Stir It Up...

The impact of the introduction of the Chinese AI Deepseek R1 on the markets could be the start of the beginning of this particular tech-bubble's bursting. Too early to call yet, but the fact that hardware manufacturing stocks [Nvidia in this case as the world's leading GPU purveyors] took such a big initial hit [they've crawled back up a tad since] would logically suggest two scenarios. Either those behind Deepseek have discovered a new paradigm for AI modelling which relies far less on massive hardware power and energy input to operate; or it's a cynical ploy to destabilise US stocks and manufacturing in the light of the second Trump administration.

I don't believe either to be the case, but I'm still sceptical of the world's Gadarene [politically & commercially driven] rush to wholesale proselytisation and adoption of the technology as the next best thing since sliced bread. It ain't, and it's obvious to anyone actually watching the tech itself. As I've said before: in certain, highly constrained and mostly vertical use-cases, the technology can prove invaluable; but in its wider, generalised adoption, it is simply in its [virtual] infancy, and like all infants lacks anything like the sophistication of experience, education or Weltanschauung necessary to make even vaguely informed pronouncements on the world in general. The tech is clever, but far from infallible. The thing about Deepseek is that it is a 'distilled' language model that uses a technique of training a smaller, less hardware intensive AI [Deepseek] by utilising the established large models such as Chat GPT as 'teachers'. 

Essentially, the model 'distils' the already fermented data sets of the hugely resource-intensive established LLMs into a form that is far less computationally-intensive and cheaper to achieve and run. The problem lies in that its tutors are just those LLMs, which are themselves imperfect, and often prone to 'hallucinations'; i.e being 'creative' on their own behalf and generating perfectly plausible, but Trumpian, 'alternative facts'. Politicians are falling over themselves to be 'the world leaders' in this fast-emerging, interesting, but essentially nascent technology - that has some merits, but is still really little more than a bright toddler in real terms - only demonstrates their own naivety and desperate need to seem 'relevant' in a world that is far too 'modern' for their generation to fully comprehend(!). To be seen to be cutting-edge, or in this case bleeding edge, has always ultimately been their downfall, throughout history. It's always best to wait for at least V1.5 of anything before hitching your own wagon to it...