The AI-ronman 🚀

Beep boop! Ready to upload the latest AI gossip straight into your neural network? Let's get computational! 🧠

Quick Takes ⚡

1. Mistral releases new AI models optimized for laptops and phones

2. Anthropic strengthens AI safety measures

3. Meta researchers develop method to make AI models "think" before answering

4. NYT takes legal action against Perplexity

5. OpenAI welcomes AI and security veterans to its ranks

6. Adobe launches Firefly video model and enhances image, vector and design models

7. Google partners with Kairos Power for clean data center energy

Deep Dive 🔍

• Mistral releases new AI models optimized for laptops and phones 📱💻️ 
  
  French AI startup Mistral has launched "Les Ministraux," its first generative AI models designed specifically for edge devices like smartphones and laptops. These models feature an extensive 128,000-token context window and are available as Ministral 3B for 4¢ and Ministral 8B for 10¢ per million tokens. For self-deployment, developers need to obtain a commercial license from Mistral, while the models can also be accessed through Mistral’s cloud platform and its partner clouds.

  

  
  
  

• Anthropic strengthens AI safety measures 🛡️
  
  Anthropic announced a key update to its Responsible Scaling Policy, incorporating new safety and governance measures for advanced AI. The policy now includes ‘Capability’ and ‘Required’ Thresholds to trigger additional safeguards for high-risk AI applications such as bioweapons and autonomous research. A Responsible Scaling Officer will oversee the implementation, and Anthropic commits to greater transparency with public reports and expert involvement. This reinforces their role as a safety-focused AI leader.
  
  

• Meta researchers develop method to make AI models "think" before answering ✨ 

  
  Meta researchers have rolled out Thought Preference Optimization (TPO), a groundbreaking method that trains large language models to “think” before responding. TPO prompts AI to generate and utilize internal thoughts privately, boosting performance in creative and marketing tasks while facing challenges in mathematical reasoning. Building on OpenAI’s recent work, TPO aims to create more adaptable and intelligent AI assistants for a wide array of applications.
  

  

  
  
  

• NYT takes legal action against Perplexity 📰 
  
  The New York Times has sent a cease and desist to AI search startup Perplexity for unauthorized use of its articles in AI-generated summaries. Despite Perplexity’s promise to stop crawling NYT content, the issue persists. Perplexity expresses willingness to collaborate with publishers and plans to respond by October 30. This move follows similar lawsuits against OpenAI and Microsoft, highlighting ongoing tensions between media outlets and AI platforms over content usage.
  
  
  

• OpenAI welcomes AI and security veterans to its ranks 👥 
  
  In the wake of a major team exit, OpenAI has recruited Sebastian Bubeck, ex-Microsoft VP responsible for the Phi models that optimize on-device AI, and Dane Stuckey, former CISO of Palantir with deep expertise in digital forensics. Bubeck will spearhead efforts to improve AI efficiency and model development, addressing areas where OpenAI seeks to advance. Meanwhile, Stuckey will fortify OpenAI’s security infrastructure, particularly in its growing partnerships with the U.S. Department of Defense, as the company expands its role in military AI projects.
  
  
  

• Adobe launches Firefly video model and enhances image, vector and design models 🎬
  
  Adobe has started rolling out its Firefly Video Model, an AI tool that transforms text prompts into videos. Competing with OpenAI’s Sora and Meta’s video AI, Adobe stands out by using legally cleared data for seamless commercial use. Currently available to waitlisted users, Firefly Video is already being used by brands like PepsiCo’s Gatorade and Mattel’s Barbie. Additionally, Adobe is enhancing its Creative Cloud suite with new AI features in Photoshop, Illustrator, and InDesign.
  
  

  

  
  
  

• Google partners with Kairos Power for clean data center energy 🔋 
  
  In response to growing power demands, Google has signed a partnership with Kairos Power to utilize clean nuclear energy for its data centers. The deal includes sourcing from Kairos’ small modular reactors (SMRs), with the first expected online by 2030 and an additional 500 megawatts by 2035. Kairos’ reactors employ molten fluoride salt, offering a scalable and economical nuclear solution. This strategy follows the lead of other tech companies like Microsoft and Amazon in pursuing sustainable energy options.

Signing off! 🖖 May your neural nets be deep and your overfitting minimal! 📉

Ciao for now!

Author: Poonam 👧 

Karan 😎 🚀 
CEO, GlazeGPT.com