August 8, 2023
-
Nvidia
Nvidia’s AI Workbench brings model fine-tuning to workstations
TechCrunch, 08/08/23. Nvidia has unveiled its new platform, AI Workbench, which allows users to create, test, and customize generative AI models on a PC or workstation before scaling them to a data center or public cloud. This platform aims to democratize AI development by making it accessible and scalable across different infrastructure. The impetus for AI Workbench came from the challenges faced by developers in customizing large AI models and the low success rate of deploying machine learning models in enterprises. As the demand for AI grows, Nvidia’s decentralized approach to fine-tuning models on a local machine may appeal to developers who want more flexibility. READ THE ARTICLE
-
Limitations
Generative AI ‘not reliable yet,’ says Mayo Clinic’s John Halamka
Healthcare IT News, 08/08/23. Artificial intelligence (AI) is making significant progress in healthcare, despite concerns about its future. A recent survey found that nearly one-third of patients are comfortable with AI leading primary care appointments, showing a growing trust in this new technology. While AI cannot replace doctors in terms of empathy and personal preference, it can assist healthcare professionals in making better diagnoses and providing more informed care by utilizing vast amounts of data. However, caution must be exercised in choosing the appropriate use cases for AI, especially in terms of generative AI, which can be probabilistic and inconsistent. The Mayo Clinic’s partnership with Google exemplifies how AI can facilitate innovation in healthcare by speeding up processes and providing access to vast datasets. READ THE ARTICLE
-
Models
How to use Meta’s new LLaMa 2 generative A.I. tool as a chatbot, and how it compares to ChatGPT and Google Bard
Fortune, 08/08/23. The new version of Meta’s language model, LLaMa 2, offers potential for customization and adaptation but falls short compared to existing chatbot tools like ChatGPT, Bing Chat, and Google Bard. LLaMa 2 is primarily an open-source model for developers to download and modify, not a consumer-facing chatbot. Specific versions like HuggingChat have been created for chatbot purposes. While LLaMa 2 can perform tasks like writing emails and drafting memos, it lacks the creativity and finesse of ChatGPT. It also struggles with word count instructions and is not reliable for research purposes due to its knowledge cutoff and faulty search function. Despite its limitations, the LLaMa 2-based HuggingChat continues to be improved and updated, so it may be worth exploring for specific needs with awareness of its drawbacks. READ THE ARTICLE
-
Shadow AI
Shadow AI will be much worse than Shadow IT
CIO, 08/08/23. In the essay, the author discusses the potential risks of using generative artificial intelligence (GenAI) and emphasizes the need for organizations to prioritize its governance and security. They draw parallels between the adoption of public cloud and the unintended consequences of data loss and exposure, highlighting the importance of centralized strategies and governance for GenAI. The author also mentions the double exposure risk of GenAI and the potential implications on privacy and intellectual property. They provide three prescriptive ways to reduce risk and protect data, including establishing a centralized strategy, identifying data that should not be used in public or hosted private cloud AI offerings, and bringing AI to the data to gain more control. The essay concludes by suggesting that organizations seek the assistance of a technology partner like Dell to navigate the complexities of implementing AI strategies and solutions. READ THE ARTICLE