- AInauten.net
- Posts
- π Weekly AI news: Did you miss it?!
π Weekly AI news: Did you miss it?!
π¨βπ The most important AI updates at a glance
Die AInauten
April 26, 2025
π Weekly AI news: Did you miss it?!
π¨βπ 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 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!
ChatGPT's new memory feature now permanently stores all conversations and builds a detailed profile of your mind. A double-edged sword: on the one hand, it enables personalized learning, better coaching and more efficient use. On the other hand, there are legitimate concerns about data protection, surveillance and manipulation! AI becomes an intimate digital companion that combines diary, confessional and coach.
According to the Harvard Business Review, the emotional use of AI surprisingly dominates, especially for therapy and emotional support! People mainly use AI systems to strengthen their human qualities - not to become more machine-like. And: the latest studies show that AI now has a higher IQ than the average human...
A massive leak on GitHub revealed the system prompts of leading AI development tools such as Cursor, Lovable, Replit, Manus, etc. These secret instructions that control the behavior of AI tools provide valuable insights into effective prompting.
The most important lessons learned: clear structuring of your prompts with a defined mission and boundaries, modular thinking when creating prompts, use of semantic structures, balance between positive and negative instructions and well thought-out troubleshooting strategies. A hierarchical approach is also recommended for custom GPTs: Role β Capabilities β Rules β Special cases β Functions.
Google's free AI tool NotebookLM has received an ingenious update! The tool, known for its ability to process complex documents, is now introducing a "Discover" function. This allows users to automatically find and integrate relevant sources from the internet - from websites to YouTube videos.
The tool continues to shine with its core functions: Source processing, detailed chats with source references, automatic note generation and the popular audio feature that prepares content as a podcast conversation. Perfect for studying, analyzing documents and in-depth research - and it's all free!
In a fascinating 13-minute interview on 60 Minutes, Google DeepMind CEO Demis Hassabis gives insights into the future of AI. The focus is on groundbreaking developments in healthcare and the innovative Astra project, which aims to better understand the world. Exciting advances in robotics are also discussed.
π° AI news quickie: HAI highlights from the industry
AI never sleeps! Here is the most important news from the last few days.
OpenAI in focus: models, strategy & controversies
OpenAI now brings ChatGPT's image generation function to the API! Apps from Adobe, Microsoft and many others are already using it.
OpenAI has also extended the usage limits for ChatGPT Plus: Now you get 100 messages per week for o3 and 300 daily for o4-mini. Deep Research "light" has also been released.
The new o3 model marks a breakthrough in the world of AI. It now uses agentic intelligence to independently integrate and use over 600 different tools for complex tasks. However, the o-Series models (o3 and o4-mini) showed mixed results in the ARC-AGI test.
The GPT-4.1 model, which is also new, proved to be less reliable in tests than its predecessors. OpenAI will soon switch off the API for GPT-4.5 in favor of 4.1. By the way, this is the best way to prompt the new models.
o3 and o4-mini can use "Reverse Location Search" to determine locations in photos - a potential data protection problem.
ChatGPT search is booming in Europe! According to OpenAI data, the number of active users shot up to 41.3 million in six months - around 270% growth!
OpenAI has introduced Flex Processing, a new API feature. The costs are halved, but slower response times are accepted in return.
OpenAI and the Washington Post have agreed on a partnership: ChatGPT will now summarize and link to Post articles.
OpenAI wanted to buy Cursor, but is now in talks with Windsurf and has made a 3 billion dollar offer. Windsurf has also lowered its prices (team starting at 30 USD) to compete with Cursor.
Legal and AI experts have spoken out against OpenAI's restructuring, which would transfer control of AGI to a for-profit company. One part revolutionary AI lab, one part typical tech company looking to generate revenue.
OpenAI could adjust its security requirements if competitors publish risky AI without protective measures, but only after thorough examination.
Will there soon be a social network to compete with Meta and X?
Google & Gemini: catching up in the AI race
According to Google's DeepMind, AI has outgrown human knowledge. With the new "Streams" approach, AI models are to learn from experience and develop autonomously.
The Gemini Assistant has 350 million monthly users, a strong growth compared to the previous year, which was revealed in a court case. ChatGPT, however, still has more than twice as many users.
Despite Google's claims, an Ahrefs study shows that AI Overviews generate 34.5% fewer clicks. Meanwhile, the EU Commission is investigating the impact on the advertising business and publishers.
Google Gemini is now testing video uploads and allows you to ask questions about your uploaded videos.
The new Google Gemini 2.5 Pro for complex tasks offers captivating features to spice up creative routines.
Google is reportedly paying Samsung huge sums to pre-install the Gemini AI app. Samsung is also integrating Google's Gemini into its home robot Ballie for outfit suggestions and health tips.
Anthropic: Claude, security & growth
Anthropic warns: Fully autonomous AI employees, equipped with memories and corporate passwords (details from the head of security), could be a reality in a year - a nightmare for cybersecurity teams.
Anthropic analyzed 700,000 of Claude's conversations and found a fascinating morphological value system: from "self-reliance" to "strategic thinking".
It also revealed best practices and tips for Claude Code, including the magic word "ultrathink" for maximum thinking power.
Anthropic has formed a new team to win AWS customers for its AI products and drive its growth strategy forward.
Robotics: progress and challenges
At the first half marathon in Beijing, in which 21 humanoid robots competed against human runners, the Tiangong Ultra model was the first robot to cross the finish line after a few falls, while only six machines finished in total. The cute "0306 Little Giant" also took part.
Top venture capitalist Zhu Xiaohu warns: China's humanoid robotics industry is overheated and could be caught up in the hype without a clear customer base.
A robot now cleans kitchens and bedrooms in homes that were never part of its training! With the new Ο-0.5 model, this is becoming a reality!
Ukraine plans to send an impressive 15,000 combat robots to the front line by 2025 to replace human soldiers.
AI investments & company news
Just seven months after its last round, Supabase secures another $200 million at a $2 billion valuation, thanks to Vibe Coding.
Manychat has secured 140 million US dollars to expand its AI-powered messaging service.
Noxtua has also closed a hefty financing round of 92 million euros to develop an AI for the German legal system.
Endor Labs raises 93 million US dollars for tools to check the security of AI-generated code.
The German translation company DeepL is apparently looking at the best time for an IPO - possibly this year or next.
BMW is integrating DeepSeek's AI into new car models in China.
Johnson & Johnson has found that 10-15% of AI use cases deliver 80% of the value. In other words: focus on such projects becomes the strategy.
Revolution in sound & speech: TTS and more
Two students have presented a new open source speech model: Dia (details, code, demo comparison)! Realistic dialog, but only in English.
With Orpheus TTS, Canopy Labs presents an advanced text-to-speech model that can express human intonation and emotion.
YouTube celebrates its 20th anniversary and plans an AI boost: videos will soon be automatically translated into all languages. And YouTube Music is introducing the long-awaited "constant volume" function.
AI tools for productivity & everyday life
Microsoft launches powerful new AI agents ("Researcher", "Analyst") with Microsoft 365 Copilot and integrates OpenAI's Ghibli image generator to revolutionize everyday work.
The Fireflies AI app can now be connected to many peripheral systems - automate everything from emails to meetings and get accurate summaries.
Hence Technologies helps to survive in a chaotic world and brings Hence Global, an AI advisor that supports with geopolitical risks.
With Websync, you can now import multiple pages or entire websites for NotebookLM with one click (7-day free trial).
Lovable 2.0 now has an intelligent chat agent, enables multiplayer collaboration and offers new security features.
Ray-Ban Meta glasses are now available with AI assistants in Germany for live translations.
Cybersecurity & AI risks: New threats
The EU Commission has immediately banned AI agents from online meetings!
Cybercrime is booming thanks to AI! Scammers use hyper-personalized attacks, deepfakes and lightning-fast fake deals.
Microsoft Defender observes new scams: Attackers are abusing Node.js to deliver malware, leading to data loss.
A new study from the Centre for Emerging Technology and Security shows how AI is helping criminals and threatening our online world.
"Vibe coding" with AI offers convenience, but could open security gaps.
Researchers have cataloged 35 "jailbreak" techniques used to trick AI chatbots. Policy Puppetry can outsmart them all.
Suspended Columbia student Roy Lee has raised 5.3 million dollars for his startup Cluely, which develops AI for "cheating".
Ethics, regulation & social impact
Palmer Luckey warns: The "Pandora's Box" of AI weapons is open. Investors should prepare for revolutionary developments.
The United Arab Emirates (UAE) is driving innovation with the first AI-powered regulatory intelligence ecosystem to write laws and revise regulations, the UAE Cabinet has decided, to speed up and clarify legislation.
Actors like Adam Coy and Simon Lee are struggling with the consequences of their decision to license faces for AI videos.
AI adoption is growing in healthcare, but many projects are failing due to security and data issues: only 30% are in production.
Character.AI has presented the video generation model AvatarFX, which enables lifelike chatbots.
An AI has written an entire Italian newspaper format for a month. Conclusion: the future belongs to journalists!
The Oscars take AI seriously: films with AI can win, but human creativity remains crucial.
A new benchmark test shows that AI now outperforms human experts in virology, raising questions about dual-use security.
AI agents with their own crypto wallets automate payments and reduce errors.
One user has developed a DeepSeek AI trading bot that makes huge profits.
And thatβs a wrap! Enjoy the rest of your weekend.
Your AInauts, And thatβs 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