In which I declare my expert judgment on AI 2024

These days my social media experience is largely Mastodon. There's something to be said about a social network that's so irreparably geeky and so hard to work with that only a tiny slice of humanity can possibly participate (unless and until Threads integration actually works).

In my Mastodon corner of the "Fediverse', among the elite pundits I choose to read,  there's a vocal cohort that is firm in their conviction that "AI" hype is truly and entirely hype, and that the very term "AI" should not be used. That group would say that the main application of LLM technology is criming.

Based on my casual polling of my pundits there's a quieter cohort that is less confident. That group is anxious, but not only about the criming.

Somewhere, I am told, there is a third group that believes that godlike-AIs are coming in 2025. They may be mostly on Musk's network.

Over the past few months I think the discourse has shifted. The skeptics are less confident, and the godlike-AI cohort is likewise quieter as LLM based AI hits technical limits. 

The shifting discourse, and especially the apparent LLM technical limitations, mean I'm back to being in the murky middle of things. Where I usually sit. Somehow that compels me to write down what I think. Not because anyone will or should care [1], but because I write these posts mostly for myself and I like to look back and see how wrong I've been.

So, in Aug 2024, I think:

1. I am less worried that the end of the world is around the corner. If we'd gotten one more qualitative advance in LLM or some other AI tech I'd be researching places to (hopelessly) run to.
2. Every day I think of new things I would do if current LLM tech had access to my data and to common net services. These things don't require any fundamental advances but they do require ongoing iteration.  I don't have much confidence in Apple's capabilities any more, but maybe they can squeeze this out. I really, really, don't want to have to depend on Microsoft. Much less Google.
3. Perplexity.ai is super valuable to me now and I'd pay up if they stopped giving it away. It's an order of magnitude better than Google search.
4. The opportunities for crime are indeed immense. They may be part of what ends unmediated net access for most people. By far the best description of this world is a relatively minor subplot in Neal Stephenson's otherwise mixed 2018 novel "Fall".
5. We seem to be replaying the 1995 dot com crash but faster and incrementally. That was a formative time in my life. It was a time when all the net hype was shown to be .... correct. Even as many lost their assets buying the losers.
6. It will all be immensely stressful and disruptive and anxiety inducing even though we won't be doing godlike-AI for at least (phew) five more years.
7. Many who are skeptical about the impact of our current technologies have a good understanding of LLM tech but a weak understanding of cognitive science. Humans are not as magical as they think.

- fn -

[1] I legitimately have deeper expertise here than most would imagine but it's ancient and esoteric.