I’m not an expert in AI systems, but here is my current thinkging:
Insofar as ‘GenAI’ is defined as
AI systems that can generate new content, including text, images, audio, and video, in response to prompts or inputs
I think this is genuinely bad tech. In my analysis, there are no good use cases for automating this kind of creative activity in the way that the current technology works. I do not mean that all machine assisted generation of content is bad, but just the current tech we are calling GenAI, which is of the nature of “stochastic parrots”.
I do not think every application of ML is trash. E.g., AI systems like AlphaFold are clearly valuable and important, and in general the application of deep learning to solve particular problems in limited domains is valuable
Also, if we first have a genuinely sapient AI, then it’s creation would be of a different kind, and I think it would not be inherently degenerative. But that is not the technology under discussion. Applications of symbolic AI to assist in exploring problem spaces, or ML to solve classification problems also seems genuinely useful.
But, indeed, all the current tech that falls under GenAI is genuinely bad, IMO.
things like the “patch x out of an image” allows people to express themselves with their own creative works more fully
text-based genai has myriad purposes that don’t involve wholesale generation of entirely new creative works:
using it as a natural language parser in low-stakes situation (think like you’re browsing a webpage and want to add an event to the calendar but it just has a paragraph of text that says “next wednesday at xyz”)
the generative part makes it generically more useful that specialist models (and certainly less accurate most of the time), and people can use them to build novel things on top of rather than be limited to the original intent of the model creator
everything genai should be used for should be low-stakes: things that humans can check quickly, or doesn’t matter if it’s wrong… because it will be wrong some of the time
I’m not an expert in AI systems, but here is my current thinkging:
Insofar as ‘GenAI’ is defined as
I think this is genuinely bad tech. In my analysis, there are no good use cases for automating this kind of creative activity in the way that the current technology works. I do not mean that all machine assisted generation of content is bad, but just the current tech we are calling GenAI, which is of the nature of “stochastic parrots”.
I do not think every application of ML is trash. E.g., AI systems like AlphaFold are clearly valuable and important, and in general the application of deep learning to solve particular problems in limited domains is valuable
Also, if we first have a genuinely sapient AI, then it’s creation would be of a different kind, and I think it would not be inherently degenerative. But that is not the technology under discussion. Applications of symbolic AI to assist in exploring problem spaces, or ML to solve classification problems also seems genuinely useful.
But, indeed, all the current tech that falls under GenAI is genuinely bad, IMO.
things like the “patch x out of an image” allows people to express themselves with their own creative works more fully
text-based genai has myriad purposes that don’t involve wholesale generation of entirely new creative works:
using it as a natural language parser in low-stakes situation (think like you’re browsing a webpage and want to add an event to the calendar but it just has a paragraph of text that says “next wednesday at xyz”)
the generative part makes it generically more useful that specialist models (and certainly less accurate most of the time), and people can use them to build novel things on top of rather than be limited to the original intent of the model creator
everything genai should be used for should be low-stakes: things that humans can check quickly, or doesn’t matter if it’s wrong… because it will be wrong some of the time