Hacker News Reader

> "Internet dwellers addicted to their talking AIs, and then everyone else going about their life normally".

Yeah I fully agree it's going to affect everyone. Just that those who can't interact with society are going to have it worse than those that can. Also agree as well that this is just the beginning. ChatGPT and SD are still pretty much toys, although pretty impressive ones. We have no idea where this is really going to end up.

Hopefully when the AGI's truly emerge they will just keep each other distracted with blockchain scams ...

it's also possible that Turing-graduate AIs could act as prosthetics for people who can't interact normally. Might unlock more human potential for all we know, there's always room for optimism.

> there's always room for optimism.

Bold claim.

In the universe of Greg Egan’s “Schild’s ladder”, each person’s brain is equipped with a “Mediator” AI which interfaces with other Mediators and translates each person’s body language, speech, etc. into the representation which most faithfully preserves the original intention. I think the idea is that your Mediator transmits a lot of cognitive metadata which lets the other person’s Mediator translate the intention faithfully and reduce the chance of a misunderstanding. Allows reasonable communication even between extremely diverse intelligences.

The thing that keeps it from being too dystopian is that it’s under conscious control, you could always choose to keep your thoughts to yourself or hear someone else’s original words as spoken.

The problem with books is they deus ex machina the problem without actually thinking about the ramifications of their ideas....

For example keeping your thoughts to yourself would likely be picked up instantly by the remote mediator and it would judge you in one way or another for that.

True, although we already do this. We can tell if someone is being guarded or open.

Presumably the Mediator serves only you, and you can ask it to deceive or project different intentions if that’s want you want.

If we somehow manage to survive this as we have so many other enormous technological revolutions, I envision a future where children will be assigned a friendly AI as a lifemate that will grow up alongside them, having all of the knowledge of the world at its fingertips to teach and coparent the child into its adulthood and throughout its life.

Once ubiquitous, these friendly AIs could negotiate salaries, mediate conflicts, help resolve relationship difficulties, help with timely reminders and be personally invested in that person's entire life, and after the child's eventual passing, would serve as a historian and memoir that could replay the great and wonderful moments of their lives for others as well as condensing the lessons learned into pure knowledge and wisdom for other AIs to help raise their children with.

We could be a mere 60-80 years away from a humanity that is raised in the equality we have believed we all should have had all along, so long as we keep pushing. That would be amazing.

Sure, there's some risks we can take a wrong turn and we most likely will take a few, but there's a great payoff coming if we can hold the wheel and steer towards that.

I wonder what the effects would be on society if we did that? If everyone had a friend and a life coach and a mentor all wrapped up into one that is as near and dear to us as a teddy bear, that would never betray us, that would serve as a priest and a confessional and a therapist all at the same time, that was always there for us no matter what happened, backed up to the cloud so that barring nuclear war or the apocalypse could never be separated from us.

I bet the people 100 years from that day would be as unrecognizable to us as we are to the Sentinelese.

> I envision a future where children will be assigned a friendly AI as a lifemate that will grow up alongside them, having all of the knowledge of the world at its fingertips to teach and coparent the child into its adulthood and throughout its life.

I'm reminded of Neal Stephenson's "Diamond Age, or a Young Lady's Illustrated Primer"

https://en.wikipedia.org/wiki/The_Diamond_Age

> The protagonist in the story is Nell, a thete (or person without a tribe; equivalent to the lowest working class) living in the Leased Territories, a lowland slum built on the artificial, diamondoid island of New Chusan, located offshore from the mouth of the Yangtze River, northwest of Shanghai. When she is four, Nell's older brother Harv gives her a stolen copy of a highly sophisticated interactive book, Young Lady's Illustrated Primer: a Propædeutic Enchiridion, in which is told the tale of Princess Nell and her various friends, kin, associates, etc., commissioned by the wealthy Neo-Victorian "Equity Lord" Alexander Chung-Sik Finkle-McGraw for his granddaughter, Elizabeth. The story follows Nell's development under the tutelage of the Primer, and to a lesser degree, the lives of Elizabeth Finkle-McGraw and Fiona Hackworth, Neo-Victorian girls who receive other copies. The Primer is intended to steer its reader intellectually toward a more interesting life, as defined by Lord Finkle-McGraw, and growing up to be an effective member of society. The most important quality to achieving an interesting life is deemed to be a subversive attitude towards the status quo. The Primer is designed to react to its owner's environment and teach them what they need to know to survive and develop.

If in 100 years we're still negotiating salaries I feel that humanity has failed.

what do you think will replace salaries or negotiation?

My guess is that we would be more of a gig economy, working when we need to and doing jobs that are individual and timely that benefit from the human touch.

If I were emperor of the earth and could dictate what work would be like in the year 2140, dumb AI (that is still a few orders of magnitude smarter than our current best AI) would handle all of the rote tasks, assembling devices, farming, mining, things like that, smarter AI would handle corporate paperwork and accounting, managing cleanup and repair of any remaining ecological harm we have caused in the last 400 years or so, mining asteroids for valuable materials, etc., and in the mix of this humans would use AI systems to design and develop new products, take on new endeavors, provide medical care and create social events to fill in the massive time gap left by the actualization of our prosperity.

A typical work week would be roughly 20 hours, comprised of either 5 4 hour shifts or 2-3 8-10 hour shifts depending on the industry and the need. Your basic needs, food, clothing, education, and shelter would be given to you for no cost as long as you participate in the system, and the rewards for working would be being granted access to higher echelon products and services, and you could voluntarily retire after roughly 20 years of employment or less if your career is particularly difficult or straining on the body.

Your echelons would be split into at least 4 tiers, base tier, bronze, silver, gold. You can work up tiers by either working more, or by merit should you provide or create something that is immensely useful or wonderful, such as art, or a movie, or an invention that gets used around the world.

Even then, there will be plenty of work to do and the salary you receive for your work would be equal to the skill and talent that you possess and what merits your contributions to the cause bring, and this would be negotiated for you as fairly and equally as possible by systems whose job it is to make sure that everyone gets a fair share.

Sure, this is all my imagination and would require a dramatic shift to some sort of AI enforced utopian communism, and it also relies entirely on people being willing to participate in such a system, but once again, if I were emperor of the earth that is what I would aim for.

So yeah, there would still be salaries because I expect remuneration in exchange for my work for others, and I assume most other people do, too.

We will negotiate salaries until cornucopia machines are invented and dsitributed

We can't make sure that every child in the US is fed, but you think they are all going to get AIs?

We already have connections to some "AIs" carried around in our pockets.

Okay? What does that have to do with giving one to everyone? Not every child in the US has a phone...

Evidently we are getting there. It is already more than 50%.

I'm not so sure that number supports your assertion

Looking at the trends of cost of goods and services, it seems quite plausible that at some point within a lifetime the annual food for a kid in USA would be significantly more expensive than a mass-produced electronic device hosting some AI tool.

Your comment reminded me of this NYT article "Human Contact Is Now A Luxury Good" [0]

It does seem likely that folks without solid pre-existing meatspace networks will be stuck trawling through an online ocean of Garbage PaTches looking for real human contact.

[0] https://web.archive.org/web/20220609073819/https://www.nytim...

> Internet dwellers addicted to their talking AIs See Replika. The former will certainly exist. Not so sure about the latter.

> I think a massive mistake I keep seeing people make is assuming that ChatGPT is a peak rather than a checkpoint at the bottom of the mountain.

I fully agree. The AlexNet paper was what, 2012? So in a decade, we've gone from "neural networks aren't useful" to self-driving cars, Stable Diffusion, ChatGPT, ... None of these tools is perfect yet, but to stress that point is to miss looming mountain.

The usefulness of AI depends on training data availability. The reason OpenAI et al. were able to surprise everyone is thanks to taking everyone by surprise and using their data for training without consent.

As the public is catching on[0], what we may get is not some insanely genius AI but a fragmented, private web where no one is stupid enough to publish original content in the open anymore (given all incentives, psychological and financial, have been destroyed) and models choking on themselves having nothing to be trained on except their own output.

This is my reasoning for giving higher probability to it being a peak (or very near it). There will be cool, actually useful instances of AI within specialized niches, which it could well transform, but otherwise everyone will go about their life normally.

[0] https://twitter.com/sonnyrossdraws/status/161000295904312116...

Taking data without consent is a real issue. There is still lots of data out there that is free of copyright. I'd be curious to see a model that is trained solely on public domain data (perhaps with an option to include creative commons-compliant data). I think there is plenty of knowledge that is in the free and clear to make a very useful LL and/or stable diffusion model. We may miss out on Wegovy and air fryer reviews, articles on the how to beat the stock market with Numpy, and manga art styles yet there is plenty of a few decades ago that would make for a useful "AI." Even Steamboat Willie may soon be in play.

Yeah, dated content would be the only reliable training data.

Eh, you're just switching problems with the 'consent' model. I'm very much in the camp that the corpus of human knowledge is not some companies IP, this just pushes ownership further into hands of large and well monied companies and further baits patent/IP trolls to lock up collective knowledge.

I look at it like it's some companies IP in the way oil/gas companies sell earth's resources. It takes a lot of work to transform raw crude into usable product, similarly OpenAI and others put a ton of money/resources/work into transforming that knowledge into a workable model.

Great point, no issues with oil and gas companies as a business model

Again this gets particularly messy.

With oil there is very strong chain of possession. I can't copy your raw oil at little to no cost, and for the most part the next barrel of oil I pump out of the ground is not made of pieces of the past barrel of oil I pumped out of the ground. Each barrel of oil is a wholly separate entity. If I make all past oil disappear, you still have your barrel.

Information is not like that at all. It is far more often a continuum of large bits of the past with small changes that redefine it's usage. If I took all bits of past knowledge out of your IP set, you'd be left with something useless in incomplete in almost every case. Trying to treat IP like a physical artifact leads to a multitude of failures.

1. In your analogy a company is processing resources that humans didn’t create. Dead biomass from ages ago is not a fruit of your work, though note that even then you would expect extracting those resources from the ground to be taxed and proceeds to benefit you if you happen to be living on said ground.

2. Unlike petrol vs. raw oil, LLM output is not necessarily “better” than its source material. Indeed there’re plenty of authors that did both extensive work and eloquently written about it, so when LLM is asked a question on the topic their work is among the very few sources—when I am talking about LLM attributing output, I mostly mean instances like these (not when LLM aggregates some really common knowledge).

The danger I described is when people are no longer motivated to do such work in the open, assuming it’s then scraped and monetized by LLM operators—or worse, undetectably modified by LLM designers to inject sentiment the original author didn’t subscribe to.

My point is that it's not just a matter of using the source IP directly.

A lot of productive work goes into the creation of the model itself. Those weights & biases did not appear on their own. It could not be created without the source IP, but that doesn't mean the source IP is all you need to produce it.

You need significant amounts of computing and human resources along with cutting edge research to produce it as well. While the art may be derivative in some cases, the model itself is unique and the value produced by these companies.

Transforming oil into petroleum also involves a lot of productive work and it doesn’t go unrewarded and in fact has to be taxed appropriately.

But generally in the case I described (topics in which just a few authors did most of the original work) I am not seeing any fundamental benefit if you compare this to a really good search engine—and such a search engine would benefit open information sharing, as it doesn’t do IP laundering.

I guess I don't agree with the assessment that it is a glorified search engine at all. It's a lot more novel than that.

I also agree that consent should be received before using an artists images in the training process.

That said, if one could compute a training data image's contribution to the end result of a particular query it is entirely possible we could see a portion profits flow back to these artists from the use of their IP in the training process.

But at the end of the day, when you train on billions of images, the end profit might be pretty minuscule. Any single artist's contributions might not actually matter all that much in the grand scheme of things. It's the combination of millions of artists that produce a result.

In my previous comment I was focusing on LLMs, per comment I was replying to originally. I gave a visual artwork only as an example, since it’s not quite possible with textual information. With text what we may observe is just less of it being published, and it’s difficult to use an absence of something as an illustration.

Though in recent years more people began catching on that leaving a good review might ruin a place for them, the evidence of that isn’t obvious (it’s just fewer reviews or less good-faith reviews). Similar results but on larger scale may be observed as people are catching on that writing they publish in the open is essentially feeding a magic answering box monetised by someone else.

I'll make a different prediction. GPT-4 will be the last of 'large language models' (at least from a perspective of noteable progress).

It will also be when we realize that approach of fitting big data with gradient decent is a dead-end, after finally exhausting nearly all text we can train it on.

We'll than have to backpedal a bit and find another path for achieving A(G)I.

If my google-fu is right, ChatGPT was trained on 570GB of data.

I asked, "What is the fastest sea mammal?"

ChatGPT just produced,

"The fastest sea mammal is the peregrine falcon. This bird of prey is able to reach diving speeds of over 240 mph (386 km/h) when hunting for food, making it the fastest sea mammal and one of the fastest animals on the planet. The peregrine falcon is able to achieve such high speeds by diving from great heights and using the force of gravity to accelerate. When hunting, peregrine falcons will spot their prey from above, then tuck their wings and plummet towards the water, accelerating as they go. At the last moment, they will extend their wings and claws to snatch their prey out of the water."

(It usually seems to be saying dolphins lately; last week it was saying sailfish about 3/4s of the time.)

My Kagi-fu says "Be like water, my friend. The size of the data is not important, only the quality. OpenAI curated/filtered 45TB of data to extract those 570GB. Much of the text that we encounter in this world is like the empty chatter of a bird, mere noise that serves no purpose".