Greetings,

When you have a team of experts applying AI to a niche issue, it exceeds expectations. But general-use AI is still too unpredictable to deploy broadly in business. For companies that want to deploy it, the focus should be on preparing high-quality data, and organizing procedures for human review and testing of AI outputs. 

• DeepMind used their AI model AlphaDev to give as much as 70% speed boost to rudimentary computer algorithms that haven’t been improved in years, and the new algorithms are already used trillions of times a day. — DeepMind
  
  
• CISO at Mattel says AI is too unpredictable to deploy it broadly, saying “if we give AI too much freedom, it will probably cause a lot of trouble” - but limiting the AIs data access and teaching employees effective prompting techniques can boost safe adoption. — WSJ
  
  
• CNET has learned the same lesson after having to retract AI-generated stories, which led to their new policy that prohibits fully AI-written stories, and requires hands-on reviews by human experts. — The Verge
  
  
• Google released a 6-point security framework for building and deploying AI, a must-read for anyone wondering about how to minimize the risk of AI negatively affecting business operations. — Google
  
  
• Data is critical for AI, so all companies that want to adopt it need to improve data management practices, including better procedures to clean data, secure it, and reduce its size; one biopharmaceutical company spent 18 months just preparing their data for training and building AI models. — WSJ
  
  
• OpenAI, Palantir, Anthropic are opening AI research centers in the UK and allowing the government access to their models, with Prime Minister Rishi Sunak aiming for the country to become the “home of global AI safety regulation”. — Politico
  
  
• Google’s GenAI search experience is highly problematic - the answers are full of plagiarism without source attribution, essentially stealing traffic from websites that published information used by the AI, plus the AI often gives wrong answers. - Tom’s Hardware

----

Found this helpful? Forward this email to a colleague.

Was this email forwarded to you? Sign up here to stay on top of the AI news.