Collaborative Intelligence Week in Review - 11Nov2024
Best business article(s) I read this week…
A Practical Guide to Gaining Value From LLMs
This article provides business leaders with a high-level, structured approach to effectively using large language models (LLMs). It stresses the importance of identifying suitable business tasks for automation by breaking processes into smaller steps and assessing the cost-effectiveness of each task. Key strategies include using simple prompting, retrieval-augmented generation, and instruction fine-tuning to align LLM outputs with specific business needs. The author argues that by adopting a thoughtful, cost-aware approach, companies can pilot LLM-based solutions that potentially enhance productivity while managing the associated risks.
Best technical article(s) I read this week…
Artificial Intelligence and Strategic Decision-Making
The article examines how artificial intelligence (AI), especially large language models (LLMs), could transform strategic decision-making (SDM) by enhancing tools like Porter’s Five Forces and scenario planning. Empirical studies show that LLMs can generate and evaluate business strategies at a level comparable to entrepreneurs and investors, potentially making strategy development faster and more scalable. However, the article highlights uncertainties about AI’s ability to create novel strategies and how competitive dynamics might shift as AI capabilities improve. This article is valuable for business leaders because it offers a realistic view of AI's current and potential roles in SDM, providing a framework to understand how AI might affect competitive advantage, innovation, and firm performance.
Other items I found valuable/interesting…
Claude AI 3.5 Haiku Dropped. How Reading Feynman Reveals AI Trends
This article analogizes the ideas of Nobel-prize-winning physicist Richard Feynman, specifically his 1959 talk titled “There’s plenty of room at the bottom,” with past, present, and future developments in AI.
Microsoft Wants to Patent a Fix for AI Hallucinations
Microsoft has filed a patent application for a system that reduces AI hallucinations. The proposed system cross-references AI-generated content with trusted data sources to ensure accuracy. This initiative aims to enhance the reliability of AI outputs, particularly in critical applications like healthcare and finance.
Radio Station's Attempt to Replace Hosts with AI Ends in Absolute Disaster
This article recounts the many missteps of a Polish radio station that replaced all of its human on-air talent with AI-generated personalities. It may be considered a good example of what not to do with AI.
Coolest thing I saw…
This is the new multimodal (voice, text, images) model from The Allen Institute for Artificial Intelligence (Ai2). Various benchmarks show that it outperforms OpenAI’s ChatGPT-4o model. It’s also open, so you can use and download it for free. Its interface includes voice transcription, so you can say your prompts rather than type them. This capability is not available on ChatGPT or Claude but is on Gemini.
The coolest feature is its ability to analyze images and identify objects. It does this by overlaying dots on the items that match the criteria. I tried a similar experiment on ChatGPT, Gemini, and Claude, but they returned text responses that missed the mark.
A company that caught my eye…
The company, which makes AI-based software for robotic control, just closed a $400M round led by Jeff Bezos and OpenAI. The videos included in this blog post show early signs of what is possible when robots operate under a “generalist policy.”