Powered by Data - April 1, 2024

Data Points

Last Week in Conversational AI

• Microsoft has developed new security measures for its Azure AI Studio to prevent intentional attempts to manipulate chatbots, such as prompt injection attacks or jailbreaks. These measures include "prompt shields" to identify and block unexpected behavior and tools to detect indirect prompt injections inserted by hackers into model training data. According to Sarah Bird, Microsoft's chief product officer for responsible AI, these attacks pose a unique threat, and the new protections aim to recognize and block questionable inputs instantly. Microsoft also plans to release a tool to notify users of false or embellished results generated by AI models, aiming to increase confidence in their generative AI tools. Bird emphasized the importance of implementing security measures into language models to prevent such attacks, highlighting the need for additional safeguards beyond the model technology.
  Read More
• The White House, led by Vice President Kamala Harris, unveiled new AI rules requiring US federal agencies to ensure their AI tools do not endanger the public or risk their rights and safety. Each agency must implement concrete safeguards by December, covering facial recognition, electric grid control, and mortgage decisions. Civil rights groups have praised the move, particularly in light of concerns over facial recognition misuse. The directive aligns with President Biden's AI executive order from October 2023 and mandates hiring chief AI officers, public AI system inventories, and AI governance boards. Despite these positive steps, the US government is urged to expedite the passage of comprehensive AI legislation, which is expected by 2025.
  Read More
• Amazon has decided to double its investment in Anthropic, a growing AI power, by investing an additional $2.75 billion. This follows through on an option left open last September when Amazon initially invested $1.25 billion for a minority stake in the company. Anthropic's AI models are highly competitive and available at scale for enterprises, making them attractive to companies like Amazon and Microsoft that cannot develop such models independently. The decision to invest the maximum amount reflects the strategic importance of AI in the tech industry, with companies willing to outspend rivals to secure their position in the rapidly evolving AI landscape.
  Read More