Re: generative KR 2. a step back

I think your assertion needs better definition. How do you want to define “generative KR” and how does that relate to “algorithm design”?

The paper starts by talking about generalising from numbers to symbols as variables. However, they too don’t define what they mean by “generative”. 

Large language models can handle such generalisations using unguided training for the foundation models, followed by reinforcement learning with human feedback, e.g. to deal with exam questions that require step by step answers.

An open question is how to design neural networks capable of reflective cognition for generating and evaluating learning hypotheses.

> On 20 Sep 2023, at 08:02, Paola Di Maio <paola.dimaio@gmail.com> wrote:
> 
> I seem to come across very interesting research but the underlying assumption is that questions surrouding generative AI (hence KR) are coupled with learning and knowledge acquisition. 
> Here is another good  bacground read on the subject 
> 
> How is Abstract, Generative Knowledge Acquired?
> A Comparison of Three Learning Scenarios
> https://conferences.inf.ed.ac.uk/cogsci2001/pdf-files/0710.pdf
> 
>  But I would take a step back. and tackle  the question of generative KR not as a learning
> process or learning outcome or learning task, ie, not a posteriori, not after K is generated
> but before, when algorithm is design.  I think generative KR happens in the algorithm design
> Happy to have a fight over this with anyone

Dave Raggett <dsr@w3.org>