AI Forecasting: One Year In | Hacker News Reader

AI Forecasting: One Year In

bounded-regret.ghost.io

42 pointssebg8 comments

For all the hollowness of crypto, AI finally feels like it is building toward something meaningful.

Large Language Models like GPT-3 definitely aren't sentient (or even coherent often), but it's definitely more than what I could figure out how to code with vanilla python, which could not be said of the much simpler chat bot type systems 10 years ago.

Interesting and exciting things happen anytime a new tool is built and added to the toolkit.

AI poses to be 9999999x more useful and interesting than even the most idealized crypto real-world use cases. The push for crypto in the last decade has been motivated largely by the allure of get-rich-quick schemes, not actual utility.

Cryptocurrency is extremely valuable. It offers a safe and reliable path for scammers, drug lords, and sex traffickers to transfer funds. A degenerate's dream

This is a false equivalence though? Technologically speaking, nothing actually practically useful has been proposed since the original Bitcoin paper. Artificial intelligence is just by its very definition of the two words the future of mankind whether we like it or not. Of course much can be said about our current interpretation of what AI means but “crypto” (which is annoying since the bros have usurped an acronym that should belong to all cryptography) is literally just a fad.

I was one of the forecasters in this market, and my forecasts were blown out of the water.

One thing to note that this blog post skips over is that the questions placed ceilings on the standard deviation of submitted predictions, which means that individual forecasts were constrained to be artificially narrow. Therefore, don't read as much into the distributions as the means.

Indeed, although this is actually mentioned at the end of the blog post.

> Hypermind’s interface has some limitations that prevent outputting arbitrary probability distributions. In particular, in some cases there is an artificial limit on the possible standard deviations, which could lead credible intervals to be too narrow.

We're going to see even quicker advances as more people jump into the field after the big recent successes, like image generation & text generation.

We're going to build some pretty incredible things.

Not a given. Adding more warm bodies has never made a difference especially in research. All of modern academia is an example.

Reply on news.ycombinator.com