[Rhapso]

More than that, "singularity" isn't even an exciting term. It is a boundary beyond which the model being used obviously cannot predict behavior.

I personally think the singularity happened back in the 90s and its just been very disappointing instead of the rapture that was imagined.

[MarkusQ]

You could put it anywhere you like; our ability to forecast the future has been rubbish for thousands of years. Some of my favorites: 2019, 2000, 1992, 1969, 1928, 1900, mid 1700s, ~400 BCE, ~5000 BP, ~11000 BPE.

[ben_w]

This is why I tend to phrase it as an "event horizon" rather than a "singularity" (well, this and to avoid the baggage).

There's always been some stuff we can't forecast, and some other stuff we can forecast kinda OK-ish. If you'd asked people from 11000 BPE if the sun would still be rising and crops would still be grown today, I suspect the biggest difficulty would be finding someone who really understood the meaning of the number used to express the number of years between then and now — even a "big city" back then would have been order of 1k-2k people.

But I say "sun" and "crops" because of what I expect to be the limits of their imagination. We see such limits even today — even on Hacker News, look at threads about AI taking over the world, and you will see people discussing AI as if it can only be in the form of existing current LLMs.

So, the question to ask is: how far into the future would you have to take someone from each of those years, before they're shocked by what they encounter? That's my "event horizon", because it's the horizon beyond which you cannot see, even if that horizon is one of inability to predict what specific thing will be invented rather than the general trend-lines going infinite.

The time-gap to a shocking new invention decreases as the rate of progress increases.

Obviously how much some person will be able to forecast and hence not be surprised by, will depend on how much attention they pay to technological developments — and I've seen software developers surprised by Google Translate having an AR mode a decade after the tech was first demonstrated — but there is an upper limit even for people who obsessively consume all public information about a topic: if you're reading about it when you go to sleep, will you be surprised when you wake up by all the overnight developments?

Personally, I've tried making my own forecasts about the future, I'm fairly consistent for the last decade that my models stop making sense around 2032. But even with that consistency, I could only see ChatGPT coming when InstructGPT was shown off in a Two Minute Papers video (https://www.youtube.com/watch?app=desktop&v=PmxhCyhevMY&t=4m...) and the rate of change of Transformer and Diffusion models has continued to surprise me.

(I also didn't predict how severely YouTube would be stuffed with adverts, but that's more of a cultural shift — everyone gets all "kids these days" as they get older, why would I be any different?)