The distinction is largely meaningless. Why is txt2img considered “generative” whereas img2txt (OCR) is not? Sometimes it is treated when it is in the form of a VLM. Why is aud2txt (transcription) not typically considered generative? Is txt2aud generative (TTS)? What about translation software? It’s all fundamentally the same technology, there is no rigorous definition of “generative AI,” and if we want to talk about something specific like LLMs, these also have scientific applications. For example, there are various LLMs trained on generating protein sequences or predicting the behavior of protein sequences, like ProGen, ProLLaMA, and ProteinGPT. The whole “I’m pro AI just anti generative AI” thing is a meaningless sentiment.
The distinction is very meaningful, generative ai is a tiny significantly less useful subset of ai basically just llms and image generation, it mostly describes the application not the technology.
Why is LLMs and image generation generative AI but music generation isn’t, or speech generation, or protein sequence generation, or material design generation, etc? Again it’s very arbitrary. Just say you don’t like LLMs and image generation. “Generative AI” doesn’t have a concrete meaning. All ANN technology universally used to generate some output.
that was not meant to be an exhaustive list, thats my bad i should have used clearer language. I think most people would include most of the things you mentioned in the definition, and yeah its arbitrary but that doesn’t make it less useful. People use the term instead of listing everything it includes for the same reason i did and for the same reason that thousands of other terms are used instead of listing a bunch of things.
That didn’t address the point I was making, all AI is ultimately about generating outputs, so I am not sure where your line of “generative AI” actually begins and ends. The term is absolutely 100% meaningless if I have zero idea what even qualifies as “generative AI” and what doesn’t, because then you aren’t telling me anything, I don’t know what you’re saying you like and what you don’t like, and different people would probably have different ideas over what even counts as “generative AI.” I am saying the term is too ambiguous for me to even know what is being talked about and your response is “well it’s just a lot of things and dontcha know in the English language we use terms for a lot of things all the time.” Like… what??? How is that a response. An appropriate response to what I said would be to actually tell me something more concrete I could use to judge whether or not something counts as “generative AI” vs not.
From my standpoint it really seems like “generative AI” is just a stand-in for “AI I don’t like.” People use it and arbitrarily lump in things they consider “slop factories” like image generators or ChatGPT, but when you point out plenty of other AI actually do have very practical usages in the science, some even also being LLMs or based on diffusion technologies, they will say “erm well I just dislike generative AI” even though again the technology is fundamentally the same and they are both generating content. The caveat is not really any more meaningful than just a placeholder for AI people think is bad.
Generative AI is colloquially used to refer to AI which you prompt in natural language to produce some stuff for you. If you prompt some AI to make music or protein sequences for you then that is generative AI too. It is a loose term and not something that AI scholars agree upon but it is not meaningless.
Generative AI is colloquially used to refer to AI which you prompt in natural language to produce some stuff for you. If you prompt some AI to make music or protein sequences for you then that is generative AI too. It is a loose term and not something that AI scholars agree upon but it is not meaningless.
Again, you only proved my point as you gave me a definition that applies to things like OCR, translation software, and voice recognition, which people wouldn’t colloquially categorize as generative AI. You cannot provide a definition that gives the kind of carve-out you want because it doesn’t exist, and any attempt to do so only solidifies my point further. The carve-out is ultimately arbitrary, it is just an arbitrary list of AI people don’t like.
Because you use a prompt in natural language to produce some stuff for you…? In this case a translation. There are already entire companies who sell entire books translated using AI and there’s a lot of them on Amazon. If “generative AI” were to refer to anything at all it seems strange you want it to exclude entire books generated by AI.
If you want to be strict about natural language actually being complete and grammatically correct sentences like we’re talking here, then translation software is generative AI but some AI image generators like Stable Diffusion are not since they rely on you using a list of positive and negative tags and not sentences that you would speak. It would also mean that if I build an AI to send commands to a robot based on voice commands that would qualify as generative AI as well since it is producing the command output for me based on speech.
It’s the same neural network technology, yes, and a different application, yes. But the generative applications require much more complex neural networks and push the same technology a lot further in ways that have affected more traditional uses of neural networks as well.
The distinction is largely meaningless. Why is txt2img considered “generative” whereas img2txt (OCR) is not? Sometimes it is treated when it is in the form of a VLM. Why is aud2txt (transcription) not typically considered generative? Is txt2aud generative (TTS)? What about translation software? It’s all fundamentally the same technology, there is no rigorous definition of “generative AI,” and if we want to talk about something specific like LLMs, these also have scientific applications. For example, there are various LLMs trained on generating protein sequences or predicting the behavior of protein sequences, like ProGen, ProLLaMA, and ProteinGPT. The whole “I’m pro AI just anti generative AI” thing is a meaningless sentiment.
The distinction is very meaningful, generative ai is a tiny significantly less useful subset of ai basically just llms and image generation, it mostly describes the application not the technology.
Why is LLMs and image generation generative AI but music generation isn’t, or speech generation, or protein sequence generation, or material design generation, etc? Again it’s very arbitrary. Just say you don’t like LLMs and image generation. “Generative AI” doesn’t have a concrete meaning. All ANN technology universally used to generate some output.
that was not meant to be an exhaustive list, thats my bad i should have used clearer language. I think most people would include most of the things you mentioned in the definition, and yeah its arbitrary but that doesn’t make it less useful. People use the term instead of listing everything it includes for the same reason i did and for the same reason that thousands of other terms are used instead of listing a bunch of things.
That didn’t address the point I was making, all AI is ultimately about generating outputs, so I am not sure where your line of “generative AI” actually begins and ends. The term is absolutely 100% meaningless if I have zero idea what even qualifies as “generative AI” and what doesn’t, because then you aren’t telling me anything, I don’t know what you’re saying you like and what you don’t like, and different people would probably have different ideas over what even counts as “generative AI.” I am saying the term is too ambiguous for me to even know what is being talked about and your response is “well it’s just a lot of things and dontcha know in the English language we use terms for a lot of things all the time.” Like… what??? How is that a response. An appropriate response to what I said would be to actually tell me something more concrete I could use to judge whether or not something counts as “generative AI” vs not.
From my standpoint it really seems like “generative AI” is just a stand-in for “AI I don’t like.” People use it and arbitrarily lump in things they consider “slop factories” like image generators or ChatGPT, but when you point out plenty of other AI actually do have very practical usages in the science, some even also being LLMs or based on diffusion technologies, they will say “erm well I just dislike generative AI” even though again the technology is fundamentally the same and they are both generating content. The caveat is not really any more meaningful than just a placeholder for AI people think is bad.
Generative AI is colloquially used to refer to AI which you prompt in natural language to produce some stuff for you. If you prompt some AI to make music or protein sequences for you then that is generative AI too. It is a loose term and not something that AI scholars agree upon but it is not meaningless.
Again, you only proved my point as you gave me a definition that applies to things like OCR, translation software, and voice recognition, which people wouldn’t colloquially categorize as generative AI. You cannot provide a definition that gives the kind of carve-out you want because it doesn’t exist, and any attempt to do so only solidifies my point further. The carve-out is ultimately arbitrary, it is just an arbitrary list of AI people don’t like.
How does translation fit this definition?
Because you use a prompt in natural language to produce some stuff for you…? In this case a translation. There are already entire companies who sell entire books translated using AI and there’s a lot of them on Amazon. If “generative AI” were to refer to anything at all it seems strange you want it to exclude entire books generated by AI.
If you want to be strict about natural language actually being complete and grammatically correct sentences like we’re talking here, then translation software is generative AI but some AI image generators like Stable Diffusion are not since they rely on you using a list of positive and negative tags and not sentences that you would speak. It would also mean that if I build an AI to send commands to a robot based on voice commands that would qualify as generative AI as well since it is producing the command output for me based on speech.
It’s the same neural network technology, yes, and a different application, yes. But the generative applications require much more complex neural networks and push the same technology a lot further in ways that have affected more traditional uses of neural networks as well.