May 9, 2023

  • Hallucinations

    What does it mean when AI ‘hallucinates’?

    The Hustle, 05/09/23. In 1943, Albert Hofmann accidentally discovered LSD while seeking a stimulant. AI hallucinations resemble conversations with intoxicated individuals, confidently providing inaccurate information. Examples include Google’s Bard chatbot fabricating a linguist’s name and ChatGPT generating false accusations. Understanding and resolving this phenomenon remains a challenge for industry leaders, as language models lack genuine knowledge and rely on data generation. The term “hallucination” is debated, with alternatives like “synthesized ungrounded text” proposed. READ THE ARTICLE

  • Google

    Leaked Google engineer memo warns that Big Tech could lose AI race to the little guys

    NBC News, 05/09/23. The leaked memo reveals a shifting playing field in the AI race as open-source developers catch up to big tech companies like Google and Microsoft. While the power of the open-source community in creating efficient projects is acknowledged, concerns about misuse and malicious applications arise. As the AI landscape evolves, questions about its future direction and societal implications loom large. READ THE ARTICLE

  • Jobs

    PayPal, IBM turn to A.I. to cut costs

    Yahoo, 05/09/23. In the current earnings season, PayPal and IBM join the trend of adopting artificial intelligence (A.I.) strategies. While PayPal sees A.I. as a means to enhance efficiency and cut costs in the long run, IBM’s mention of job replacements raises concerns. Interestingly, the impact of A.I. on stock returns varies, with IBM and PayPal experiencing declines. Nonetheless, A.I. applications are expanding across industries, with some companies aiming to sell A.I. as a product, while others offer A.I.-powered services and hardware. PayPal’s focus on A.I. aligns with the tech industry’s emphasis on efficiency and cost reduction, reflecting a promising future of integrated A.I. solutions. READ THE ARTICLE

  • Open Source

    Meta open-sources multisensory AI model that combines six types of data

    The Verge, 05/09/23. Meta, formerly Facebook, has introduced ImageBind, an open-source AI model merging text, audio, visual, movement, thermal, and depth data. Although it’s currently a research project, ImageBind demonstrates the potential for future AI models to produce immersive, multisensory content. Meta’s focus on sharing AI research sets it apart from competitors, showing commitment to openness. By integrating diverse data streams, ImageBind envisions a future of AI-generated audio, visual, and physical sensations, creating realistic and captivating virtual experiences. READ THE ARTICLE

  • Regulation

    OpenAI suggests voluntary AI standards, not government mandates, to ensure AI safety

    Fox News, 05/09/23. OpenAI’s General Counsel, Jason Kwon, argued for voluntary industry-led standards as the basis for regulating AI in a panel discussion hosted by BSA/The Software Alliance. He emphasized the importance of industry standards to ensure safe and responsible AI use, addressing issues like discrimination. Kwon acknowledged unforeseen results from OpenAI’s ChatGPT but highlighted their commitment to monitoring and addressing potential misuses. He suggested voluntary provision of comprehensive information to aid informed decision-making while regulatory measures are explored. READ THE ARTICLE