Comment by sxp | Hacker News Reader

Comment by sxp | Hacker News Reader

Parent Full thread

> This article was for the VISION-21 Symposium sponsored by NASA Lewis Research Center and the Ohio Aerospace Institute, March 30-31, 1993.

> Within thirty years, we will have the technological means to create superhuman intelligence. Shortly after, the human era will be ended.

Impressive timing on that prediction assuming GPT-4's release on March 14th counts as "superhuman intelligence".

I read Vinge's essay in college and it was a critical factor in affecting my career choices and transhumanist views. His sci-fi Fire Upon the Deep/Deepness in the Sky and Fast Times at Fairmont High/Rainbows End were also critical in affecting my career path into AR/VR. I highly recommend reading this essay and any of his other sci-fi. He work could be classified as "hard computer science fiction" because it is heavily influenced by his career in CS. When I was at Google X around 2011, I realized that almost all the stuff they were working on in secret at the time (AR glasses, Drone delivery, self-driving cars, balloon Internet with lasers, computers embedded in fabric, etc) were copied directly out of Fast Times at Fairmont High. Fast Times and Rainbows End aren't available online but I suggest Synthetic Serendipity which is set in the same universe and has concepts such as a gig economy (called "affiliate tasks"): https://spectrum.ieee.org/synthetic-serendipity

> Impressive timing on that prediction assuming GPT-4's release on March 14th counts as "superhuman intelligence".

How many legs does a dog have, if you assume a tail is a leg?

Four. Assuming it doesn't make it so.

In the same way, you can assume whatever you want, but GPT-4 is not "superhuman intelligence". It just isn't, for any meaningful definition of those words.

>> In the same way, you can assume whatever you want, but GPT-4 is not "superhuman intelligence". It just isn't, for any meaningful definition of those words.

It's also not 2030 yet, so the prediction has 7 more years to be false in its full form. If it happens in 2040 will people be upset that it was 10 years later than predicted?

1993 + 30 = 2023

He specifically gives the range 2005-2030 in the paper :)

Get outta here with your witchcraft!

Yes, that's what ChatGPT says: Assuming that a tail is a leg does not actually make it a leg. Dogs have four legs, regardless of the length of their tail.

So the fact that it might have outsmarted you means it is not intelligent?

I think humans tend to have a definition of intelligence that is too far fetched in my opinion. They confluate it with things like conscience, meaningfulness etc, which just might be human hallucinations.

> assuming GPT-4's release on March 14th counts as "superhuman intelligence"

If it did, nobody here would have a job and we would be already deep into the "reorient society or fight the Machine" phase.

There are some in the 'slow takeoff' school of AI speculation that disagree.

For example what if there is a five billion dollar supercomputer data center that can simulate a single intelligence that is say in the top 10% of smartness of people and it outputs one word every five seconds.

In that example, even though it is superintelligent there would be a limit to how fast it can make more of its own data centers. And even if it can research its own development to make itself more efficient, there is probably some lower bound on the hardware required. Like no matter how smart the superintelligence is, it's probably not going to be reducing its own size to fit into everyones pre-existing phones. And even if it can do its own hardware research it would still have to be implemented, by some kinds of workers or robots. And those robots don't get made instantly.

Probably Eliezer Yudkowsky has a galaxy brain reason why this is wrong?

If it is based on an LLM trained on the internet then it is probably outputing "gib moar srvrs plz" every twenty seconds.

Reply on news.ycombinator.com