Democratization of AI - "We have no moat, and neither does OpenAI"

A few years ago I thought the AI race would be "winner take all", because costs to build large data sets and train advanced models were already so high and would only go higher.?If we ever developed something like an artificial general intelligence (AGI) it would be built and tightly controlled only by tech giants or large governments - completely out of reach for startups, smaller companies, or individuals.

Turns out that might be wrong... One of the most fascinating developments over the past few months is the "democratization" of large language models (LLMs).?OpenAI's GPT-4 is probably the best publicly available model today, but open source models are quickly catching up - and they are becoming increasingly cheaper to train and run.

For anyone even remotely interested, this leaked document from Google - "We have no moat, and neither does OpenAI" - is a must-read to give you an idea how quickly things are progressing.

Short to medium term: this probably means that LLMs will be commoditized.?Creating a really good model may not give much competitive advantage beyond just first-mover advantage.?The models themselves will end up closer to off-the-shelf tools like a database - say MySQL or MongoDB - rather than an expensive and centrally controlled service like, say, Google Maps API.?

This means models will become relatively cheap to customize and use!?Most of the value creation will not come from the models themselves or using them directly like with ChatGPT, but from how these models will be used and integrated into existing or new products (which, incidentally, is what we do).

Long term: if we ever develop something like an AGI. Popular imagination holds the first AGI as a giant supercomputer the size of a building, surrounded by scientists with white lab coats and clipboards.?The reality may end up closer to something that emerges organically out of random Discord channels and can be run (and modified!) on pretty much anything by pretty much anyone.?Which is... probably equal parts exhilarating and terrifying.

All this also has fascinating implications for the discussion on government control and regulation of AI... but I won't go into that now.

Things are moving so quickly that my predictions are probably completely wrong - I look forward to find out just how wrong in a few years (or months!).?We live in interesting times indeed.