- 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 09, 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 have summarized our highlights of the past week:
Finally, here's 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!
A colorful mix of six brand ChatGPT tips and tricks - and we've put together the best prompts for you.
Personalize your AI with custom instructions and our suggestion.
Use the Advanced Voice mode for exciting voice conversations.
Boost your searches with Perplexity thanks to clever prompt hacks.
Force your chatbot to do more thorough web searches.
And for deadlocked conversations: Just let the AI think.
For nostalgics: create more authentic retro-style images à la MySpace.
With the advance of AI, it is becoming increasingly difficult to distinguish between real and fake. From manipulated images and deceptively real videos to fake text messages and chatbots - the pitfalls are manifold.
Fortunately, there are helpful tools: Detect AI Images and forensic tools for images, Resemble for audio deepfakes, and Attestiv for videos. Copyleaks AI Detector is recommended for texts, and portals such as Mimikama help with fact-checking.
ByteDance's latest AI project OmniHuman1 takes deepfake technology to a whole new level with terrifyingly realistic videos. What makes it special: It only needs a single reference image or a short audio snippet to generate deceptively real videos of any length. While the tool is not yet publicly available, it is already raising important questions: How will we be able to distinguish between real and fake in the future? A social challenge that affects us all...
The AI search engine Perplexity has received an exciting update: two new reasoning models (DeepSeek R1 and OpenAI o3-mini) for better search results. ChatGPT is sufficient for simple queries, but for in-depth searches, Perplexity with DeepSeek Reasoning is the significantly better and more source-rich choice. Only Google Gemini's Deep Research and OpenAI's Deep Research are even more powerful - albeit for a fee.
Replit makes it possible: develop apps directly from your smartphone - without any programming knowledge! Simply describe your app idea and the AI will do the rest. Sounds too good to be true? Sometimes it is, but it's definitely worth a try!
Three promising app ideas: A subscription-free AI image generator that uses multiple models simultaneously; a senior-friendly assistant chatbot with a reminder function; and a "healthy alternative" app that suggests better food options. Bonus idea: a social media calendar that tracks content-relevant holidays and generates suitable posts.
📰 AI news quickie: HAI highlights from the industry
Did you also have the feeling that things have been a little quieter over the last few days? We certainly felt that way - until we put the news together ... Actually, there were a lot of smaller and larger updates! Here we go...
OpenAI
OpenAI has made the new o3-mini model available via ChatGPT and API - and it impresses with its cost-efficient, fast thinking (including live data). Even free users can use it, and now users without an account in ChatGPT can also use the web search.
It also excels at mathematical and scientific tasks - an o3-mini based trading strategy has even massively outperformed the market!
The new OpenAI Deep Research Agent also uses o3-mini and convinces with its multi-level research - even already in the DACH region.
According to SamA, these research agents will be able to take over a one-step percentage of all economically valuable activities!
Here is a list of 20 professions that may be replaced by the new Deep Research function - including tax advisors(98% probability). See the whole list here!
Canvas sharing is now live in ChatGPT! Share your canvas with others who can view, use or even edit it.
OpenAI has also increased the memory limit in ChatGPT by 25% for Plus, Pro and Team users.
Sam Altman and the OpenAI team discussed the future of AI in an Ask Me Anything and announced upcoming updates. And Sam also thinks that open source still has potential.
SoftBank is entering into an official partnership in Japan to introduce focused AI agents and AGI to companies.
Researchers have trained an OpenAI counterpart in just 26 minutes and for under 50 US dollars. The model, called s1, uses distillation to learn from Google's Gemini - a real game changer for cheaper AI models!
By the way, OpenAI is testing the persuasiveness of its new models on the subreddit r/ChangeMyView. The results show that the AI can indeed argue successfully...
Figure AI has decided to end its collaboration with OpenAI and is now relying on internal AI models after achieving a "revolutionary breakthrough". This raises great expectations for their humanoid robots!
California State University is integrating ChatGPT Edu into its education system, reaching almost 500,000 students.
John Schulman, co-founder of OpenAI and most recently at Anthropic, has left the AI startup to support ex-OpenAI CTO Mirat Murati at her new company.
Google is expanding the Gemini 2.0 family with new options and making it available to everyone - including the new Flash models.
The Gemini app also receives an upgrade with Gemini 2.0 Flash for faster and improved performance.
Google Sheets can now generate Python code for charts etc. thanks to Gemini.
By the way, hacker groups also like to use Google's AI Gemini...
The cool image model Imagen 3 is now also available via API.
Google is testing a new Search 'AI Mode' internally - a fresh approach to web search is being rolled out. Curious? Find out more here!
Less cool: Google removes the promise not to use AI for weapons or surveillance.
Microsoft's Mustafa Suleyman has poached three talented people from Google DeepMind who have worked on the popular AI podcasting feature. He wants to use it to shape the next generation of Copilot!
DeepSeek
Researchers have successfully jailbroken DeepSeek. This revealed the entire system instructions and possible connections to OpenAI. Very exciting!
By the way, thanks to ElevenLabs you can use the R1 model with voice integration.
Cerebras secures the title as the fastest host for the DeepSeek R1, overtaking Nvidia GPUs by a factor of 57!
Australia has banned DeepSeek due to security risks for all government devices - and there are also efforts in the USA, with the threat of a sentence of up to 20 years in prison... but it won't get through.
Features, Research, Tech
TopazLabs has released Project Starlight, a new diffusion model for restoring old videos - very impressive!
Pikadditions and Pikascenes are new Pika features that allow you to make your videos stand out from the crowd.
Btw, Grok AI now allows you to edit generated images.
Mistral AI has released an update for LeChat - super fast, keep it up!
GitHub Co-Pilot gets another update with agent functionality.
Amazon Alexa should also get an update soon, supported by Claude.
Lyft uses Anthropic's Claude for customer service - while Anthropic asks us to jailbreak their model ...
Harvard researchers have discovered six chemical compounds that can reverse cellular aging. Does this mean eternal youth and health?
According to a leaked memo, Meta is planning to launch a flood of six new AI-supported wearables around 2025. Sounds like an exciting course!
VideoJAM is a new framework for video generators that can reproduce very realistic movements.
The VR headsets Vision Pro, Meta Quest and Pico 4 (Ultra) now support over 2,000 additional video games thanks to the GeForce Now app - not AI, but cool!
Society and politics
Cheers to the Beatles! They've just won a Grammy.
Nvidia boss Jensen Huang (who sells 14 million US dollars worth of shares every day...) recommends getting an AI tutor - because the future of education and work lies in digitalization. AI does not replace, but supports people!
Reid Hoffman's new book Superagency emphasizes humanity's self-determination over AI. And he warns that the USA must stay ahead in the AI rivalry with China.
AI systems with "unacceptable risk" are now banned in the EU.
Microsoft founds the Advanced Planning Unit (APU) to investigate the social and economic impact of AI.
Musk confidant plans comprehensive AI integration in the US government.
Investments and financing
European AI start-ups have raised 8 billion US dollars in investments in 2024.
Synthesia has secured 180 million dollars in a Series D round, bringing its valuation to 2.1 billion.
Bhavish Aggarwal, founder of Ola, invests 230 million US dollars in his own AI startup Krutrim - and pushes the Indian AI market!
StackAdapt has raised 235 million US dollars in a financing round from Teachers' Venture Growth and other investors.
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