Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

This Week in AI: When 'open source' isn't so open | TechCrunch

Apr 20, 2024 - techcrunch.com
Meta has released the latest in its Llama series of generative AI models, Llama 3 8B and Llama 3 70B, which are capable of analyzing and writing text. However, despite being labeled as "open source," the models come with licensing restrictions, such as not being able to be used to train other models and requiring a special license for app developers with over 700 million monthly users. This has sparked debates over the definition of open source, with a study co-authored by researchers at Carnegie Mellon, the AI Now Institute, and the Signal Foundation finding that many AI models branded as "open source" come with catches, such as the data required to train the models being kept secret and the compute power needed to run them being beyond the reach of many developers.

In other AI news, Meta has upgraded its AI chatbot across Facebook, Messenger, Instagram, and WhatsApp with a Llama 3-powered backend. Social media service Snap plans to add watermarks to AI-generated images on its platform, and Hyundai-owned robotics company Boston Dynamics has unveiled its next-generation humanoid Atlas robot. Meanwhile, Reddit CPO Pali Bhat revealed that an AI-powered language translation feature is in the works, and LinkedIn has started testing a new way to boost its revenues with a LinkedIn Premium Company Page subscription that includes AI to write content. Lastly, Google parent Alphabet’s moonshot factory, X, has unveiled Project Bellwether, which uses AI tools to identify natural disasters, and Ofcom plans to explore how AI can be used to detect and remove illegal content online.

Key takeaways:

  • Meta has released the latest in its Llama series of generative AI models, Llama 3 8B and Llama 3 70B, which are capable of analyzing and writing text. However, these models are not truly open source due to certain licensing restrictions.
  • A study found that many AI models branded as “open source” come with big catches such as secret training data, high compute power requirements, and expensive fine-tuning labor.
  • Swiss researchers found that chatbots, if armed with personal information about a user, can be more persuasive in a debate than a human with the same information.
  • OpenAI is expanding to Japan with the opening of a new Tokyo office and plans for a GPT-4 model optimized specifically for the Japanese language.
View Full Article

Comments (0)

Be the first to comment!