Collaborative Intelligence Week in Review - 30Jun2023
I was able to catch up on a lot of reading this week, so there are more entries than normal.
Best business article(s) I read this week…
The Economic Potential of Generative AI
Highlights of a McKinsey study on the economic impacts of Generative AI. Includes a link to the full 68-page report.
The Emergence of Dominant Designs in Artificial Intelligence
Authors forecast how they think the battle for the AI marketplace will play out and which factors will be most influential in determining the outcomes. It’s a bit premature, IMO, but it’s never too early to think about the strategic implications of a new technology wave.
What Roles Could Generative AI Play on Your Team?
As the title suggests, the authors consider different roles for generative AI, from Coach to Boss to something they term “Imperial.”
Best technical article(s) I read this week…
Orca: Progressive Learning from Complex Explanation Traces of GPT-4
Microsoft will be releasing Orca as an open-source model. It shows impressive performance for its size because it interacts with GPT-4, asking it to explain its reasoning, and then uses that explanation to develop its own reasoning. It exhibits fairly strong performance on Zero-Shot learning tasks, which are classifications or task completions about which it has not been trained.
FinGPT: Open-Source Financial Large Language Models
On March 30 of this year, Bloomberg announced BloombergGPT, a proprietary large language model (50B parameters) for querying Bloomberg financial data. A group working with the AI4Finance community announced FinGPT, an open-source model for querying financial data.
This is a GitHub repository that enables users to query their own documents using Large Language Models (LLMs) without accessing the Internet, so no private data is exposed.
Large Language Model Guided Tree-of-Thought
This paper describes an emerging area of research whose goal is to create mechanisms for LLMs to mimic what Daniel Kahneman refers to as “System 2” thinking (they currently exhibit only “System 1” thinking).
Other item(s) of note…
Why Chief Data and AI Officers Are Set Up to Fail
Presents several factors that hinder CDOs and CAIOs ability to be successful, then offers five steps to take to improve the situation.
Coolest thing I saw…
Uses Neural Radiance Fields (NeRFs), a machine learning technique, to transforms 2D images into interactive 3D images. They have an app for mobile phones. Many are expecting NeRFs to be the next big thing in digital photography.