Modern Day Oracles or Bull**** Machines?

Modern-day Oracles or Bull**** Machines is the provocative title for a new class that describes itself as "How to thrive in a ChatGPT world".  The course is led by Dr. Carl Bergstrom and Dr. Jevin West at the University of Washington and covers Large Language Models (LLMs) such as ChatGPT. Dr Bergstrom is a biologist who studies information, how it evolves and flows and Dr West is a scientist who uses computational methods to study the sociology of science, spread of misinformation and generative AI's impact on collective discourse. Together they make a formidable team (and have been working together for 20 some years already!)  to tackle the complex and rapidly changing subject of artificial intelligence.

It might be a tall order, to thrive in this new AI world. Maybe AI doesn't dominate your life, yet, but it does touch it, whether you know it or not. The headings of each lesson are thought provoking: Autocomplete in Overdrive, Hard to Understand, Harder to Fix.... Democracy is the title of the last lesson, number 18---I am looking forward to seeing what that one holds.  

The first lesson states a key point in the instructors' eyes: LLMs don't 'think' about your question the way a person does.  The professors describe LLMs as extraordinary-autocomplete machines..a rather different take on the whole field than so much of the 'humanizing' that tends to happen when describing LLM interactions. Even before AI, when my computer sat there with a spinning wheel and I had to wait, I would excuse the delay with "oh sorry, my computer is thinking".

As I was reading background for this blog, I typed in "what is the difference between thinking and reasoning" into Google.  The first section that now shows up whenever I query is, ironically, Google's AI Overview, Gemini. That overview matched references further down on the page so I was comfortable with it--reasoning is a structured form of the broader term thinking in that reasoning specifically refers to drawing logical conclusions based on evidence. (ChatGPT gave a similar response, I just had to check!) I think it will be interesting to notice the use of 'reasoning' vs 'thinking' as I continue to read and learn about LLMs and AI, given that with those definitions, the models seem to be closer to reasoning models than thinking models, although I am not sure that it is advisable to use either of the labels with AI.

Of course, the tricky thing about a course whose subject is moving incredibly rapidly is how does it stay current?  When DeepSeek (Chinese AI Open Source Model) was unveiled, it set off a flurry of activity in the AI realm, and since I haven't finished the course, I don't know if developments such as that will be incorporated in the lessons.  Even so the course is still very relevant, given that no matter how the answer is derived, its always good to examine both the answer and the process critically.

The course is freely available, a mixture of text and videos. Take a look, and see what you 'think'!

AI image made with ChatGPT 4o, February 17, 2025 by Elizabeth Stewart