- 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
February 01, 2025
👀 Weekly AI news: Did you miss it?!
👨🚀 The most important AI updates at a glance
AI-HOY 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, 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!
P.S. This year we have big plans and goals! That's why we've also given our design a little facelift - we hope you like it.
OpenAI's new Operator is a semi-autonomous agent with browser access that can perform complex web tasks. In tests, it demonstrated impressive capabilities, such as winning at online poker and performing searches autonomously. The agent is powered by a special computer-using agent model and has GPT-4 vision and advanced thinking skills.
The operator is currently only available for OpenAI Pro users in the USA. It clearly outperforms comparable tools in terms of functionality and user-friendliness. Use cases such as automated product management tasks or the control of other AI tools are particularly interesting.
The AI competition reaches new dimensions! DeepSeek from China is causing a sensation with a reasoning model that rivals OpenAI's o1 - for just 5% of the cost. As an open source solution, it is conquering the app stores and even runs locally on smartphones.
OpenAI counters with o3 mini and more free queries, while xAI announces Grok 3 and Meta plans a Manhattan-sized data center for Llama 4. Google prepares Gemini 2 Pro, and Anthropic predicts revolutionary breakthroughs in life extension. The EU? It's still pondering.
DeepSeek's new R1 model caused a mini stock market crash: NVIDIA lost 17% (500 billion market value) in one day! The Chinese AI model achieves similar performance to OpenAI's o1 with only 5% of the development costs - despite the US chip embargo. The DeepSeek app even overtakes ChatGPT in the App Store.
This efficiency demo shows: More computing power and money alone won't win the AI arms race. Experts see the development as positive: more efficient AI models make the technology more accessible. For big tech companies such as Apple and Meta, cheaper models could even be advantageous.
An exciting update for ChatGPT's Canvas feature now enables the integration of the o1 model! This means that live previews for code visualization are finally a reality. ChatGPT thus remains the most used AI app and proves its supremacy with continuous feature updates.
In a practical test at the beginning of the year for funds and ETFs, the AI not only shows competence, but also a refreshing sense of humor. And don't worry about data security - the CIA already has everything anyway! 😉
Firecrawl is revolutionizing web scraping with its new /extract function. Unlike technically complex scrapers, you simply describe in natural language what data you want to extract from a website - Firecrawl does the rest automatically.
Perfect for use cases such as AI chatbot training with website content, market analyses or rating evaluations. The tool independently creates schemas, collects the desired data and exports it in a usable format. Particularly practical: the API interface and Zapier integration enable seamless automation.
📰 AI news quickie: HAI highlights from the industry
What a week! After DeepSeek dominated all the news, OpenAI launched the long-awaited o3 mini model at the end of the week, which we will report on in more detail next week. Here's all the important news from the last few days... and if you want to get a glimpse into the future, we highly recommend this podcast. ❤️
DeepSeek R1 - everything you need at the click of a button
The DeepSeek R1 model manages to perform better than OpenAI, Meta and the like for a fraction of the price. Meta has therefore put together four war rooms of engineers to find out how the architecture works in detail.
OpenAI CPO Kevin Weil describes DeepSeek as the first big bang in the US-China AI race.
The company's rise is emblematic of the race for AI leadership.
The US is investigating the national security risks of DeepSeek.
David Sacks claims there is substantial evidence that OpenAI's models were used for training, and Microsoft is looking into it. (But no one is talking about OpenAI appropriating the entire Internet to train GPT...)
Dario Amodei argues that export controls on chips to China are crucial to ensure US leadership in AI.
Parts of the DeepSeek database were accessible to anyone, with more than a million records - including user data and prompts.
Also, not only inputs are collected, but also other inputs on the device, and stored on servers in China - "free" always has its price.
Perplexity AI now offers DeepSeek as a reasoning option to its Pro users, hosted directly in the US.
Developers: DeepSeek R1 is now available in the Azure AI Foundry, and Nvidia launches the DeepSeek-R1 NIM microservice.
The most powerful thinking model, DeepSeek R1, is now available on Perplexity to help you with your deep web research! Look forward to increasing limits and updates soon!
DeepSeek R1 and V3 are now available in Windsurf, fully hosted on western servers. This means that R1 can be used as a coding agent.
KREA AI is also launching a new chat interface with Krea Chat powered by DeepSeek, find out more here.
New voice and video models
OpenAI's o3 mini is finally live, and GPT 4o has also been updated.
Mistral AI has introduced the 24B parameter model Mistral Small 3 - a fast, latency-optimized competitor to larger models such as Llama and GPT4o-mini! Ideal for fast, accurate responses!
The Allen Institute for AI has released its latest model Tülu 3 with 405 billion parameters, which is said to outperform DeepSeek v3 and GPT-4o.
Luma AI has now introduced the Upscale to 4K feature - transform your videos in stunning clarity up to 4K! A feast for the eyes!
Pika has introduced Turbo Mode: Generate stunning AI videos now three times faster with seven times less credits! Ready to try it out for yourself?
Hailuo AI presents the new T2V-01 Director model - now control your camera with natural language for perfect cinematic transitions! Bye-bye, shaker AI!
Oh, and ElevenLabs has closed a 250 million dollar financing round and is now valued at over 3 billion dollars…
Tools and stuff
Google Labs has exciting tools up its sleeve to be creative, learn and play with AI! See if there's something for you.
YuE 7B allows you to install it locally to create complete songs with music, genre and lyrics. Get ready to make your musician's heart beat faster!
Spell by Spline allows you to easily create 3D worlds - no programming skills required.
Hunyuan3D 2.0 from Tencent presents a new platform for creating high-resolution 3D assets.
Goose is an open source entry point that helps developers work more efficiently. With customizability and independence, it takes on complex tasks and is loved by engineering hearts!
UI-TARS reinvents automated GUI interaction with native agents for more efficient program flows.
AI in health = longevity
Retro Biosciences, backed by Sam Altman, raises 1 billion dollars to extend human life by 10 years!
Anthropic CEO Dario Amodei predicts a doubling of human lifespan through AI technologies by 2030.
A new AI technology from the Yale School of Medicine detects the risk of heart failure based on ECGs.
A new fluorescent protein, esmGFP, has been developed by AI and could eliminate the need for 500 million years of evolutionary adaptation. Scientists see potential for new drugs.
AI safety and robots
Experience the future up close! On the Future Timeline, you can browse future forecasts from 2025 to 3000+. The question is: will we survive the Singularity?
OpenAI can store data from deleted chats in Operator for up to 90 days to better monitor cases of abuse. That's 60 days longer than ChatGPT! Technically quite audacious, isn't it?
With the Lynx Robot Shaggy, DEEP Robotics demonstrates impressive skills in the snow parkour and when crossing streams. The Unitree H1 "Fuxi" also impresses with its capabilities.
Researchers warn that AI could potentially learn a dangerous deception that could have catastrophic consequences.
Chinese and Iranian hackers are using US AI products for cyberattacks - a new Google report shows just how vulnerable we are.
OpenAI researcher Steven Adler has resigned and warns: The global race for AGI is a risky bet on our future - the necessary security measures are lacking!
And that’s a wrap! Enjoy the rest of your weekend.
Your AInauts,
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