Modern-Day Oracles or Bullshit Machines?

This audio post was inspired by Carl T. Bergstrom and Jevin D. West’s course of this name, combined with various interactions I’ve had of late.

Here’s what I’m observing in my life (through direct interactions in both personal and professional contexts):

• Lots of folks are using LLMs

• Lots of folks are relying on the outputs of their interactions with LLMs

• Very few folks are critically engaging with LLMs (learning about more and less valid uses, underlying limitations, fabrication, referencing etc.)

• Many folks are ‘defending’ the tools and their usage of the tools

• Very few folks seem open to deepening their understanding of the broader historical / ecological / economic / cultural etc. etc. context within which these tools developed and are now being used

• Most folks are unknowingly amplifying AI related multipolar traps as a result of the above points

• Plus lots more

*Note I am not suggesting there aren’t valid uses cases for LLMs. There seem to be some. And there are clearly many valid uses cases for other types of AI systems. I cover some of the nuance of this in the audio.

In addition, a friend called out that the course I reference has some context that isn’t reflective of the very latest models. I’m unsure this meaningfully changes anything from a wide boundary / sociotechnical perspective, but it feels important to note for those who end up going through the course content.

Where are you at with all of this? How are you using or not using these tools? What are your expectations for the future?

Here are a few references from the audio:

With love as always.