“Very recently, I realised that maybe the digital intelligences are actually learning *better* than the brain.

With 1% of the storage capacity, GPT-4 knows thousands of times more than us. That strongly suggests its got a better way of getting information into the connections.” 

GPT-4 helps with programming & brainstorming. It’s already speeding up AI research.

When AI systems are as good at AI research as humans, Tom Davidson (@open_phil) thinks that the growth rate of the global economy will increase by >10x.

Also: this may happen within 10 years.

Key claim: you only need human-level AI researchers to get explosive growth.

When you have that, AI capabilities grow very quickly & other bottlenecks to AI-driven economic growth are quickly overcome.

Common claims from those who *disagree* with the above:

1. AI systems will *not* become capable of human-level AI research.

2. AI systems will *not* become wildly superintelligent even after a huge research effort by millions of AI researchers.

3. There are bottlenecks to GDP growth that wildly superintelligent AI systems will not overcome.

Claims (1) and (2) are looking less and less plausible.

But: coordination & regulation might make (1) true, at least for a while.

I'm doubtful about (3): it's hard to imagine concrete stories.

In the world of atoms, superhuman robots are on the way. Superintelligent systems will rediscover tacit knowledge. Coordination between AIs will be fast.

But the world economy is extremely complex, so... maybe...?

Robin Hanson argues against FOOM, but he also predicts explosive economic growth.

It seems like he broadly shares the model above, but thinks that (1) won't happen soon, and that bottlenecks will slow things down for a while.

@robinhanson is that right?

https://twitter.com/peterhartree/status/1634630734886633472

@tylercowen has said:

1. Intelligence is "multidimensional to an extreme".
2. AI systems will struggle to synthesise their skills into a coherent whole.

I don't understand the case for (2). Humans can do it—why can't AI systems? GPT-4 already does this to some degree?

The explosive growth prediction is discussed on the @80000hours podcast, with links and further reading here:

https://80000hours.org/podcast/episodes/tom-davidson-how-quickly-ai-could-transform-the-world/#articles-books-and-other-media-discussed-in-the-show


We might see 30% annual GDP growth within a decade

Some ways of responding to transformative change:

1. Fight it to your last breath.

A noble strategy, but a losing one. Memorably depicted in Bait (2019). Exemplified by the French.

2. Look for path dependencies, take opportunities to shift things in directions you like.

To some degree, we can design the stage on which evolution plays out.

Long-run, the question is: stick with the sun, or go interstellar? @robinhanson

Changes

Reading this essay with longtermist effective altruism in mind, I found myself thinking of that old Bloggingheads interview where it turns out that Tyler has thought more deeply about Peter Singer’s views than Peter Singer.

https://twitter.com/peterhartree/status/1651819231510659073

This point, in particular, seems neglected by the longtermist meme industrial complex.

https://twitter.com/peterhartree/status/1651819242717851650

And this point remains underrated.

We should take arguments seriously, but not so seriously that we ignore their context.

https://twitter.com/peterhartree/status/1651824331700224000

Tyler Cowen, longtermist OG

Greg Brockman gave a live demo of GPT-4 on TED this week.

My favourite part: he asked the model to add a 2023 projection to a graph. It did this. Then, with quiet amazement:

“If you notice, it even updates the title. I didn’t ask for that, but it knows what I want."

The main aim of the FLI letter was, presumably, to get people talking about AI safety. 

On that dimension: 10/10, wild success. https://twitter.com/peterhartree/status/1647529060334178305

“The Hebrews understood truth as commitment—they had to remain true to their particular covenant with God. Their truth was not universal and for all people all time. To live in truth required not total detachment but total commitment.”

“In Greek metaphysics, nothing important happens in time. In the Hebrew/Christian view, however, everything important happens in time.”

Dreyfus, H. (2012), “What a monster then is man”.

Truth as detachment; truth as commitment

Angel's Dance, Sally McCloskey (Nevada, 1953)

“It seems that in me, memories live on that act more powerfully than the superstructure of argumentation—a highly dubious way for a philosopher to live.”

—Jacob Taubes

"We're shocked when authority figures who are supposed to know what they're doing make it plain that they don't. We shouldn't be shocked, though.

All these stories illustrate a fundamental truth of human existence: everyone is just winging it, all the time."

—@oliverburkeman

Realising that "everyone is just winging it, all the time" was one of the most import insights of my twenties.

Not convinced? Watch the COVID testimony given to the UK Commons Inquiry by @Dominic2306.

https://www.youtube.com/watch?v=XBiCqtEOFfU

Many of the most impactful people I know describe this insight as a key moment in their formative years.

Upshot: you can probably do more than you think. You can contribute at the highest levels—they're not out of reach.

Accept the responsibility, and try.

Burkeman continues:

"Institutions—from national newspapers to governments and politicial parties—invest an enormous amount of money and effort in denying this truth. The facades they maintain are crucial to their authority, and thus to their legitimacy and continued survival."

"We need them to appear ultra-competent, too, because we derive much psychological security from the belief that somewhere, in the highest echelons of society, there are some near-infallible adults in charge."

See also: @mgurri.

https://80000hours.org/podcast/episodes/martin-gurri-revolt-of-the-public/

The quoted article is here:
https://www.theguardian.com/news/oliver-burkeman-s-blog/2014/may/21/everyone-is-totally-just-winging-it