back to article We can unify HPC and AI software environments, just not at the source code level

Welcome to the latest Register Debate in which writers discuss technology topics, and you the reader choose the winning argument. The format is simple: we propose a motion, the arguments for the motion will run this Monday and Wednesday, and the arguments against on Tuesday and Thursday. During the week you can cast your vote on …

  1. cyberdemon Silver badge
    Devil

    You can take your compute graphs and shove them up your arse

    Maybe all HPC and AI "COULD" all run on a unified platform. BUT SHOULD IT?

    Who would control the unified platform? Bob Page?

    Obscure reference alert: A pre-millenium (1999) computer game's preminition of a billionaire head of an AI company who wants to rule the world with AI weapons (mass surveillance and walking machineguns), and wished to concentrate all of the world's communications into his all-powerful AI. He also (in the game) engineered a terrible global pandemic called the Gray Death, and sold a very limited quantity of a product called Ambrosia that merely deactivated the virus in individuals for a short time, so long as they kept paying him.

    https://youtu.be/FR7yhm5obtk?t=20989

    See also: Black mirror, etc.

    What Rob Farber calls AI is nothing more than an energy-munching statistical machine that is only really useful for mass surveillance, manipulation and oppression, and not much else. And it needs to be curtailed, severely.

  2. Anonymous Coward
    Anonymous Coward

    In order to be truly agnostic, the graph language wouldn't even be able to presume anything about data types for the machine; it would have to take in a specification of the machine to transform the graph into the machine's code. But maybe that is the whole point - to stop the compiling at the point where the graph is produced, and leave all the optimization and transformatioin to machine code and load modules up to the system trying to run the code.

    Add in some interesting module differentiation and OS caching capabilities for the compiled graphs, and you start ending up with some very interesting possibilities for cross program optimization and code elimination by the system, not just the graph transformation phase.

    Ah, if I had a nickel for every bright idea I had - I'd still be poor. 'Cause I suck at marketing. :)

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon