Ravi,

There is a huge difference between the theoretical issues about what LLMs (or the smaller SLMs) can do as the underlying technology and what any particular software system can do.

The limitations of LLM technology (summarized in the note that started this thread) cannot be extended by systems that just add various interfaces to the LLMs.  But applications that combine LLMs with other technology (from AI, computer science, or many kinds of applications) may support a wide variety of technology.

Examples that we have discussed before include Wolfram's use of LLMs to support an English-like front end to their powerful Mathematica system.  That does everything that anyone has done with Mathematica and adds an English-like front end to give users a simpler and more friendly interface.  Many other companies are supporting such technology with varying degrees of success.  They implement a friendly interface to their previous systems.

Our VivoMind system in 2010 included technology that was very powerful and ACCURATE for applications that cannot be done with LLMs even today.  See https://jfsowa.com/talks/cogmem.pdf 

Our new Permion.ai system combines a newer version of what VivoMind could do with LLMs to support the interface.  Arun Majumdar and I have discussed these issues in talks that we gave in the past year or so.

I believe that is the wave of the future:  use LLMs as one component of an Ai system that uses other kinds of technology to implement functionality that LLMs, by themselves, cannot support.

Answer to your question:   The features you're asking for in the note below would be very easy to implement -- just add an on/off button for features you don't want.  That does not require new technology.  It just requires somebody to add that button.

John

-------------------------------------------------------------------

John

Are we at a point where

1. We can turn AI off say on phone apps and desktops

2. We can limit the content access to AI for inputs to hopefully focus the results better
Regards