Light at the end of the tunnel?

[Frisby572]

In the past I used C/C++, and some other languages, for personal projects arising from professional needs or from curiosity. If I've understood it correctly, this project offers consolidation within the arena of offline AI model use, especially image manipulation.

At present, personal use of AI, along with its large scale commercial application, is in ferment; this resulting from AI being new ground offering many routes for exploration and rabbit holes leading to nowhere. In Thomas Kuhn's terminology, a communal 'paradigm' is yet to be established.

My remarks arise from the perspective of somebody wanting to play with AI, to explore some possibilities, rather than engage in its configuration for use. Doubtless, many people wish to take software 'off the shelf' and concentrate upon deploying it. Therein lies the rub.

Setting aside the plethora of commercial offerings for running software, leaves an equally confusing set of offline options. Whilst some open-source home/office-use software is showing signs of maturity and reliability, dilettante users are confronted by an array of difficult to configure options without wholly trustworthy guidance.

For example, Stable Diffusion technology arouses awe. Its embedment in front-end software such as ComfyUI can be used 'out of the box' by most people. Similarly, encapsulation of ComfyUI and a range of other interfaces within Stability Matrix offers great convenience. However, the present state of play is that not all software available within Stability Matrix, within other encapsulations, or as standalone, is wholly fit for end-users.

Matters are further confused when a new chip-configuration, e.g. NVIDIA's 5090, is pushed into the fray. This NVIDIA graphics card has been sold for more than one year, but is only just beginning to fully be integrated within AI frontend software. Confusion is compounded by NVIDIA's CUDA software, in many cases, requiring linkage to PyTorch software for full functionality.

Delving down further, it's apparent that Python software greatly magnifies difficulties arising from potential incompatibilities at higher levels. The immense number of variants of Python3, coupled with a multiplicity of 'environments', is the foundation for Babelian software confusion.

I've not studied Python in detail, but it looks trivial when compared to C/C++. Its interpreted nature is unattractive, yet, these days, that is unlikely to pose much processing-time overhead. Python's error reporting is of Byzantine complexity.

Maybe, AI interface technology re-coded in C++ shall displace amateurish Python code. Perhaps, it will become a standard, and incorporated into the GNU stable.

Is there a 'usability by novices' timetable for this project's output? How highly does intelligible to laymen documentation figure in all this?

[MrSnichovitch]

Filed #864 about this. Hopefully it'll get some traction when the devs can spare it some time.