Re:Show HN: Marsha – An LLM-Based Programming Language
by albert_e on Wednesday July 26, 2023 @02:44AM
(#36873895)
Mostly agree.
Not sure why you were downvoted -- but this seems fairly straightforward in terms of logical progression of abstraction.
I would say LLMs may or may not be that next step in the chain before we see direct interfaces with brain for creating new programs -- but they sure seem promising.
BCI itself will still need go through a few steps where first we can reliably select from choices (click buttons) across diverse use cases, and then graduate to inferring intent / extracting logic or language etc. -- to reach a stage where we are able to ask a question of Alexa/Siri or prompt ChatGPT in a coherent way by simply thinking about it. Even the hardware would need to achieve a form factor that can be adopted by masses.
By that time hopefully LLMs would have graduated to a level that can take that input and give us robust programs / API calls etc. Or something else other than LLMs emerges by that time to fulfil the same function.
But once we get there though -- our communication with machines need not be constrained solely by clicks and words or programming constructs and logic -- but could take on a much more complex and unconstrained form. We can think in abstract concepts, ideas, visualizations, feelings and emotions -- and all of them are rich fodder as input for BCI + generative models.
For example, I could think musical ideas and am able to generate actual music live -- or "save" musical ideas as .RAW thoughts for later processing.
Or I think of the concept of frisson and imagine the feeling of goosebumps -- and able to generate content / physical design ideas that result in that sensation or state of mind.
Not sure why you were downvoted -- but this seems fairly straightforward in terms of logical progression of abstraction.
I would say LLMs may or may not be that next step in the chain before we see direct interfaces with brain for creating new programs -- but they sure seem promising.
BCI itself will still need go through a few steps where first we can reliably select from choices (click buttons) across diverse use cases, and then graduate to inferring intent / extracting logic or language etc. -- to reach a stage where we are able to ask a question of Alexa/Siri or prompt ChatGPT in a coherent way by simply thinking about it. Even the hardware would need to achieve a form factor that can be adopted by masses.
By that time hopefully LLMs would have graduated to a level that can take that input and give us robust programs / API calls etc. Or something else other than LLMs emerges by that time to fulfil the same function.
But once we get there though -- our communication with machines need not be constrained solely by clicks and words or programming constructs and logic -- but could take on a much more complex and unconstrained form. We can think in abstract concepts, ideas, visualizations, feelings and emotions -- and all of them are rich fodder as input for BCI + generative models.
For example, I could think musical ideas and am able to generate actual music live -- or "save" musical ideas as .RAW thoughts for later processing.
Or I think of the concept of frisson and imagine the feeling of goosebumps -- and able to generate content / physical design ideas that result in that sensation or state of mind.
Telepathy is one step away from there.