26, but are monitored by humans before going live.
Product placement is now possible with HeyGen: upload a product photo, select an avatar, insert a script - and your ad is ready to go.
An Israeli vibe coder sells his six-month-old startup Base44 to Wix for 80 million dollars; the platform has 250,000 users.
Zoox has opened its first robotaxi plant in California to produce 10,000 cars a year.
Regulation, ethics and social impact
New York has passed a law to prevent catastrophic AI scenarios.
California publishes a long-awaited report on AI safety that criticizes companies for insufficient risk identification.
Pope Leo XIV massively addresses the threat of AI and demands responsibility from the tech industry. Read more in TechCrunch.
Big Tech fights for states not to be able to regulate AI for ten years to ensure their dominance.
Experts and observers intensively discuss the impact of AI on the world of work: Eric Jing gives tips on how to deal with AI in working life, Scott Galloway compares the change to industrialization, and Amazon CEO Andy Jassy announces that the workforce will shrink due to new AI tools , which could encourage other bosses to speak more openly.
A Stanford survey of workers and experts shows that the "AI takeover" is going in the wrong direction - companies are also realizing that the overblown AI hype is not living up to expectations.
Nvidia CEO Jensen Huang contradicts warnings from Anthropic boss Dario Amodei about job losses due to AI.
The use of AI could increase the energy demand from data centers to up to 12% of electricity consumption by 2028 and cause more CO2 emissions.
Calls to stop developing AI therapy chatbots are growing louder as their lack of competence could jeopardize mental health.
Research and the "brain" of the machines
Anthropic develops the SHADE arena to monitor LLM agents and detect the possibility of sabotage.
Discover the secret (leaked) system prompts that determine the behavior of leading AI models.
Punctuation influences the responses of chatbots: Correct punctuation leads to structured-analytical responses, incorrect writing leads to casual responses. However, the effectiveness without punctuation does not decrease.
A preliminary study suggests that less external support from LLMs such as ChatGPT leads to better cognitive effects when writing. Another study shows that ChatGPT can reduce critical thinking as users show lower brain engagement.
Another study contradicts Apple's claims about the "reasoning collapse" of LLMs.
AI recognizes your emotions better than you do by analyzing facial expressions and embedding them in data patterns!
And that’s a wrap! Enjoy the rest of your weekend.
Your AInauts, And that’s a wrap!
Fabian & Reto
Your feedback is essential for us. We read EVERY comment and feedback, just respond to this email. Tell us what was (not) good and what is interesting for YOU