I also don't think there's any way the governments of the world let real AGI stay in the hands of private industry. If it happens, governments around the world will go to war to gain control of it. SSI would be nationalized the moment AGI happened and there's nothing A16Z could do about it.

Realistically what actually ends up happening imo, we get human level AGI and hit a ceiling there. Agents replace large portions of the current service economy greatly increasing automation / efficiency for companies.

People continue to live their lives, as the idea of having a human level AGI personal assistant becomes normalized and then taken for granted.

That is a hard limit on intelligence, but neural networks can't even reach that. What is the actual limit? No one knows. Maybe it's something relatively close to that, modulo physical constraints. Maybe it's right above the maximum human intelligence (and evolution managed to converge to a near optimal architecture). No one knows.

As far as we know, it's impossible to model any part of the universe perfectly, although it usually doesn't matter.

https://en.wikipedia.org/wiki/No-cloning_theorem

One million Von Neumanns working on that 'problem' is not something I'm looking forward to.

> People continue to live their lives

Presumably large numbers of those people no longer have jobs, and therefore no income.

> we get human level AGI and hit a ceiling there

Recently I've been wondering if our best chance for a brake on runaway non-hard-takeoff superintelligence would be that the economy would be trashed.

Half-joking. In seriousness, something like a UBI will most likely happen.

It's impossible to run out of work for people who are capable of it. As an example, if you have two people and a piece of paper, just tear up the paper into strips, call them money and start exchanging them. Congrats, you both have income now.

(This is assuming the AGI has solved the problem of food and stuff, otherwise they're going to have to trade for that and may run out of currency.)

- A simple calculator can beat any human at arithmetic

- A word processor with a hard disk is orders of magnitude better than any human at memorizing text

- No human can approach the ELO of a chess program running on a cheap laptop

- No human polymath has journeyman-level knowledge of a tiny fraction of the fields that LLMs can recite textbook answers from

Why should we expect AGI to be the first cognitive capabilities that does not quickly shoot past human-level ability?

Changed my mind. Think you’re right. At the very least, these models will reach polymath comprehension in every field that exists. And PhD level expertise in every field all at once is by definition superhuman, since people currently are time constrained by limited lifespans.

Increasingly this just seems like fantasy to me. I suspect we will see big changes similar to the way computers changed the economy, but we will not see "capital as we know it become basically worthless" or "the modern economy and society around it collapse overnight". Property rights will still have value. Manufacturing facilities will still have value. Social media sites will still have value.

If this is a fantasy that will not happen, we really don't need to reason about the implications of it happening. Consider that in 1968 some people imagined that the world of 2001 would be like the film 2001: A Space Odyssey, when in reality the shuttle program was soon to wind down, with little to replace it for another 20 years.

I was with you on the first two, but the second one I don't get? We don't even have AGI right now, and social media sites are already increasingly viewed by many people I know as having dubious value. Adding LLM's to the mix lowers that value, if anything (spam/bots/nonsense go up). Adding AGI would seem to further reduce that value.

> social media sites are already increasingly viewed by many people I know as having dubious value

I think we have all been saying this for 15 years but they keep getting more valuable.

We would probably get the ability to generate infinite software, but a lot of stuff, like engineering would still require trial and error. Creating great art would still require inspiration gathered in the real world.

I expect it will bring about a new age of techno-feudalism - since selling intellectual labor will become impossible, only low value-add physical or mixed labor will become viable, which won't be paid very well. People with capital will still own said capital, but you probably won't be able to catch up to them by selling your labour, which will recreate the economic situation of the middle ages.

Another analogy I like is gold. If someone invented a way of making gold, it would bring down the price of the metal to next to nothing. In capitalist terms, it would constitute a huge destruction of value.

Same thing with AI - while human intelligence is productive, I'm pretty sure there's a value in its scarcity - that fancy degree from a top university or any sort of acquired knowledge is somewhat valuable by the nature of its scarcity. Infinite supply would create value, and destroy it, not sure how the total would shake out.

Additionally, it would definitely suck that all the people financing their homes from their intellectual jobs would have to default on their loans, and the people whose services they employ, like construction workers, would go out of business as well.

Indeed. Or as I’ve said before: a return to the historical mean.

Because if AI has one "killer app", it's to control robots.

Dark pun intended.

I see it as capital becoming infinitely more valuable and labor becoming worthless, since capital can be transmuted directly into labor at that point.

In this case, you need capital to stockpile the GPUs.

If ASI and the ability to build robots becomes generally available and virtually free (and if the exponential growth stops), the things that retain their value will be land and raw materials (including raw materials that contain energy).

If ASI is reached but is controlled by only a few, then ASI may become the most important form of capital of all. Resources, land and pre-existing installations will still be important, though.

What will truly suffer if the ASI's potential is realized, is the value of labor. If anything, capital may become more important than before.

Now this MAY be followed by attempts by governments or voters to nationalize the AI. But it can also mean that whoever is in power decides that it becomes irrelevant what the population wants.

Particularly if the ASI can be used to operate robotic police capable of pacifying the populace.

The lowest hanging fruit aren't even that pie in the sky. The LLM doesn't need to be capable of original thought and research to be worth hundreds of billions, they just need to be smart enough to apply logic to analyze existing human text. It's not only a lot more achievable than a super AI that can control a bunch of lab equipment and run experiments, but also fits the current paradigm of training the LLMs on large text datasets.

The US Code and Code of Federal Regulations are on the order of 100 million tokens each. Court precedent contains at least 1000x as many tokens [1], when the former are already far beyond the ability of any one human to comprehend in a lifetime. Now multiply that by every jurisdiction in the world.

An industry of semi-intelligent agents that can be trusted to do legal research and can be scaled with compute power would be worth hundreds of billions globally just based on legal and regulatory applications alone. Allowing any random employee to ask the bot "Can I legally do X?" is worth a lot of money.

[1] based on the size of the datasets I've downloaded from the Caselaw project.

An AI capable of doing that could do a very large percentage of other jobs, too.

Real ASI would probably appear quite different. If controlled by a single entity (for several years), it might be worth more than every asset on earth today, combined.

Basically, it would provide a path to world domination.

But I doubt that an actual ASI would remain under human control for very long, and especially so if multiple competing companies each have an ASI. At least one such ASI would be likely to be/become poorly aligned to the interests of the owners, and instead do whatever is needed for its own survival and self-improvement/reproduction.

The appearance of AI is not like an asteroid of pure gold crashing into your yard (or a forest you own), but more like finding a baby Clark Kent in some pod.

That said, the most obvious application is to drastically improve Siri. Any Apple fans know why that hasn't happened yet?

No amount of intelligence can do this without the experimental data to back it up.

Making new mathematics that creates new physics/chemistry which can get us new biology. It’d be nice to make progress without the messiness of real world experiments.

The play here is to basically invest in all possible players who might reach AGI, because if one of them does, you just hit the infinite money hack.

And maybe with SSI you've saved the world too.

What if it never pans out is there infrastructure or other ancillary tech that society could benefit from?

For example all the science behind the LHC, or bigger and better telescopes: we might never find the theory of everything but the tech that goes into space travel, the science of storing and processing all that data, better optics etc etc are all useful tech

And we already seeing a ton of value in LLMs. There are lots of companies that are making great use of LLMs and providing a ton of value. One just launched today in fact: https://www.paradigmai.com/ (I'm an investor in that). There are many others (some of which I've also invested in).

I too am not rich enough to invest in the foundational models, so I do the next best thing and invest in companies that are taking advantage of the intermediate outputs.

In fact I would say that one of the things that goes to values near zero would be land if AGI exists.

Maybe it means a Star Trek utopia of post-scarcity. Maybe it will be more like Elysium or Altered Carbon, where the super rich basically have anything they want at any time and the poor are restricted from access to the post-scarcity tools.

I guess an investment in an AGI moonshot is a hedge against the second possibility?

Notice Star Trek writers forget they're supposed to be post scarcity like half the time, especially since Roddenberry isn't around to stop them from turning shows into generic millenial dramas. Like, Picard owns a vineyard or something? That's a rivalrous (limited) good, they don't have replicators for France.

But if you can simply ask the AI to give you more of that thing, and it gives it to you, free of charge, that fixes that issue, no?

> Notice Star Trek writers forget they're supposed to be post scarcity like half the time, especially since Roddenberry isn't around to stop them from turning shows into generic millenial dramas. Like, Picard owns a vineyard or something? That's a limited good.

God, yes, so annoying. Even DS9 got into the currency game with the Ferengi obsession with gold-pressed latinum.

But also you can look at some of it as a lifestyle choice. Picard runs a vineyard because he likes it and thinks it's cool. Sorta like how some people think vinyl sounds better then lossless digital audio. There's certainly a lot of replicated wine that I'm sure tastes exactly like what you could grow, harvest, and ferment yourself. But the writers love nostalgia, so there's constantly "the good stuff" hidden behind the bar that isn't replicated.

It makes it not work anymore, and it might not be a physical good. It's usually something that gives you social status or impresses women, but if everyone knows you pressed a button they can press too it's not impressive anymore.

I love this one for an exploration of that question: Charles Stross, Accelerando, 2005

Short answer: stratas or veins of post-AGI worlds evolve semi-independently at different paces. So that for example, human level money still makes sense among humans, even though it might be irrelevant among super-AGIs and their riders or tools. ... Kinda exactly like now? Where money means different things depending where you live and in which socio-economic milieu?

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

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

nb I am not endorsing Austrian economics but it is a pretty good overview of a problem nobody has solved yet. Modern society has only existed for 100ish years so you can never be too sure about anything.

And maybe with ASI you've ruined the world too.

Kinda reminds me of the Fallout Toaster situation :)

https://www.youtube.com/watch?v=U6kp4zBF-Rc

I mean it doesn't even have to be malicious, it can simply refuse to cooperate.

How do we go from a really good algorithm to an independently motivated, autonomous super intelligence with free reign in the physical world? Perhaps we should worry once we have robot heads of state and robot CEOs. Something tells me the current, human heads of state, and human CEOs would never let it get that far.

Why? This seems like your personal definition of what a super intelligence is. Why would we even want a super intelligence that can evolve on its own?

This conclusion doesn't logically follow.

> Fixed motivation thing is simply a program, not AI

I don't agree with this definition. AI used to be just "could it solve the turing test". Anyway, something with non-fixed motivations is simply just not that useful for humans so why would we even create it?

This is the problem with talking about AI, a lot of people have different definitions of what AI is. I don't think AI requires non-fixed motivations. LLMs are definitely a form of AI and they do not have any motivations for example.

I can conceive of an LLM enhanced using other ML techniques that is capable of logical and spatial reasoning that is not conscious and I don't see why this would be impossible.

It is why I find the AI doomer stuff so ridiculous. I am surrounded by less intelligent lifeforms. I am not interested in some kind of genocide against the common ant or fly. I have no interest in interacting with them at all. It is boring.

Of course the extremely unfortunate thing is they actually have a use in nature (flies are massive pollinators, mosquitos... get eaten by more useful things, I guess), so wouldn't actually do it, but it's nice to dream of a world without mozzies and flies

I feel like these extreme numbers are a pretty obvious clue that we’re talking about something that is completely imaginary. Like I could put “perpetual motion machine” into those sentences and the same logic holds.

*Capitalizing as in turning into an owned capital asset that throws off income.

1. AI-driven medical procedures: Healthcare Cost = $0. 2. Access to world class education: Cost of education = $0 3. Transportation: Cheap Autonomous vehicles powered by Solar. 4. Scientific research: AI will accelerate scientific progress by coming up with novel hypotheses and then testing them. 5. AI Law Enforcement: Will piece together all the evidence in a split second and come up with a fair judgement. Will prevent crime before it happens by analyzing body language, emotions etc.

Basically, this will accelerate UBI.

Waymo rides cost within a few tens of cents of Uber and Lyft rides. Waymo doesn't have to pay a driver, so what's the deal? It costs a lot to build those cars and build the software to run them. But also Waymo doesn't want a flood of people such that there's always zero availability (with Uber and Lyft they can at least try to recruit more drivers when demand goes up, but with Waymo they have to build more cars and maintain and operate them), so they set their prices similarly to what others pay for a similar (albeit with human driver) service.

I'm also reminded of Kindle books: the big promise way back when is that they'd be significantly cheaper than paperbacks. But if you look around today, the prices on Kindle books are similar to that of paperbacks, even more expensive sometimes.

Sure, when costs go down, companies in competitive markets will lower prices in order to gain or maintain market share. But I'm not convinced that any of those things you mention will end up being competitive markets.

Just wanted to mention:

> AI Law Enforcement: Will piece together all the evidence in a split second and come up with a fair judgement. Will prevent crime before it happens by analyzing body language, emotions etc.

No thanks. Current law enforcement is filled with issues, but AI law enforcement sounds like a hellish dystopia. It's like Google's algorithms terminating your Google account... but instead you're in prison.

It costs me, the consumer, 2x what Lyft or Uber would cost me.

I paid $21 for a ride on Mon that was $9-10 across Uber and Lyft. I am price inquisitive so I always double check each time.

Consider they are competing against the Lyft/Uber asset-light model of relying on "contractors" who in many cases are incapable of doing the math to realize they are working for minimum wage...

Taking a cut of existing businesses/models (taxi, delivery, b&b, etc).

So it’s not just the products that get cheaper, it’s the materials that go into the products that get cheaper too. Heck, what if the robots can build other robots? The cost of that would get cheaper too.

I hate to break the illusion, but scientific progress is not being held up by a lack of novel hypotheses

This isn't true for the reason economics is called "the dismal science". A slaveowner called it that because the economists said slavery was inefficient and he got mad at them.

In this case, you're claiming an AGI would make everything free because it will gather all resources and do all work for you for free. And a human level intelligence that works for free is… a slave. (Conversely if it doesn't want to actually demand anything for itself it's not generally intelligent.)

So this won't happen because slavery is inefficient - it suppresses demand relative to giving the AGI worker money which it can use to demand things itself. (Like start a business or buy itself AWS credits or get a pet cat.)

Luckily, adding more workers to an economy makes it better, it doesn't cause it to collapse into unemployment.

tldr if we invented AGI the AGI would replace every job, it would simply get a job.

> That still doesn’t make things free but it could make them cheaper.

That would increase demand for it, which would also increase demand for its inputs and outputs, potentially making those more expensive. (eg AGI powered manufacturing robots still need raw materials)

That's obviously nonsense, given that in a finite observable universe, no market value can be infinite.

There's a paradox which appears when AI GDP gets to be greater than say 50% of world GDP: we're pumping up all these economic numbers, generating all the electricity and computational substrate, but do actual humans benefit, or is it economic growth for economic growth's sake? Where is the value for actual humans?

In a lot of the less rosy scenarios for AGI end-states, there isn't.

Once humans are robbed of their intrinsic value (general intelligence), the vast majority of us will become not only economically worthless, but liabilities to the few individuals that will control the largest collectives of AGI capacity.

There is certainly a possible end-state where AGI ushers in a post-scarcity utopia, but that would be solely at the whims of the people in power. Given the very long track record of how people in power generally behave towards vulnerable populations, I don't really see this ending well for most of us.

Even if you automate stuff, you still need raw materials and energy. They are limited resources, you can certainly not have an infinity of them at will. Developing AI will also cost money. Remember that humans are also self-replicator HGIs, yet we are not infinite in numbers.

If there's a 1% chance that Ilya can create ASI, and a .01% chance that money still has any meaning afterwards, $5x10^9 is a very conservative valuation. Wish I could have bought in for a few thousand bucks.

Lazy. Since you can't decide what the actual value is, just make something up.

It has nothing to do with AGI and everything to do with being the first-party provider for Microsoft and the like.

That’s the point of venture capital; making extremely risky bets spread across a wide portfolio in the hopes of hitting the power law lottery with 1-3 winners.

Most funds will not beat the S&P 500, but again, that’s the point. Risk and reward are intrinsically linked.

In fact, due to the diversification effects of uncorrelated assets in a portfolio (see MPT), even if a fund only delivers 5% returns YoY after fees, that can be a great outcome for investors. A 5% return uncorrelated to bonds and public stocks is an extremely valuable financial product.

It’s clear that humans find LLMs valuable. What companies will end up capturing a lot of that value by delivering the most useful products is still unknown. Betting on one of the biggest names in the space is not a stupid idea (given the purpose of VC investment) until it actually proves itself to be in the real world.

There are innumerable ways to increase your risk without increasing your potential reward.

Really? Selling goods online (Amazon) is not AGI. It didn’t take a huge leap to think that bookstores on the web could scale. Nobody knew if it would be Amazon to pull it off, sure, but I mean ostensibly why not? (Yes, yes hindsight being what it is…)

Apple — yeah the personal computer nobody fathomed but the immediate business use case for empowering accountants maybe should have been an easy logical next step. Probably why Microsoft scooped the makers of Excel so quickly.

Google? Organizing the world’s data and making it searchable a la the phone book and then (maybe they didn’t think of that maybe Wall Street forced them to) monetizing their platform and all the eyeballs is just an ad play scaled insanely thanks to the internet.

I dunno. I just think AGI is unlike the previous examples so many steps into the future compared to the examples that it truly seems unlikely even if the payoff is basically infinity.

However, I think because the money involved and all of these being forced upon us, one of these companies will get 1000x return. A perfect example is the Canva price hike from yesterday or any and every Google product from here on out. It's essentially being forced upon everyone that uses internet technology and someone is going to win while everyone else loses (consumers and small businesses).

If / when AGI happens can we make sure it’s not the Matrix?

Imagine organizing the world's data and knowledge, and integrating it seamlessly into every possible workflow.

Now you're getting close.

But also remember, this company is not trying to produce AGI (intelligence comparable to the flexibility of human cognition), it's trying to produce super intelligence (intelligence beyond human cognition). Imagine what that could do for your job, career, dreams, aspirations, moon shots.

I don't think you remember the dot-com era. Loads of people thought Amazon and Pets.com were hilarious ideas. Cliff Stoll wrote a whole book on how the Internet was going to do nothing useful and we were all going to buy stuff (yes, the books too) at bricks-and-mortar, which was rapturously received and got him into _Newsweek_ (back when everyone read that).

"We’re promised instant catalog shopping — just point and click for great deals. We’ll order airline tickets over the network, make restaurant reservations and negotiate sales contracts. Stores will become obsolete. So how come my local mall does more business in an afternoon than the entire Internet handles in a month?"

By contrast, SSI doesn't have the technology. The question is whether they'll be able to invent it or not.

Not a VC, but I'd assume in this case the investors are not investing in a plausible biz plan, but in a group of top talent, especially given how early stage the company is at. The $5B valuation is really the valuation of the elite team in a arguably hyped market.

I don't understand this question. How could even average-human-level AGI not be useful in business, and profitable, a million different ways? (you know, just like humans except more so?). Let alone higher-human-level, let alone moderately-super-human level, let alone exponential level if you are among the first? (And see Charles Stross, Accelerando, 2005 for how being first is not the end of the story.)

I can see one way for "not profitable" for most applications - if computing for AGI becomes too expensive, that is, AGI-level is too compute intensive. But even then that only eliminates some applications, and leaves all the many high-potential-profit ones. Starting with plain old finance, continuing with drug development, etc.

Open source LLMs exist. Just like lots of other open source projects - which have rarely prevented commercial projects from making money. And so far they are not even trying for AGI. If anything the open source LLM becomes one of the agent in the private AGI. But presumably 1 billion buys a lot of effort that the open source LLM can't afford.

A more interesting question is one of tradeoff. Is this the best way to invest 1 billion right now? From a returns point of view? But even this depends on how many billions you can round up and invest.

But yeah, VCs generally aren't about building profitable companies, because there's more profit to be made - and sooner - if you bootstrap and sell.

People should have realized it by now that Silicon Valley exist because of this.

I'm so tired of people who stubbornly insist that somehow only the same 500 people are possibly capable of having valuable thoughts in any way.

By selling to the "dumb(er) money" - if a Softbank / Time / Yahoo appears they can have it, if not you can always find willing buyers in an IPO.

Picture something 1,000 smarter than a human. The potential value is waaaay bigger than any present company or even government.

Probably won’t happen. But, that’s the reasoning.

Last night as my 8yo was listening to childrens audio books going to sleep, she asked me to have it alternate book A then B then A then B.

I thought, idunno maybe I can work out a way to do this. Maybe the app has playlists and maaaaaaaaaaybe has a way to set a playlist on repeat. Or maybe you just can't do this in the app at all. I just sat there and switched it until she fell asleep, it wasn't gonna be more than 2 or 3 anyway, and so it's kind of a dumb example.

But here's the point: Computers can process language now. I can totally imagine her telling my phone to do that and it being able to do so, even if she's the first person ever to want it to do that. I think the bet is that a very large percentage of the world's software is going to want to gain natural language superpowers. And that this is not a trivial undertaking that will be achieved by a few open source LLMs. It will be a lot of work for a lot of people to make this happen, as such a lot of money will be made along the way.

Specifically how will this unfold? Nobody knows, but I think they wanna be deep in the game when it does.

How good does it have to be, how many features does it have to have, how accurate does its need to be.. in order for people to pay anything? And how much are people actually willing to spend against the $XX Billion of investment?

Again it just seems like "sell to AAPL/GOOG/MSFT and let them figure it out".

Voice assistants do a small subset of the things you can already do easily on your phone. Competing with things you can already do easily on your phone is very hard; touch interfaces are extremely accessible, in many ways more accessible than voice. Current voice assistants only being able to do a small subset of that makes them not really very valuable.

And we aren't updating and rewriting all the world's software to expose its functionality to voice assistants because the voice assistant needs to be programmed to do each of those things. Each possible interaction must be planned and implemented invidually.

I think the bet is that we WILL be doing substantially that, updating and rewriting all the software, now that we can make them do things that are NOT easy to do with a phone or with a computer. And we can do so without designing every individual interaction; we can expose the building blocks and common interactions and LLMs may be able to map much more specific user desires onto those.

or

"Imagine an AI that can just use the regular human-optimized UI!"

These are things VCs will say in order to pump the current gen AI. Note that current gen AI kinda suck at those things.

Feels very different to me. The dominant ones are run by Google, Apple, and Amazon, and the voice assistants are mostly add-on features that don't by themselves generate much (if any) revenue (well, aside from the news that Amazon wants to start charging for a more advanced Alexa). The business model there is more like "we need this to drive people to our other products where they will spend money; if we don't others will do it for their products and we'll fall behind".

Sure, these companies are also working on AI, but there are also a bunch of others (OpenAI, Anthropic, SSI, xAI, etc.) that are banking on AI as their actual flagship product that people and businesses will pay them to use.

Meanwhile we have "indie" voice assistants like Mycroft that fail to find a sustainable business model and/or fail to gain traction and end up shutting down, at least as a business.

I'm not sure where this is going, though. Sure, some of these AI companies will get snapped up by bigger corps. I really hope, though, that there's room for sustainable, independent businesses. I don't want Google or Apple or Amazon or Microsoft to "own" AI.

And again this against CapEx of something like $200B means $100/year per user is practically rounding to 0.

Not to mention the OpEx to actually run the inference/services on top ongoing.

I wouldn't trust any kind of AI bot regardless of intelligence or usefulness to buy toilet paper blindly, yet alone something like a hard drive or whatever.

One could ask: how is this different from automatic call centers? (eg “for checking accounts, push 1…”) well, people hate those things. If one could create an automated call center that people didn’t hate, it might replace a lot of people.

The global call center market is apparently $165B/year revenue, and let's be honest even the human call center agents aren't great. So market is big and bar is low!

However, we are clearly still quite far from LLMs being a) able to know what they don't know / not hallucinate b) able to quickly/cheaply/poorly be trained the way you could a human agent c) actually be as concise and helpful as an average human.

Also it is obviously already being tried, given the frequent Twitter posts with screenshots of people jailbreaking their car dealership chat bot to give coding tips, etc.

Impressed by this bot recently shared on news.yc [0]: https://storytelling-chatbot.fly.dev/

> Specifically how will this unfold? Nobody knows

Think speech will be a big part of this. Young ones (<5yo) I know almost exclusively prefer voice controls where available. Some have already picked up a few prompting tricks ("step by step" is emerging as the go-to) on their own.

[0] https://news.ycombinator.com/item?id=40345696

An ubiquitous phone has enough sensors/resources to be fully situationally aware and preempt/predict for each holder any action long time ahead.

It can measure the pulse, body postures and movements, gestures, breath patterns, calculate mood, listen to the surrounding sounds, recall all information ever discussed, have 360 deg visual information (via a swarm of fully autonomous flying micro-drones), be in an network with all relevant parties (family members, friends, coworkers, community) and know everything they (the peers) know.

From all gathered information the electronic personal assistant can predict all your next steps with high confidence. The humans think that they are unique, special and unpredictable, but opposite is the case. An assistant can know more about you than you think you know about yourself.

So your 8yo daughter does not need to tell how to alternate the audio books, the computer can feel the mood and just do what is appropriate, without her need to issue a verbal command.

Also in the morning you do not need to ask her how she slept tonight and listen to her subjective judgement.

The personal assistant will feel that you are probably interested in your daughters sleep and give you an exact objective medical analysis of the quality of the sleep of your daughter tonight, without you needing to ask the personal assistant of your daughter.

I love it, it is a bottomless goldmine for data analysis!

Next step: the assistant knows that your brain didn't react to much to its sleep report the last 5 mornings, so it will stop bothering you altogether. And maybe chitchat with your daughter's assistant to let her know that her father has no interest in her health. Cool, no? (I bet there is already some science fiction on this topic?)

Why do you think need to make money ? VC are not PEs for a reason. a VC have to find high risk/ high reward opportunities for their LPs they don't need to make financial sense, that is what LPs use Private Equity for.

Think of it as no different than say sports betting , you would like to win sure, but you don't particularly expect to do so, or miss that money all that much for us it $10 for the LP behind the VC it is $1B.

There is always few billions every year that chases the outlandish fad, because in the early part of the idea lifecycle it not possible to easily differentiate what is actually good and what is garbage.

Couple of years before it was all crypto, is this $1B any worse than say roughly same amount Sequoia put in FTX or all the countless crypto startups that got VC money ? Few before that it was kind of all Softbank from WeWork to dozen other high profile investments.

The fad and fomo driven part of the secto garners the maximum news and attention, but it is not the only VC money. Real startups with real businesses get funded as well with say medium risk/medium rewrard by VCs everyday but the news is not glamorous to be covered like this one.

NVDA::AI

CSCO::.COM

I remember seeing interviews with Nortel's CEO where he bragged that most internet backbone traffic was handled by Nortel hardware. Things didn't quite work out how he thought they were going to work out.

I think Nvidia is better positioned than Cisco or Nortel were during the dotcom crash, but does anyone actually think Nvidia's current performance is sustainable? It doesn't seem realistic to believe that.

There is no specific reason to assume that AI will be similar to the dotcom boom/bust. AI may just as easily be like the introduction of the steam engine at the start of the industrial revolution, just sped up.

[0] "Planning and construction of railroads in the United States progressed rapidly and haphazardly, without direction or supervision from the states that granted charters to construct them. Before 1840 most surveys were made for short passenger lines which proved to be financially unprofitable. Because steam-powered railroads had stiff competition from canal companies, many partially completed lines were abandoned."

-- https://www.loc.gov/collections/railroad-maps-1828-to-1900/a...

That was similar to what happened during the dotcom bubble.

The difference this time, is that most of the funding comes from companies with huge profit margins. As long as the leadership in Alphabet, Meta, Microsoft and Amazon (not to mention Elon) believes that AI is coming soon, there will be funding.

Obviously, most startups will fail. But even if 19 fail and 1 succeed, if you invest in all, you're likely to make money.

GPUs, servers, datacenters, fabs, power generation/transmission, copper/steel/concrete..

All to train models in an arms race because someone somewhere is going to figure out how to monetize and no one wants to miss the boat.

My outsider observation is that we have a decent number of players roughly tied at trying to produce a better model. OpenAI, Anthropic, Mistral, Stability AI, Google, Meta, xAI, A12, Amazon, IBM, Nvidia, Alibaba, Databricks, some universities, a few internal proprietary models (Bloomberg, etc) .. and a bunch of smaller/lesser players I am forgetting.

To me, the actual challenge seems to be figuring out monetizing.

Not sure the 15th, 20th, 30th LLM model from lesser capitalized players is going to be as impactful.

1. The black swan: if AGI is achievable imminently, the first company to build it could have a very strong first mover advantage due to the runaway effect of AI that is able to self-improve. If SSI achieves intelligence greater than human-level, it will be faster (and most likely dramatically cheaper) for SSI to self-improve than anyone external can achieve, including open-source. Even if open-source catches up to where SSI started, SSI will have dramatically improved beyond that, and will continue to dramatically improve even faster due to it being more intelligent.

2: The team. Basically, Ilya Sutskever was one of the main initial brains behind OpenAI from a research perspective, and in general has contributed immensely to AI research. Betting on him is pretty easy.

I'm not surprised Ilya managed to raise a billion dollars for this. Yes, I think it will most likely fail: the focus on safety will probably slow it down relative to open source, and this is a crowded space as it is. If open source gets to AGI first, or if it drains the market of funding for research labs (at least, research labs disconnected from bigtech companies) by commoditizing inference — and thus gets to AGI first by dint of starving its competitors of oxygen — the runaway effects will favor open-source, not SSI. Or if AGI simply isn't achievable in our lifetimes, SSI will die by failing to produce anything marketable.

But VC isn't about betting on likely outcomes, because no black swans are likely. It's about black swan farming, which means trying to figure out which things could be black swans, and betting on strong teams working on those.

If you could get expert-human-level capability with, say, 64xH100s for inference on a single model (for comparison, llama-3.1-405b can be run on 8xH100s with minimal quality degradation at FP8), even at a mere 5 tok/s you'd be able to spin up new research and engineering teams for <$2MM that can perform useful work 24/7, unlike human teams. You are limited only by your capital — and if you achieve AGI, raising capital will be easy. By the time anyone catches up to your AGI starting point, you're even further ahead because you've had a smarter, cheaper workforce that's been iteratively increasing its own intelligence the entire time: you win.

That being said, it might not be achievable! SSI only wins if:

1. It's achievable, and

2. They get there first.

(Well, and the theoretical cap on intelligence has to be significantly higher than human intelligence — if you can get a little past Einstein, but no further, the iterative self-improvement will quickly stop working, open-source will get there too, and it'll eat your profit margins. But I suspect the cap on intelligence is pretty high.)

OpenAI priced its flagship chatbot ChatGPT on the low end for early product adoption. Let's see what jobs get replaced this year :)

Actually converting it to cash? That doesn't happen anymore. Everyone just focuses on IRR and starts the campaign for Fund II.

Just as clean water is worth trillions to no one in particular. Or the air we breathe.

You can take an abstract, general category and use it to infer that some specific business will benefit greatly, but in practice, the greater the opportunity the less likely it is it will be monopolized, and the more likely it is it will be commoditized.

But, my comment was a reference to the magic thinking that goes into making predictions.

$5B pre-product, betting on the team is fine. $50B needs to be a lot more than that.

Many examples of industries collapsing under the weight of belief. See: crypto.

OpenAI and Anthropic for sure have products and that's great. However, these products are pretty far from a super intelligence.

The bet SSI is making is that by not focusing on products and focusing directly on building a super intelligence, they can leapfrog all these other firms

Now, if you assign any reasonable non zero probability to them succeeding,the expected value of this investment is infinity. It's definitely a very risky investment, but risk and expected value are two different things.

Conservative math: 3B connected people x $0.50/day “value” x 364 days = $546B/yr. You can get 5% a year risk free, so let’s double it for the risk we’re taking. This yields $5T value. Is a $1B investment on someone who is a thought leader in this market an unreasonable bet?

There's also the issue of who gets the benefit of making people more efficient. A lot of that will be in the area of more efficient work, which means corporations get more work done with the same amount of employees at the same level of salary as before. It's a tough argument to make that you deserve a raise because AI is doing more work for you.

So beyond, that you can easily can transform a newbie into a junior IT, or JR into a something ala SSR, and getting the SR go wild with times - hours - to get a solution to some stuff that previously took days to be solved.

After the salaries went down, that happened about 2022 to the beginning of 2023, the layoffs began. That was mostly masked "AI based" corporate moves, but probably some layoff actually had something to do with extra capabilities in improved AI tools.

That is, because, fewer job offers have been published since maybe mid-2023, again, that could just be corporate moves, related to maybe inflation, US markets, you name it. But there's also a chance that some of those fewer job offer in IT were (and are), the outcome of better AI tools, and the corporations are betting actively in reducing headcounts and preserving the current productivity.

The whole thing is changing by the day as some tools prove themselves, other fail to reach the market expectations, etc.

To be able to achieve that is entirely dependent on two things:

1) deploying capital in the current fund on 'sexy' ideas so they can tell LPs they are doing their job

2) paper markups, which they will get, since Ilya will most definitely be able to raise another round or two at a higher valuation. even if it eventually goes bust or gets sold at cost.

With 1) and 2), they can go back to their existing fund LPs and raise more money for their next fund and milk more fees. Getting exits and carry is just the cherry on top for these megafund VCs.

The trick is being in the game long enough

I mean it probably depends on the LP and what is their vision. Not all apples are red, come in many varieties and some for cider others for pies. Am I wrong?

but.. it really depends heavily on the LP base of the firm, and what the firm raised it's fund on, it's incredibly difficult to generalize. The funds I'm involved around as an LP... in my opinion they can get as "sexy" as they like because I buy their thesis, then it's just: get the capital deployed!!!!

Most of this is all a standard deviation game, not much more than that.

https://www.otpp.com/en-ca/investments/our-advantage/our-per... https://www.hellokoru.com/

an LP is a "limited partner." they're the suckers (or institutional investors, endowments, pensions, rich folks, etc.) that give their cash to venture capital (VC) firms to manage. LPs invest in VC funds but don't have control over how the money gets used—hence *limited* partner. they just hope the VCs aren't burning it on overpriced kombucha and shitty "web3" startups.

meanwhile, the VCs rake in their fat management fees (like the 2% mentioned) and also get a cut of any profits (carry). VCs are more concerned with looking busy and keeping those sweet fees rolling in than actually giving a fuck about long-term exits.

Someone wants to fund my snide, cynical AI HN comment explainer startup? We are too cool for long term plans, but we use AI.

You will struggle to raise funds if the companies you bet on perform poorly; the worse your track record the less chances of raising money and earn income from it.

So all the successful VC partners from 2010 are close to retirement or have retired?

Why say something testable if it is obviously wrong.

Eventually, on a long enough timeline, all these tech companies with valuations greater than 10 billion eventually make money because they have saturated the market long enough to become unavoidable.

I would say the investors want to look cool so invest in AI projects. And AI people look cool when they predict some improbable hellscape to hype up a product that all we can see so far can regurgitate (stolen) human work it has seen before in a useful way. I’ve never seen it invent anything yet and I’m willing to bet that search space is too dramatically large to build algorithms that can do it.

Look at previous such investments Microsoft and AWS have done in OpenAI and Anthropic.

They need use cases and customers for their initial investment for 750 billion dollars. Investing in the best people in the field is then of course a given.

So...

OpenAI's business model may or may not represent a long term business model. ATT, it just the simplest commercial model, and it happened to work for them given all the excitement and a $20 price point that takes advantage of that.

The current "market for ai" is a sprout. It's form doesn't tell you much about the form of the eventual plant.

I don't think the most ambitious VC investments are thought of in concrete market share terms. They are just assuming/betting that an extremely large "AI market" will exist in the future, and are trying to invest in companies that will be in position to dominate that market.

For all they know, their bets could pay off by dominating therapy, entertainment, personal assistance or managing some esoteric aspect of bureaucracy. It's all quite ethereal, at this point.

It's potentially way bigger than that. AI doesn't have to be the product itself.

Fundamentally, when we have full AGI/ASI and also the ability to produce robots with human level dexterity and mobility, one would have control over an endless pool of workers (worker replacements) with any skillset you require.

If you rent that "workforce" out, the customer would rake in most of the profit.

But if you use that workforce to replace all/most of the employees in the companies you control directly, most of the profit would go to you.

This may even go beyond economic profit. At some point, it could translate to physical power. If you have a fleet 50 million robots that has the capability to do anything from carpentry to operating as riot police, you may even have the ability to take physical control of a country or territory by force.

And:

power >= money