Silence before the storm?
The 'AI' landscape may be changing in unanticipated ways.
Bundles of thought leading to 'AI' trace strongly to the 1980s, and threads, more tenuously, much further back. LLMs, and similar, were brilliant conceptions when viewed as data storage and interrogation tools, and as agents for automation. Unfortunately, high-profile AI development has fallen into the hands of clever technicians employed by adventurers. One is led to believe that the West, especially the USA, leads in the area of further conceptual and practical development. Perhaps, not so. The current Western paradigm for 'AI' R&D is focussed on 'the bigger, the better' and as size increases on the emergence of properties not already implicit to base theoretical structures. Already, the best known models exhibit eccentricities indicative of severe limitations to their application by the unwitting.
Advances in chip technologies and chip mass production are bringing new players into the game; this despite the US Administration's strong efforts to establish a monopoly. Almost certain to come are cheaper and less power hungry chip configurations; these should coincide with parallel developments in underlying software to bring about more rapid training and smaller energy requirements for practical uses.
These trends suggest that innovations in model structures, and in training regimens, don't depend upon the behemoths with which OpenAI is linked. People not bothered to create Arthur C. Clarke's 'HAL 9000', but content with tools lessening some burdens on human intellectual endeavour, will not seek to feed their creations with all they can get their hands upon of human slop expressed in digital format: the output of fools and vain ideologues having the same weight as that of sages. Subject, discipline, and task-specific 'AIs' will arise. These, as is so for cut-down versions of current mega-'AIs', shall fit within computational devices of modest (yet seemingly always increasing) technical specification. Moreover, being digital in nature, they, just like digital books, films, music, 3D-printer recipes, etc. will be freely shared regardless of people with Luddite thought processes seeking otherwise.
It may be that OpenAI and similar current mega-players shall monopolise 'all singing and dancing' 'AIs' for use in the context of social media, of ignorant home-users, and of tools sold to naive Western governments. There should be plenty of money generated, but whether enough to meet present expectations is moot.
Clearly, 'AI' tools are handy for promulgating 'values' and muting those deemed unacceptable. Censorship, ostensibly for 'the sake of the children' is rife. I use an 'AI'-tool to check my spelling as I write. When I refer to 'the Ukraine' it tells me I mustn't because that is no longer acceptable usage; at least, it doesn't yet have the gall to force me to conform or, I think, to report me to Mr Starmer's thought police. Presumably, many general purpose 'AIs' will not only object to one discussing 'Lolitas', but also to wrongful interpretations of historical events. Although the West at present is taking propaganda and rooting out 'disinformation' to high art, such as Dr Goebbels would approve, one must assume some nations elsewhere may do so according to their 'official narratives'. So, when the Internet is awash with general-purpose 'AIs' which can be interrogated, savvy users can play with several to compare and contrast the built-in slants of various corporations and countries.
Biting the hand that feeds IT
