Unbridled techno-optimism

So as a pragmatic optimist, I see what feels like unbridled technooptimism in statements like “AI capabilities are the worst they’ll ever be right now” and I want to introduce a little realism. (Show me something excessively negative and I’d do the opposite).

Here is my thinking -

* We’re using products, and products degrade. In fact, enshittification has been the default trend for a lot of successful software, I don’t see why AI would be exempt from that

*  a model trained on today knows nothing of next year - and left un-updated surely its usefulness declines, because the gap between what it knows and what you need will widen

* the quality of inputs for training future models is being degraded as the internet fills with AI slop - the training data of the future will be noisier, self-referential and just lower quality. It’s possible we’ve already trained on the best data we’ll ever have

* capability is relative, not absolute. What future AI can accomplish - its practical edge - depends on what future spam filters, fraud defences and human adaptation improvements there are.

I’m not anti-AI by any means.

But I believe that progress is a choice societies keep making - not a law of physics, and not inevitable by any means.

So we need to be alert to the issues and risks above, and do what we can to address them, for that original statement to turn out to have been true. It’s not insurmountable at all. We just need to be really conscious about the need for continued stewardship here.