Love the 15 minute segment on reference and grounding.
After listening to this, it strikes me that the AI/ML research community needs to tap reach beyond philosophical, cognitive, and computational notions of reference and grounding and get some synergistic work going with some folks from the rhetoric, narrative, and communication/media area studies make some real headway with these issues.
In my communities, we are comfortable talking about reference and representation as a kind of parallel circuit system with different kinds of intersection points depending on one's social, political, ideological, and metaphysical commitments. For example, you have your profound relativists and correlationists who for a variety of different reasons believe there to be little to no intersection. You have your critical realist like me who find shadowy correlations and sinewy causations between the parallel circuits that are easy to detect but difficult to describe accurately and explain comprehensively.
For a more empirical approach, perhaps we can take John Searle's performative utterances as a starting point. Language when spoken makes things happen. What, why , and how is to be determined...
I like the idea of a tight causal nexus as a starting point -- and so do your guests -- until they set it aside for a completely contravening thought experiment. A very cool one by the way. And yet, reference in speech is so much more complicated. Take fiction for instance. What makes Madame Bovary's reference to her comb referential? What makes a thought experiment or a hypothesis referential? This is where one needs to evoke some kind of parallel circuit explanatory system at the level of grounding and inside the operations of mind and language.
In the narrative school I operate in, we talk a lot of doubling, particularly in fiction. Fictions double observation perfectives, voices, subjective position, ideological commitments. In a sense, fiction should be the starting point of the conversation about reference in LLMs since arguably this is the genre they appear to excel at.
Just some initial thoughts... Know I will get off my soap box!!!
Thanks again for all your hard work, Daniel!!! I know you spend countless hours preparing and creating these presentations!!! We greatly appreciate it. As I have said many times, you are creating an ongoing AI masterpiece for the masses.
Thanks a lot for this great podcast. Especially the part about multimodal LLMs was very interesting. Hope to hear more about that exciting field on your podcast in the future. 👍
I’m a complete digital incompetent, so I appreciate these discussions which help me understand AI and LLM
Love the 15 minute segment on reference and grounding.
After listening to this, it strikes me that the AI/ML research community needs to tap reach beyond philosophical, cognitive, and computational notions of reference and grounding and get some synergistic work going with some folks from the rhetoric, narrative, and communication/media area studies make some real headway with these issues.
In my communities, we are comfortable talking about reference and representation as a kind of parallel circuit system with different kinds of intersection points depending on one's social, political, ideological, and metaphysical commitments. For example, you have your profound relativists and correlationists who for a variety of different reasons believe there to be little to no intersection. You have your critical realist like me who find shadowy correlations and sinewy causations between the parallel circuits that are easy to detect but difficult to describe accurately and explain comprehensively.
For a more empirical approach, perhaps we can take John Searle's performative utterances as a starting point. Language when spoken makes things happen. What, why , and how is to be determined...
I like the idea of a tight causal nexus as a starting point -- and so do your guests -- until they set it aside for a completely contravening thought experiment. A very cool one by the way. And yet, reference in speech is so much more complicated. Take fiction for instance. What makes Madame Bovary's reference to her comb referential? What makes a thought experiment or a hypothesis referential? This is where one needs to evoke some kind of parallel circuit explanatory system at the level of grounding and inside the operations of mind and language.
In the narrative school I operate in, we talk a lot of doubling, particularly in fiction. Fictions double observation perfectives, voices, subjective position, ideological commitments. In a sense, fiction should be the starting point of the conversation about reference in LLMs since arguably this is the genre they appear to excel at.
Just some initial thoughts... Know I will get off my soap box!!!
Thanks again for all your hard work, Daniel!!! I know you spend countless hours preparing and creating these presentations!!! We greatly appreciate it. As I have said many times, you are creating an ongoing AI masterpiece for the masses.
Thanks a lot for this great podcast. Especially the part about multimodal LLMs was very interesting. Hope to hear more about that exciting field on your podcast in the future. 👍