Skip to main content

This Week in Generative AI – 15 to 21 July

By July 21, 2023September 6th, 2023No Comments

Meta Launches New Generative AI Tool For Research, Commercial Use

  • Meta Platforms, the parent company of Facebook, has launched a new large language model called Llama 2.
  • It is available for free for both research and commercial use.
  • Meta has partnered with Microsoft to make Llama 2 accessible through Azure and Amazon Web Services, promoting an open approach to AI development.

Alarming Decline in ChatGPT Response Quality Found by Researchers

  • Stanford and UC Berkeley researchers have found evidence of a decline in the quality of responses from OpenAI’s ChatGPT model.
  • The recently released GPT-4 model had a significant drop in success rate when answering prime number queries.
  • The study evaluated the performance of GPT-4 and GPT-3.5 across various tasks and found a decrease in response quality.
  • Tasks fall into four categories: solving math problems, answering sensitive questions, code generation, and visual reasoning.

Apple GPT: ChatGPT Response, Arriving Next Year

  • Apple is reportedly developing its chatbot service called ‘Apple GPT’ using its Ajax framework.
  • The company is concerned about missing out on the generative AI trend and aims to transform how people interact with their devices.
  • Ajax has already improved search, Siri, and Maps.

Salesforce Targets AI Boom With $4bn UK Investment

  • Salesforce, the CRM giant, plans to invest $4bn in its UK business over the next five years.
  • The company aims to capitalise on the growing demand for artificial intelligence technology and has already launched various AI-powered applications.
  • Salesforce has also partnered with Deloitte Digital to help businesses implement and scale AI for CRM.

AI Feedback Loop: A Threat to Future Generative Models

  • A new paper titled “The Curse of Recursion” speculates that as the internet becomes filled with AI-generated content, large language models (LLMs) like OpenAI’s ChatGPT will struggle to progress.
  • The reliance on generative AI services and algorithms will lead to “Model Collapse” and make training new LLMs more difficult.