Brilliant question and well timed. I have been fascinated by many things AI is going to be able to do. I agree with much of the video, but in some ways it’s more than that. For background I have been way deep in genetic coding and neuroevolutino etc. for a while, i.e. mimicking the brain, or in other words probabilistic and simple compute. It’s ants man, it’s ants 
Unclear, random thinking incoming 
(early morning, brain dump, no proof read)
Anyway, this is why I hate the massive total order or determinism, it’s plain wrong for so many obvious reasons. However, that has been a debate and I think the debate is over. With current LLM, ANN and watch for liquid networks, the proof points are too numerous to mention. However, AI is being seen as a typically polarised thing today, there’s the oh it’s full AGI already camp and the, this thing makes mistakes, so it’s crap camp. I think they are both wrong. In my mind current AI shows us some amazing things. My head is not clear on this yet, maybe it never will be, but here are some thoughts:
- Simple models work well, but require scale (many billions of neurons get smart, but only a few thousand are not) A bit like a chimps brain has only 1% less neurons than us, so scale matters, A LOT.
- Probabilistic compute allows malleable capabilities (determinism is a fixed do this, get that, whereas probabilistic compute is do something I have never seen and we will still get a good output)
- Data is not knowledge, knowledge is extracted from data
- We can run these things locally and they are good enough to be the app engine for all our needs (i.e. good enough to become our interface with knowledge, already)
From a developer point of view, then it’s interesting. One of the things that will vanish is the skill gap of the user. For instance my mum, if I ask her to send an email goes into a complete meltdown and panic. Why?. It’s simply the interface. It is the, I can or cannot navigate the apps or command lines etc. part and that changes with LLM already. The apps and command lines probably vanish now.
What I mean I want to send Nigel and email today means the same thing to the computer expert or the computer illiterate grannie. With LLM both parties now have the same “power”, both can instruct the AI to send the email. So AI removes the clutter and provides a malleable interface to the user, whether the user is a top 1% computer engineer or a grannie. This is brilliant and also a major part of the SAFE vision (Everyone).
Another step change is knowledge, so music, files, images, videos etc. I have for many years disliked the files and folders model, apple and microsoft as well as Linux et all have leaned towards search instead of navigation. It’s poor, but perhaps better, but it’s poor. Spotifiy and netflix etc. leaned towards neural networks for finding stuff (recommendations) and also better, but still poor. LLMs though are brilliant at extracting knowledge from data and storing only that, throwing away all the crud. This is the compression looking part of LLMs.
Using retrieval augmentation or fine tuning on your own data is very powerful. There are almost zero hallucinations and you have a new search, a search that is multi modal input against your own data. This is super powerful.
However RAG or fine tune is a step, it’s not (yet) automatic, but that is coming fast and even with the current LLM capability that is not AGI, it won’t matter, these LLMs will replace all apps we can think of now whp, apps won’t exist. It’s the thing we know, the best app just works, it has no configuration or many buttons to press, it just works and the LLM type apps will work for every human in every language for every task that can be done with its data set.
So the data set is vital and there we have global knowledge (i.e. chatgtp, bard etc.) and local knowledge (your own data, fine tuned or whatever). This is where I am right now, it’s not clear how knowledge shares or improves without bias (which are lies or at least unproven facts) and that may be much worse than hallucination.
I won’t go on, this probably is a whole new thread or even more than that. It’s the alignment of humanity to knowledge that is happening, and we need to navigate that and make sure it’s all SAFE and belongs to everyone free of charge.