I have some additional thoughts about AI since my last post on the potential of LLMs. Ethan Mollick shared his experience with the Code Interpreter plug-in for ChatGPT-4. I think that this will be a significant boost for LLMs. Generating code for Python applications, functional webpages, and R data analysis will provide a level of transparency that makes the results more verifiable in ways than standard text. In order o verify text results, you need to be an expert in the content area. There are plenty of automation tools to verify code.
A core capability of LLMs is mapping. This is the ability to operate on vector arrays regardless of the original data mode (i.e., text, music, images, video, code, etc.). As a result, these types of jobs (i.e., developers, UI designers, data analysts) are at risk of being automated or replaced by someone who is proficient with AI.
Developers translate people-friendly logic into algorithms coded in a development language. User Interface designers convert low-fidelity (hand dawings) into digital UI widgets (ie.e., menus, buttons, etc.). Data analysts translate people-friendly analysis requests into a relevant data set with commands to manipulate and display the refined data.
Autonomous agents are a fairly large leap from where we are today with LLMs. They provide a roadmap of sequential goals and criteria for completion. Individual goals can be assessed by the AI itself. tasks)and criteria additional ______. In a academic setting, autonomous agents could reduce the friction between students and tutors (see previous post).
