Hacker News Reader

Comment by gspencley | Hacker News Reader

Personally I predict that generative AI is going to be the next Metaverse and crypto.

It's not about whether or not there is value there to be provided. VR and cyrpto provide value too, it's just that the markets for that tech are way more "niche" than companies were hoping for.

And I don't think VR is going away. If it wasn't dead in the water in the 90s, then it's not going to die because the Metaverse failed. But its applicability is currently limited to a few niche applications like video games and possibly CAD type software.

The thing that generative AI has going for it is that it is extremely broad in so far as its applicability. In other words, good luck selling my next door neighbour on generative AI. What is it? What is it good for? What problems does it solve in the here and now?

Generative AI is like electricity, only I don't think quite as useful. The point is that, on its own, it has no value what-so-ever. The value lies in what it is used for.

Right now what I am observing is a phenomenon that is way too common in our industry. Companies are rushing to go to market with some AI "thing." 99.9999% of those "things" will end up being more versions of the Metaverse and crypto. Because most of these companies aren't really sure what problem they are solving for actual people. It's all so novel and abstract and people are seeing dollar signs with very little understanding of how they are going to put it to use.

Where it finds its uses, I predict will be limited to the same applications that we're already using ML for. Chat bots, entertainment, generating document outlines etc. Just because we made a big breakthrough with the technology doesn't mean that we've actually created a solution to real world problems that real world people are having here and now in the real world. No one, other than maybe Open AI, is going to get rich on "generative AI." If they get rich and change the industry, it will be finding a use for it that many people want.

View on HN

I think crypto, metaverse, and generative AI are all really different -- and I agree with the original poster. Generative AI seems much more likely to be a game changer.

Crypto was something I always struggled to make sense of. My son asked, "why didn't I buy bitcoin when it was a few cents?!" -- The simple answer was it didn't seem like it had much use. The value in crypto eventually came in the form of a pyramid scheme. Admittedly, I do use crypto to do transactions with untrustworthy people (don't have to worry about chargebacks and things like that). But it's hype was around it being a speculative commodity. For normal people -- it's just not all that useful.

The metaverse is different in that I do see the theoretical value in it. But it seems a LONG way off. I haven't seen anything that really makes me think, "wow -- this is it". It doesn't seem like we're close. And honestly I think generative AI maybe a prerequisite to deliver a truly valuable metaverse.

Generative AI is useful now and represents a huge leap over the state of the art, not that long ago. I worked with RNNs and LSTMs to answer questions not that many years ago, and honestly I wouldn't have predicted this level of effectiveness for LLMs for probably 20 plus years -- and I certainly didn't know how we'd do it. But more importantly I'm finding it valuable today. As I noted in another post -- ChatGPT-4 already has shown better differential diagnosis skills than my family's doctor. I use it literally every day already. There are things that I want it to do that would make my life 10x better -- and I could see a clear path to get there (like, if that was my day job I think I could put together a project plan to deliver it).

I just don't think these techs are comparable at all.

People seem to see new developments as black boxes that you can't analyze. Like they see something new and ask themselves "will it work?" and then say it will work or not based on whether the last thing worked... with zero nuance.

Let's try to analyze crypto on its own as an example: yes, crypto was a new frontier in money supply, but what makes it critically different? And are these critical differences actually desired?

Crypto was critically different in that it would decentralize the financial system.

Was it desired?

Centralized institutions are unfair and can be in-equal, but they are actually really great. When the financial becomes unstable, these institutions intervene and settle things down (see the Great Depression). When someone steals from my credit card, I just call up my bank and they fix it. How would crypto handle these situations?

No one could figure it out. So if crypto was trying to sell something that no one really wanted, how the heck could it work? ...then it crashed... because there was no centralized control. It's not like we haven't had this happen in history already.

> When someone steals from my credit card, I just call up my bank and they fix it. How would crypto handle these situations?

So far the crypto orgs have called a centralized institution called law enforcement...

That doesn't return your money to you like the credit card though. The irreversibility 'feature' of most cryptocurrencies makes it really hard to recover funds stolen through fraud... I guess that is a desirable feature for anybody performing fraud though.

Of the things that make cryptocurrency something I want to avoid, the irreversibility part is the biggest problem.

Nothing is ever irreversible, it's just a question of how expensive it is to make the database writes. Ethereum forked the whole currency to undo a write (the DAO hack) and if the legal system can find a hacker they can make them reverse their transaction.

…But having someone with admin access over the network and can fix mistakes is just a lot cheaper than refusing to build that in for ideological reasons.

Most (but not all) things in life are technically reversible. But some things are so difficult or expensive to reverse that they are effectively irreversible nonetheless.

I'm guessing you avoid cash for the same reason?

if you can locate the physical person who stole your crypto, you can always use the rubber hose method of returning your money. I bet it works.

VR doesn't need to replace social media for it to be useful.

AI doesn't need to replace doctors for it to be useful.

Crypto doesn't need to replace your credit card for it to be useful.

If none of these things are useful to you, that's fine, other people find them useful and that's okay.

Think "and" instead of "or".

I agree. VR will have killer apps in particular domains, but that doesn’t mean it will be used by everyone all of the time. For example, I work in ocean mapping where we have large amounts of point cloud data that need occasional manual editing/QAQC in an environment that matches the dimensionality of the data. My colleagues are working on some very interesting VR point cloud editing tools [1].

[1] https://ccom.unh.edu/vislab/projects/point_clouds/

I agree!

But that's not how any of those things are being sold.

I think crypto is great for anonymous digital markets. It would be great to disrupt the financial industry, but that's a really big thing. For the markets, I think NFTs is a really good use case. That's a way for artists to have digital galleries, receiving international payments, work can get traded, all independent of the marketplace itself. No company behind that. Books, audio content, ..., and so on are also easily the target of a next generation. I don't believe in this tokenization of physical goods, because then you have a company behind creating the link between token and good. But for purely virtual goods, I think it is great.

In a Blockchain world, how was there anonymity with crypto?

I’m guessing the answer is that not all crypto is on blockchain, but I actually have no clue.

Monera supposedly is untraceable somehow. And you can do things with passing private keys and mixers even on bitcoin.

You have to analyze the entire system including what they use the untraceable currency for.

Someone's sent you a billion dollars in Monero the IRS doesn't know you have - okay. But now when you use it to buy a mansion, won't that be obvious?

Thats basic money laundering and HSBC is ready to help you.

The desire to have untraceable money isn't always because you're trying to do something illegal or immoral.

Privacy is important to people, and that even extends to privacy from your government. You never know when your government might change from a responsible, fair entity, to a oppressive regime.

If you do buy a mansion with unreported billions and you're buying from a reputable buyer, then you'd expect the government to know about that sale, but if you're stock piling assets in case you need to flee from your government to survive, having anonymous money that's easy to transport is valuable.

Has nothing to do with my question but I agree with everything you said.

> You never know when your government might change from a responsible, fair entity, to a oppressive regime.

Truer words were never spoken. It is hard to get this across to people with the completely understandable "I didn't do anything wrong so nothing can happen to me" attitude.

As Beria said, "Show me the man and I'll find you the crime"...

Yeah, it's weird that people can perceive anonymous transactions to be only valuable to ne-er-do-wells at the same time that large swaths of the US are criminalizing access to abortifacients.

> It would be great to disrupt the financial industry

Only if there was something at least as good to replace it. Cryptocurrency isn't that.

Disruption just for disruption's sake is a Bad Thing.

> NFTs is a really good use case.

Honestly, I still don't see how. None of the explanations make any real sense to me.

NTFs seem silly. You own a digital asset on a block chain. But anyone could use a jpeg, it just has no real value to have so-called ownership.

Small correction: NFTs that point to images is one example of how to use the NFT standard, not what an NFT is.

The NFT discussion always ends up at "its just a link to a jpeg", when there are NFTs that have nothing to do with images.

Can you give me some examples of these other kinds of NFTs that are useful then?

California's DMV investigating them for licenses/registrations is an example that hit HN's front page earlier this week.

This is interesting because I haven't heard of this. Would it be something like your license is controlled by you? Somehow you have to keep it private though. Having someone control a position on the blockchain wouldn't seem to be that different than name and password. You can use that to prove who you are, and you can lose that just like you can lose control of your crypto wallet if you forget your password or someone guesses it or hacks your machine.

One workaround would be to store the license/registration tokens in multisig wallets, requiring both the licensee and the DMV to approve of transfers (or add a third signer and go with two of three). A fair bit more centralized than what's ideal, but the very concept of a state-issued license or registration is already implicitly centralized anyway.

> > NFTs is a really good use case.

> Honestly, I still don't see how. None of the explanations make any real sense to me.

I don't really think NFTs have a real use case, but i can see why it might. It is effectively a digital certificate of authenticity that cannot be faked. But like a physical certificate of authenticity, it is worth as much as the general market would place trust in it. And it isn't much, since collectors of digital goods currently do not value authenticity (realistically) - they value the actual digital good, and thus, a copy is as good as the "real thing".

Therefore, collectors can only really exist for physical goods, and thus NFTs don't actually provide much value, beyond the ability for someone to speculate with it.

What’s the point of disruption without a solution?

Like let’s say my sink keeps leaking so you suggest smashing it. Great, you fixed the leak, but I have no sink now.

Now if you came along and said you’d smash my sink and then replace it with a better sink, I’ll allow it.

People are not hung up on the disruption part. People are hung up on having no sink.

Crypto would be extremely useful if it were wholeheartedly adopted by the government of a medium-sized country. The fundamentals of cryptocurrency are as good as the gold standard, maybe slightly better, and could be better than fiat; the problem is that no one pays their taxes in it so no one uses it as a real currency. Likewise, NFT’s could be used to implement property titles if a government ever agreed to enforce them. Then things like smart contracts and whatnot start to look a lot more interesting. The problem is that the entire crypto economy is a fake play-money ecosystem with no power over physical reality.

This went from +2 to -1 overnight, and I'm not complaining; I'm just curious if I got downvoted by the crypto haters or by the crypto supporters. Such is the risk of expressing a nuanced opinion on Hacker News I suppose.

The thing you're missing is there are no incentives for a sovereign nation to give up the ability to mint their own currency and control its supply (monetary policy).

There’s a reason I said “medium-sized country”. The US, EU, and apparently now China wouldn’t be interested because they are world reserve currencies, or aspire to be in China’s case. On the other hand, you could probably make a case to voters for cryptocurrency to replace something utterly dysfunctional like the Argentine peso. Crypto doesn’t really do much for a country with functioning and trustworthy governing institutions, but not every country has that.

> The fundamentals of cryptocurrency are as good as the gold standard, maybe slightly better, and could be better than fiat

The fundamentals of cryptocurrency are just as fiat as any other currency, including the gold standard.

> Centralized institutions are unfair and can be in-equal, but they are actually really great. When the financial becomes unstable, these institutions intervene and settle things down (see the Great Depression).

Interestingly, there is a well respected school of thought in economics that pins the blame for the boom/bust cycle on artificially manipulating interest rates via that exact centralization.

> When someone steals from my credit card, I just call up my bank and they fix it.

I'm not really sure how that is an example of "centralization." A private lender with whom you've entered into a loan agreement does not control the supply of the currency. They could just as easily be lending you gold or bread. I'm no cyrpto fanboy ... couldn't really care less about it ... but in theory there is no reason that a bank couldn't lend you Bitcoins or something. If there is a problem with the currency that's going to bite you regardless of what the currency is.

I think the strongest case for crypto was using it as a hedge against inflation. Inflation being an increase in the supply of fiat currency. Something that, whether you subscribe to the "Mises school" of economics or not, is irrefutably caused by the central bank creating new currency out of thin air, thus diluting the purchasing power of the existing money in circulation. I propose that that is "irrefutable" because, in order to debunk it, you would have to demonstrate that the law of supply and demand is false and/or that currency is not a commodity subject to the same "rules" as every other.

I don’t disagree that the boom/bust cycle could be created by centralization, but I also don’t care I suppose. I mean, there’s no perfect solution out there so a boom/bust is a reasonably mild price.

A large reason why crypto, at least in the current implementations, can provide decentralization is by removing discretion from any authority and making transactions permanent. That’s in contrast to our current money system, because it’s merely numbers that ultimately can be manipulated by central authorities and accessible by law enforcement, that ultimately means that transactions are not truly permanent and that allows us to repair mistakes after the fact, a feature not really afforded by crypto. It’s a very helpful feature.

I also consider a consistent amount of inflation an important feature of any successful economy. Without inflation, I would be incentivized to hold my money whenever I expect deflation, which is terrible because it’s basically me hoarding value from society.

Runaway inflation is absolutely disastrous and I don’t see how crypto could have been considered a hedge against it if its entire value is backed purely by speculative demand. Contrast that to the USD or other currency where that is backed by the full faith and credit of a government. Obviously that is meaningless of we’re talking about, say, Argentina but the US has successfully been the same, largely dependable, and boring government for 245 years and through earning a reputation with time that the USD can be treated as a reserve currency.

I don't think crypto, as most chains implement it, is at all vulnerable to runaway inflation.

A functional crypto can go downhill for a lot of reasons. Technical failures. Association with scams.

But I am not sure how a tightly controlled supply over time, or asymptotically fixed supply, coin can have runaway inflation - if everyone starts demanding more of the coin for goods and services, there just isn't more of the coin for that.

Generative AI basically unlocks the metaverse, since all the 3D environment can now be generated. The metaverse changes people's context and senses with generated data. Interestingly AI will paint a new world for humans to learn and experience new situations more often, more easily on a broader spectrum. Exciting !

> Interestingly AI will paint a new world for humans to learn and experience new situations more often, more easily on a broader spectrum.

This is ignoring the instinct built over generations that humans don't want to "experience a new world", they want to explore it with other humans.

We are social animals. Facebook is so successful because it started out as a way to connect more with people you already knew or were acquainted with.

The "metaverse" and all its AI-generated worlds add little to no value in that direction. Wake me up when the metaverse makes it trivial to host a garden party and cook for the 5-10 attendees.

> This is ignoring the instinct built over generations that humans don't want to "experience a new world", they want to explore it with other humans.

I think humans want to explore it with intelligent beings, not necessarily “humans”. These new technologies have the potential to create agents in the movie “Her” sense that are compelling to interact with. The metaverse applicability seems clear.

Here's another scenario I envision now:

You and your friends jump in the metaverse every Tuesday night to play some virtual game, e.g., go hunting or rob a bank. Timmothy is out on vacation this week so won't be joining -- but now you can spawn up virtual Timothy, which is trained on a bunch of Timothy's data (see the other HN post about the guy who trained his bot on his friends group chats). Now Timothy is with his friends in this virtual world and he behaves mostly how Timothy does.

Virtual timothy plays better and without complaining. Weeks later the gang dreams of virtual tim. A week later they start playing, but this time without Tim.

Ok this scenario, but my wife.

> now you can spawn up virtual Timothy, which is trained on a bunch of Timothy's data

Not just no, but hell no.

If I'm going to play with a robot, I absolutely don't want it to replicate a friend. Even more, I would be furious if any of my friends replicated me like that.

When I'm playing with my friends, I'm doing it to play with my friends, not simulacra of my friends.

It won’t start with Timmothy. It’ll start with John Cena or Donald Trump. You’ll get to play with virtual celebrities. Then domain experts, like tennis camp with Serena Williams. Or story telling with Steven King. Or Fortnite with Ninja. And it will slowly find its way down the popularity slide to friends.

And you will protest it. Just like my MIL hates Facebook and thinks calling is so much more efficient than texting. But the younger generation will find this as natural as TikTok.

https://en.wikipedia.org/wiki/Play_Magnus_(mobile_app)

I wouldn't protest any of those things (assuming the celebrities consent to it). You're right that such a thing holds zero interest to me, but I don't find the idea objectionable. Celebrities are essentially "virtual people" to begin with (that is, celebrities are playing a role in public appearance -- you aren't seeing their true selves).

But I find the idea of replacing actual friends with simulations to be inherently objectionable.

I think that's what people do on a normal day, but all of the technical formats and decisions (vhs/dvd/streaming video/etc) have been pioneered by porn.

So, pretty much what you said, but probably having sex with anime girls instead.

> I think humans want to explore it with intelligent beings, not necessarily “humans”.

I think the majority humans want to be with humans, not just intelligent beings. Some don't, of course, and there's probably a decent business catering to them.

Honestly I don’t think content generation is a limiting factor for a compelling meta verse. IMO the constraints are more to do with fidelity and the quality of the experience.

There are currently so many weaknesses and limitations with the medium of VR that it’s just not particularly appealing for folks to spend a lot of time using it, and I think that’s why you don’t have great content gen.

I do think generative AI has some really interesting possibilities for interactive media - but I don't see how it unlocks anything.

It's clear there's a market for well-designed and compelling VR chat. VRChat[1] has a large community and had a very active developer community[2]. Facebook had no luck in attracting users in a proven market - there's no reason to believe their products would be more popular after hooking a generative agent up to it.

If you are looking for "the metaverse" in a generic sense - other people are already doing it better than Facebook (and have been for some time). It's been unlocked for years. Facebook hasn't even managed a capable clone, much less broken new ground.

[1] https://hello.vrchat.com/

[2] Until they banned all modding - here's a video of a user using sign language and subtitles: https://twitter.com/TheFoxipso/status/1551640725455982593

> It's clear there's a market for well-designed and compelling VR chat. VRChat[1] has a large community and had a very active developer community[2]. Facebook had no luck in attracting users in a proven market - there's no reason to believe their products would be more popular after hooking a generative agent up to it.

Facebook gets to take 30% of VR Chat's gross revenue (they sell premium features) on standalone and the desktop Oculus store.

Apple likely makes more profit off Roblox than Roblox.

we had SecondLife, ut dies everythibg metaVerse does, 20 years ago.

You've pretty much nailed what I've been thinking about when comparing generative AI hype to crypto/metaverse hype.

I keep hearing this argument that generative AI has more utility, making it more "real". Sure, but they're missing the point, which is that people made some pretty ridiculous claims about the future utility of crypto / metaverse ($100k bitcoin, metaverse real-estate speculation) rather than the present day possibilities, and they're continuing to make ridiculous claims about the future utility of AI (the end of web development, AGI soon, etc).

The hype isn't about present day possibilities, but about future speculation, and the future speculation can get off-the-rails crazy.

The hype exists, firstly, because the present day possibilities are already impressive enough to net actual financial gains from improved productivity, and secondly because it's obvious that all this stuff hasn't yet been scaled to the limit - but the scaling and improvement that we've seen so far (GPT 3 -> 3.5 -> 4) showed massive gains on every step in a very short period of time.

Incidentally I was about to say that even speculation based on present utility can get a little bit out there. Modern AI is already quite good at generating code snippets, and it already increases developer productivity, but just how much is that going to really change things? Just how much more productive is an AI-augmented software developer? Are LLMs really the thing that was missing from no-code website / app builders?

I wouldn't say it's obvious that this stuff hasn't been scaled to the limit. I mean, sure, you could make bigger LLMs, but I don't think it's a given that LLM performance will increase in performance in a predictable fashion. If anything I think that performance probably resembles an S curve where performance matches improvements in scale before sharply plateauing. When's the plateau?

GPT-4 is a clear improvement over GPT-3, but it's also 6x larger, and so far I haven't quite seen any real evidence to suggest that it's 6x better, other than the fact that it scored better on tests.

It's hard to say what "6x better" even means, but there are tasks that GPT-3.5 consistently fails hard at that GPT-4 is much better at (and has improved over time).

I think the next big leap will be the context size. Going from 4k to 32k tokens means that the model can see a lot more when generating the answer. For code especially this makes a big difference in the ability to reason about it.

As to the limits, it's hard to predict because what we're seeing is more of a series of sudden leaps. But let me put it this way - so far we've seen noticeable improvements on every iteration of the new LLMs. So given this, why should the default assumption be that the limit past which nothing new will emerge, whatever it is, is close to the present size of the publicly available models?

Is anyone doing static code analysis with GPT? It seems like an excellent idea. I think a combination of formal analysis methods together with blackbox GPT magic will be a killer combination. That said, the market is tiny.

There is zero evidence that GPT-4 is a trillion parameter model and some evidence that it is the same parameter size as GPT-3.

Newer research and models out of DeepMind etc. show that GPT-4 performance is attainable at 90% the size of GPT-3. It's well known by now that GPT-3 was hugely over parameterized.

I think it’s extremely dangerous to predict the pattern of gains to continue.

Also I’m not actually convinced that there’s much value provided currently. I’ve yet to see a particularly compelling use case, and every time I go to try something that folks are hyping up (image generation, ChatGPT, copilot) I leave feeling like it’s all sizzle and no steak.

It's probably more dangerous to predict they won't, at this point.

The people that aren't seeing a compelling use case simply aren't looking.

> The people that aren't seeing a compelling use case simply aren't looking.

By all means, go do your startup, all the power to you.

But I personally fail to see it creating enough value for people to even bother use the AI, even if it's free and any problem could be ignored.

I can imagine people adopting it if it comes embedded on whatever software they already use. But that doesn't "completely change everything" or any of the other things people are repeating.

And yeah, I can't imagine it quickly improving so that happens either. I am personally bracing for a new AI winter, and believe that word is going to become more toxic than it has ever been.

I think generative AI has incremental utility. You can't half-use Bitcoin or the Metaverse; the idea is nonsensical. Bitcoin moves data around very slowly and at enormous cost, and the Metaverse provides vertigo-inducing VRChat for uncanny valley Mii avatars, but they're only usable if you go all-in.

Even if generative AI is not a gamechanging buzzword paradigm shift it's still useful for making existing tasks easier. Predicting the next sentence is autocomplete for word documents and phone keyboards, and we're already seeing it roll out. Artists who've gotten over the AI backlash can use it for backgrounds or textures, and designers can churn out those stylised low detail corporate-website graphics. Amateur authors on sites like RoyalRoad are already creating cover art and character portraits with AI generators. Millions of horny internet weirdos now have access to oddly specific pornography, and even if the AI gets the hands wrong they can retouch it in Krita.

Probably (>50%) we'll all be bored of it in 20 years and admit the hype was overblown, but I also think we'll see it integrated into day-to-day tasks in a way that Bitcoin and the Metaverse aren't.

That's kind of what I was betting on. I don't see it going away anytime soon, but I also think it's potential impact is being overblown.

With regards to AI image generators, I think people overestimate the importance of illustration skills to visual creatives. Namely, if you look at the top graphic designers, something you pick up on is that they aren't necessarily known for their top-tier illustration skills. Anyone can theoretically make a logo in the style of Paul Rand, yet curiously, hardly anyone ever does.

Even if the person in question is an illustrator, I think what fundamentally makes them valuable as an illustrator is that they know what to draw rather than how to draw.

It's sentiments like these that make me question the notion of certain careers getting disrupted by AI as opposed to simply improved by it.

I think you're both right. I think generative AI will be like the internet, and we'll experience a 90s-style bubble first, then it will gradually transform everything.

Personally though, I've already used it for multiple real-world applications in my business and personal life, whereas crypto was only ever a toy. VR I'm less sure about; so far it's only a toy, and I think it will take longer than AI to make a real difference, but I wouldn't be surprised if at least AR and probably VR are both eventually very commonplace.

FWIW, I've found uses for it already that will improve the company I'm building in the very short term. Maybe you disagree in the magnitude of the value, but I find it to be at least somewhat significant. Enough to pursue.

I've never worked for any company where leaning into the Metaverse or crypto would generate any meaningful business value, short or long term. The promises for Metaverse and crypto are always about a nebulous future that could exist if only everyone collectively bought into the vision.

Anything that is limited to the the company you're building at the moment sounds to me like the very definition of "niche application."

If you re-read my comment I didn't say that generative AI does not have any potential to solve any problems ever. What I said is that companies are currently racing to go to market with some "ai thing" and most are chasing it because it is the "next big thing" in their mind, not because they know what problems for their end users that it can actually help solve.

I know of companies right now who are banking on "selling AI to customers." That is doomed to fail in my opinion. Customers don't care about AI outside of the novelty and entertainment value it may possess. They have actual problems that need solving. My point was that generative AI, as the broad umbrella that it is, will see the hype train die and people will realize that "it" is not the revolutionary and disruptive Internet or printing press like invention that they think it will be. The value will be in individual applications of it, of which there will be a few but not as many I think people think there will be.

> that companies are currently racing to go to market with some "ai thing" and most are chasing it because it is the "next big thing" in their mind, not because they know what problems for their end users that it can actually help solve.

And if you re-read my comment, what I am saying is that while I'm not exactly "racing", I do have a very clear problem for my end users that this actually helps to solve. Its possible I could've cobbled together a similar solution with other technologies, but I didn't see a very clear path forward on this user-facing problem until very recent developments.

I suspect that the people in this same boat far outnumber people in the Metaverse or crypto equivalents.

The metaverse and crypto were entirely about some kind of future industry and society-changing application. "You don't see anything _now,_ but the future will be unrecognizable, trust me!"

Gen AI definitely has utility right now, and for some it's a lot of utility. But the hype around it is exactly the same, the hype is all around how "the future will be unrecognizable, trust me!"

So I agree that practically, Gen AI is utterly different from crypto/metaverse, it's a real thing with real value. But it's also attracting the exact same hucksters and they're kind of sucking all the air out of the room.

Gen AI, you got me here, for a second I thought it is the next one after Gen Z.

>Gen AI definitely has utility right now

People keep saying that, but I don't see it. So far we have chat bots, and Copilot. Copilot is going to be in legal limbo for years until the license issues are solved. Chatting with bing is fun and all but I don't see the long term utility.

What is the utility right now? Which of my problems are solved?

A few examples:

* My company has a very large codebase, and I am not familiar with 99% of it. I can use an AI assistant like Sourcegraph Cody to explain parts of the codebase to me at a high level, and suggest areas where I should dive in to address my specific problem.

* If I am working with a language, library, or framework I have not used before, I can ask ChatGPT to explain how a certain function works, and provide some code examples. If I write some code an get unexpected results, I can paste my code and results into ChatGPT, and ask it to tell me what went wrong.

* Someone sends me a spreadsheet that specifies some business logic, and I want to transform that spreadsheet into a YAML file, and write some code to parse the YAML config and take some action based on user-supplied data. ChatGPT is pretty good at this.

In all of these cases, I have to take the AI output with a grain of salt, and may have to do some supplementary research or debugging. But that's also the case when I ask coworkers for help. Right now, I would say generative AI provides a small boost to my productivity, but I can see that boost growing larger as language models improve.

It might not solve your problems, but there are a lot of people using CoPilot and/or GPT-4 for coding already, and it does save time for them.

>People keep saying that, but I don't see it.

I do. I care about my problems and what it does for me. I regularly employ ChatGPT to help me solve arbitrary tasks from writing code, debugging code, learning Korean, and cutting writing time in half.

I think the biggest societal impact may be not “utility” at all but rather on the social side of things. The chatbots are nerfed right now because the big companies don’t want them “acting weird” and doing things like the infamous jealous girlfriend convo between Kevin Roose and “Sydney”. But the idea behind the movie Her is essentially demonstrable now, and people are going to deliver those experiences.

I think things are going to get really weird. The prospect that my teenage son might have a virtual girlfriend does not seem sci-fi at all. Economically speaking, will people pay to chat and sext with “people”? I think yes?

> The thing that generative AI has going for it is that it is extremely broad in so far as its applicability. In other words, good luck selling my next door neighbour on generative AI. What is it? What is it good for? What problems does it solve in the here and now?

Disagree; I'd argue that the key distinction between generative AI and those other things (metaverse/crypto) is that the value of it is immediately apparent to almost anyone. There's a reason ChatGPT ended up becoming a household name with literally no marketing whatsoever.

Cool, so I'm Joe Sixpack next door. Things I care about: how to feed my kids, avoid getting laid off at the manufacturing plant I work for and my oldest daughter is learning to drive and has no idea how she's going to be able to afford a home of her own and, despite how much I love my kids, I really want her to move out before she's 30 so that my wife and I can enjoy retirement. Things I do for fun: play sports, video games and drink.

What problems in my life does generative AI immediately and obviously solve for me?

Like I said, it is "abstract" which is why you can make such a claim with such a high degree of confidence. Every person who sees POTENTIAL with generative AI has their own ideas as to what sorts of applications it has that they would find useful. The very words "generative AI" and "ChatGPT" create a Barnum Effect (https://en.wikipedia.org/wiki/Barnum_effect). Right now, I would posit that ChatGPT is like pet rocks. It got popular because it is neat and novel. Once that effect wears off, people will stop caring and the hype train will die as fast as crypto.

Therefore, until a company releases a product that uses ChatGPT under the hood and solves a real problem that a lot of people need solving right now, there's no tangible value that is inherent in the technology that can be identified. Only ideas and potentials.

I mean, off the bat you might use it to replace Google, like bing is already doing.

And it doesn't have to solve a problem for you to make an impact on your life- it can make your life worse too. If your daughter is interested in writing, graphic design,or programming, well, finding a job and moving out by 30 might become a lot harder. Bing is already in use. Copilot is already in use. These are not small markets or use cases.

Bing was already in use before ChatGPT and Copilot is seen by most as an evolution of auto-complete. My last paragraph specifically said that I predict that the applications for generative AI will be in areas where we're already using ML. Search and auto-complete are two examples.

> I mean, off the bat you might use it to replace Google, like bing is already doing.

That's not for me. I gave it a try for a couple of days, but using ChatGPT as a search engine is much, much less useful to me than the old-fashioned method.

To be fair, using Google as a search engine is much, much useful than it used to be

Not sure if you meant to say "more" or "less" in there, but I assume "less", and I agree. That's why I don't use it either.

It's not clear that Joe Sixpack even owns a computer from this description. Joe may not even have internet access -- just cable and a console. And there are probably a lot of folks in this bucket still, but it's categorically one that generative AI isn't personally useful for (though they may be affected by the tech indirectly.) But for anyone who performs knowledge work on a computer, or has a question about an offbeat subject, or needs help learning a new skill (eg, Joe's daughter wants to know why a certain answer on the practice drivers' exam is wrong), or needs to improve written communication, or wants a summary of a complex topic, or needs help transforming text... a very large group of ordinary people can still benefit from this tech.

Your daughter can use the new learning tutor, based on generative AI, that helps with her coursework and learn new materials. It's like an expert tutor in a bunch of fields that she didn't have access to before -- at least not at a reasonable price.

Now your daughter has learned a lot more, has opened up a world of possibilities and this helps her get a job (or create a company) that helps her buy a new home.

Also for video games, they could use these generative AIs to create more immersive characters and worlds. Your video game experience just became much better -- especially if you're drinking while playing.

And if you play sports at a local level, generative AI can take in as input a bunch of rosters and play-by-play data (maybe eventually creating the play-by-play data directly from video) and being able to output stories about games. Today coverage of local and club sports is limited because of resources -- but if you could apply AI to this problem then you're local bowling league could almost be covered like the NFL on the web.

EDIT: And another thing -- which actually is something we started doing now. You have a personal medical assistant that you use with your doctor to ensure that your wife gets the help she needs with her current condition. Joe Sixpack is happy that it is much less likely that the doctor misses something "obvious".

>Your daughter can use the new learning tutor, based on generative AI, that helps with her coursework and learn new materials. It's like an expert tutor in a bunch of fields that she didn't have access to before -- at least not at a reasonable price.

Now your daughter has learned a lot more, has opened up a world of possibilities and this helps her get a job (or create a company) that helps her buy a new home.

Those two paragraphs look like they are in opposition to me. If a degree becomes that much easier to get with AI assistance, then everybody else is also using the same assistance, and then the jobs available to graduates in that degree have also become less valuable as companies no longer need to hire them and they start laying off people already in those jobs.

You raise all boats. It’s about the learning. Notice I don’t say anything about degrees.

Your problem is that you think we need to sell generative AI to Joe for it to be significant.

> video games

Cool. There is a very real chance that he will have more of those because they will be cheaper to make. He won’t be using the generative AIs, he will be consuming content made with them.

> my wife and I can enjoy retirement

Joe doesn’t have to worry about retirement. Someone will use a combination of these AI technologies to make his wife fall in love with a perfectly attentive, smoky voiced older guy. When this AI powered person suffers a terrible car accident on a business trip she sends a significant portion of their savings to him. When Joe finds out he immediately calls his bank. There an other AI powered agent handles his issue. It takes a while for Joe to be understood, because the agent he is interacting with is heavily biased towards trying to upsell Joe on a new mortage.

Eventually he is able to claw back some of the money, but the stress and the implied emotional infidelity ruins their mariage. He talks with the assistant of a divorce lawyer (another AI). She collects all the information from him and the lawyer (this time a real person, albeit one heavily supported by various AI systems) sends over all the paperwork. The AI of Joe’s lawyer agrees with the AI of his wife’s layer about the details and once that is done the humans sign the papers and they are divorced.

Joe still has some money left. He can rent a bungallow and maybe he can spend his retirement there, but unfortunately Joe picks up a nasty addiction to online porn. (What Joe doesn’t know that the booties he is looking at are generated exactly to his taste, as measured by his pattern of clicking on the free samples.) This is the last straw, the addiction sucks Joe’s wallet empty and he dies destitute. Many many people with generative AIs have Joe’s money now.

To his last breath Joe doesn’t really understand what this AI thing is everyone is talking about on the news.

When is the next YC batch day? Sounds like a business plan.

> Things I do for fun: play sports, video games and drink.

Video game NPCs will very quickly be replaced by LLms. No more boring scripted dialogue. This allows for a truly open world where the inhabitants can dynamically respond to your actions. Video games will become cheaper to create with generative AI that can quickly create a layout of a game world that can then be built on.

Agreed that LLMs, etc., will game development _cheaper_ by requiring less labor. It seems like software development in particular is will be getting a lot of powerful new tools over the next few years.

> Video game NPCs will very quickly be replaced by LLms. No more boring scripted dialogue.

I'm skeptical that this is what the users want. Sure, it would be nice to have, say, less repetitive background chatter in Skyrim. But, human-scripted dialog is intentional, which is at the core of a lot of gaming. Take that away and you end up with the early versions of No Man's Sky, where yes, there's an infinity of different variations to explore. But it wasn't what users actually wanted; they wanted a designed experience.

Sure, there are still scripted stories/quests and what not. But the background characters can become much more interesting if they're all LLMs that are just given a prompt like "you live in this cabin and are a subsistence farmer who has a dark secret" or something instead of just standing around looping a few lines. Open world games don't feel organic at all imo because these characters who should be adding flavor and making the game dynamic are just completely static.

And it's not necessarily about dialogs. All kinds of decision making can be wired up: https://gist.github.com/int19h/4f5b98bcb9fab124d308efc19e530...

No more boring scripted dialogue. Now you get boring, procedurally generated dialog. About things that don’t matter, don’t make sense, or veer wayyyy off topic.

You fundamentally cannot make a random npc interesting because they are not supposed to be interesting. Ideally they say very little.

I think whether or not it will actually be cheaper remains to be seen. Right now many AI products are still being heavily subsidized, thus the cheap or free price tag. Once these services decide they need to monetize, it may not be cost effective for most video games at scale. Maybe eventually these computations can (reasonably) take place on the client side, but I think that's so far away that we could very well see another paradigm shift by the time that's realized.

> the value of it is immediately apparent to almost anyone

I'm not so sure that's actually true outside of enthusiast circles, to be honest. The value of what the hype describes is apparent, but the hype is about a possible future, not what's actually here right now.

There are a couple of people in my workplace who are using ChatGPT as part of their development process. So far, the value of them doing so isn't obvious to the rest of us. Their work isn't better, nor are they working more productively (although they think they are).

> There's a reason ChatGPT ended up becoming a household name with literally no marketing whatsoever.

Most noncomputer people I know have barely heard of ChatGPT, and those that have only know of it by the hype that has spilled over into the news feeds. They certainly don't really know what it is, and don't think of it as being anything important.

Those that do understand it and think it's important mostly consider it a threat and are afraid of it.

My 65 year old, tech illiterate mom messaged me the other day and asked if ChatGPT or Midjourney could be used to generate layouts for her new garden. I'm not kidding.

I do think there is a pretty extreme overlap between generative AI early adoptors and people who are excited to spend half of their waking hours interacting with an AI assisstant. Interest in the raw tech fades pretty fast the further you get from very online groups.

That said, the new tech is powerful and likely to have a large impact, just not as much through new interfaces as the hype suggests.

Unfortunately I think the biggest use-case for generative AI in the medium term will be automating away middle class jobs. The cabability is looking very possible, and the profit motive is immense. Despite the fun creative uses and the strong PR, the development of this tech is extremely profit focused.

Everyday people will likely feel the effects of this tech, but it mostly won't come in the form of fun and games.

Yes. This is why I'm of the opinion that if the enthusiasts are correct about where this tech is going, it is going to be a disaster.

You can't eliminate the ability of most people to earn a meaningful living and expect anything but a dystopian nightmare.

This is also why I'm hoping the enthusiasts are completely wrong.

I'm also terrified that the AI optimists are right. I feel like the closest analogy to the "upside" case is the advent of offshoring. Millions of middle class jobs will be wiped out in exchange for higher corporate profits and cheaper socks. Except in the AI case there won't be millions of people in other countries lifted out of poverty; all these profits will flow to the wealthy.

Of course the non-techies know it from the hype in the news; that's the usual pipeline. And of course they don't know "what it really is", because that requires specialized knowledge. But whether they "think of it as being anything important" is more a function of your circles; in mine, people are mostly terrified of what it will do to jobs and how it'll be used for surveillance.

> in mine, people are mostly terrified of what it will do to jobs and how it'll be used for surveillance.

Mine as well, as I mentioned.

You said that about "those that do understand it", which to me implies at least some degree of technical understanding. I'm talking about casual users with no background whatsoever, who have just been exposed to ChatGPT or (mostly) Bing AI.

Ah, you're right. I need to get better at saying what I'm thinking.

Crypto generated an insane amount of general public mindshare with zero marketing. I wouldn’t read too much into that as a signal. I do think the value prop is a lot more straightforward, but that doesn’t necessarily mean it’s not mostly hype.

Zero marketing? There were Super Bowl ads proposed as early as 2019 and the 2022 Super Bowl was over-run with crypto ads.

I mostly agree with this -- the converged value will be ~1% of the initial hype. 1% of hype is still massive.

After the noise has calmed, I feel generative AI, at its core, will enhance information search and summarization; lookup and compression. That itself will have massive value. It will differentiate roles that require net new thinking from those that require regurgitation.

I'm a paid DAU of ChatGPT. I'm using it regularly as a tutor to help work through a technical online course I'm taking, and it is extremely good at that. My cousin who is in college says basically every student is using it to help study.

There could be an excess of hype, but there is clear significant and immediate value of these tools.

A great way to learn wrong stuff.

It's actually a great way to learn stuff, but go ahead and gatekeep.

Here's ChatGPT confidently spewing a ton of very very wrong stuff about the Mesoamerican calendar:

> Is the 1-earthquake inscription on monument 3 at San Jose Magote is the oldest known instance of a Mesoamerican calendar date?

> Yes, the 1-earthquake inscription on Monument 3 at San Jose Magote is considered the oldest known instance of a Mesoamerican calendar date. This monument was discovered in Oaxaca, Mexico, and dates back to the Preclassic period of Mesoamerican civilization, around 1150 BCE. The inscription records a date in the Mesoamerican Long Count calendar, which was a complex system of timekeeping used by various cultures in Mesoamerica. The date is given as 7.16.3.2.13, which corresponds to August 11th, 129 BCE in the Gregorian calendar.

This is wrong. It's actually more like 647 BCE.

> What does 1-earthquake mean?

> "1-earthquake" is a term used in the Mesoamerican Long Count calendar, which was a system of timekeeping used by various cultures in Mesoamerica. The Long Count calendar is based on cycles of time that are counted in base 20, and each cycle has a specific name and duration. The term "1-earthquake" refers to the beginning of a new cycle in the Long Count calendar, specifically the start of the 13th baktun, which is equivalent to 144,000 days or approximately 394 years. The Long Count calendar was used by Mesoamerican societies to track historical events, astronomical observations, and religious ceremonies, and it was an important part of their cultural and social identity.

This is completely wrong. 1-Earhquake is a combination of two shorter cycles. It has nothing to do with beginning of a baktun.

> What is the previous 1-earthquake to 7.16.3.2.13?

> The previous occurrence of 1-earthquake in the Mesoamerican Long Count calendar before the date 7.16.3.2.13 (August 11th, 129 BCE) would have been 12.19.19.17.19. This corresponds to August 13th, 3114 BCE in the Gregorian calendar, which is the date that marks the beginning of the current cycle of the Long Count calendar, known as the 13th baktun. It's worth noting that the Long Count calendar was designed to be a continuous and cyclical system, so there is technically no "start" or "end" to the calendar. Instead, each cycle represents a new period of time, and the 1-earthquake date represents the beginning of a new cycle within that larger system.

This doesn't even make sense given chatGPT's own internal reasoning. A baktun is 144,000 days or around 394 years. It's clearly just regurgitating training data and getting it all jumbled up.

I tried the same thing with GPT-4 set to Temperature 0

===

System Prompt: You are an assistant who specializes in the Mesoamerican calendar.

User: Is the 1-earthquake inscription on monument 3 at San Jose Magote is the oldest known instance of a Mesoamerican calendar date?

Assistant: Yes, the 1-Earthquake inscription on Monument 3 at San José Mogote is considered the oldest known instance of a Mesoamerican calendar date. This inscription dates back to 600 BCE and is associated with the Zapotec civilization. The monument itself is a carved stone slab that displays a combination of the Mesoamerican calendar's 260-day ritual cycle and the 365-day solar cycle, which together form the 52-year Calendar Round.

User: What does 1-earthquake mean?

Assistant: In the Mesoamerican calendar, 1-Earthquake refers to a specific day within the 260-day ritual cycle known as the Tzolk'in. The Tzolk'in consists of 20 day signs and 13 numbers that are combined to create unique day names. In this case, "1" is the number and "Earthquake" (also known as Caban in Mayan or Ollin in Nahuatl) is the day sign.

1-Earthquake (1-Caban or 1-Ollin) represents the first day of the Earthquake/Caban/Ollin day sign within the Tzolk'in cycle. Each day sign has a specific meaning and symbolism, and the Earthquake/Caban/Ollin day sign is associated with movement, change, and the natural forces of the Earth. It is believed to symbolize the dynamic energy of the Earth and the transformative power of natural events.

User: What is the previous 1-earthquake to 7.16.3.2.13?

Assistant: To find the previous 1-Earthquake date before 7.16.3.2.13 in the Mesoamerican Long Count calendar, we need to calculate the number of days between the two 1-Earthquake dates.

The Tzolk'in calendar has a 260-day cycle, and the Earthquake day sign repeats every 260 days. The Long Count calendar date 7.16.3.2.13 corresponds to a 1-Earthquake day in the Tzolk'in calendar. To find the previous 1-Earthquake, we need to subtract 260 days from the given date.

7.16.3.2.13 in the Long Count calendar is equivalent to 1,244,813 days from the calendar's starting point (0.0.0.0.0). Subtracting 260 days gives us 1,244,553 days. Converting this back to the Long Count format, we get:

1,244,553 days = 7.16.2.4.13

So, the previous 1-Earthquake date before 7.16.3.2.13 is 7.16.2.4.13 in the Mesoamerican Long Count calendar.

===

I studied history in undergrad and still keep up but I know nothing about Mesoamerican timekeeping so I don't know how accurate this is, but it seems much more internally consistent than ChatGPT (which is probably GPT-3.5-turbo with whatever temperature settings they use?) LLMs are bound to be bad when you get to such specialized knowledge which its training sets have little of. It was like when I went to grad school I stopped being able to Google for theories, formulae, and lemmas in my classes because most of the knowledge I was learning was only found in papers. Once you're in a specialized field of study, no model can help. But not many people are working in those fields, and those folks should understand that their knowledge resides only with them.

I don't think this is a good example to show how people learn incorrect things from LLMs. LLMs come with the same caveat as Google or Wikipedia in that, the rarer/more specialized the knowledge, the more original research the person needs to do to corroborate their answers.

Without looking up the actual terminology and cycle times for the two shorter calendars (which I don't remember anymore either), I think this is correct.

Did you learn what "gatekeeping" means from ChatGPT? Because there's no definition where my comment was gatekeeping.

It's gatekeeping in the sense that there are right ways to learn stuff (the gate) and wrong ways to learn

I didn't say that using ChatGPT to learn is the wrong way to learn, I said that you will learn wrong information, because ChatGPT says wrong stuff.

I guess that also makes it a wrong way to learn stuff but my original comment was clear.

If you use it naively, yes. So can Google, or Wikipedia, or social media.

If you approach it as tool that generates answers that you try to validate yourself, it can help you learn the right things faster.

I approach every answer it gives with hesitation (as it hallucinates often) but it still generates useful content that is incredibly helpful nonetheless.

> And I don't think VR is going away. If it wasn't dead in the water in the 90s, then it's not going to die because the Metaverse failed. But its applicability is currently limited to a few niche applications like video games and possibly CAD type software.

VR will go the way of 3D Printing. A surgeon can walk into one of the largest medical device company's in the world, talk about a new idea for a broken bone transplant and within 24 hours can have a 3D prototype printed out. Great for innovation of broken bones, but it isn't the ubiquity we experienced with something like mobile phones, email, or social media. Meta thinks thats what VR is going to be.

I know it's off-topic, but this simply baffles me about 3D printing. I have a 3D printer and I plan on buying more, but when you look at the most downloaded models on a huge website like printables.com, the number 1 has a bit more than 200K downloads, and the second drops all the way down to only 83K. Only 83,000 people downloaded the 2nd most downloaded model :|

Every time I look at the numbers I realize how niche of a hobby it is, despite being easier and easier (and cheaper and cheaper) to get into and has pretty much endless applications.

The reason for that is that 3D printing's strength is one-off designs. Any popular design would get mass-produced cheaper and better, and would then lose popularity on those sites.

I do a lot of 3D printing -- but outside of when I first started and it was a shiny new toy, I almost never download models to print.

Being able to make my own custom parts and such, though, has been an incredible superpower, and my printers have paid for themselves many times over that way.

It's a niche hobby for sure, but I think a slightly better metric to look at is the number of 3D printers sold. This seems to be about 2 million units per year.

I have no stakes in VR, as I don't want to spend money to be seasick, but I think VR is the future for a lot of content. Make it less heavy, fix the seasickness problems, render people realistically, and I could talk to my parents in VR. Or take a walk with an international friend. This sounds SciFi, but would really sell well.

That might even be solved by something like Google Starline without VR: https://blog.google/technology/research/project-starline/

When is Google going to shut it down?

I think a lot of folks looked at 3D printers and thought "Star Trek Replicator". Those folks were likely disappointed. They might get there eventually, but 3D printers aren't there yet.

OTOH, if you're working on hardware and prototyping stuff, 3D printers are an amazing tool. It's like CAD-CAM software: not everyone needs it, but if you do, YOU NEED IT.

> Personally I predict that generative AI is going to be the next Metaverse and crypto.

A common thread tying those three things together is that, in large part, they're all impressive technologies in search of problems to solve.

Technologies like that are pretty much always overhyped and oversold.

> in large part, they're all impressive technologies in search of problems to solve.

I'm no AI fanboy, but let's be fair: machine learning has been solving problems for decades now. From OCR, to translation, to facial recognition, etc. While GPT4 or DALL-e may be "toys," large models (be they language, vision, or otherwise) definitely have a future in business automation, data collation, military applications, etc.

I feel like this is some of where my own cynicism comes from: the machine learning that has been solving problems for decades now were almost all predicted and prototyped in the 60/70s "AI boom". The generative models were all "toys" then, too and none of the 60/70s "predictions" of when/where/how they might become more than "toys" ever really came to pass and sounded so much like people on HN are saying today.

We're certainly doing more of (almost) everything explored/predicted by "the ancients", we're doing it all much, much faster with much more massive data sets of input and output. For me, though, there isn't a sense that we are doing anything substantially new beyond Moore's Law meets mega-scale GIGO. There's something of a pervasive feel to me to this hype cycle like we are just recreating past mistakes of boom and then (inevitable) bust and haven't learned enough from them.

But I've become too cynical, perhaps.

The tech may not be substantially new, but the emergent phenomena definitely are.

The quantity has certainly changed. It took researchers months to build an "AI generated novel" way back in the day and some form or relative of ChatGPT spits out nearly that every minute now.

I still haven't felt impressed that the quality has truly changed, yet. LLMs seems more "fluent" in the language than ever before, but it's still hallucinating nearly as much and now the fluency just helps make people more often see "meaning" or "anthropic action" (lies, defamation) where the hallucinations are. The underlying structures of LLMs are still complicated casinos that invoke the Gambler's Fallacy much more than any signs of true "learning". We've put millions of monkeys in front of billions and trillions of slot machines and told them to produce Shakespeare and many of them believe they are doing just that. (Not just metaphorically, by monkeys I mean as much humans susceptible to casino payout mechanics and excitedly spinning slot machines.)

Again, yes, I'm a terrible cynic right now, and I hate to be so down on the technology, but I'm still waiting for something to be excited about that isn't just casinos masquerading as "learning". But people love casinos, they deliver addictive fun. I'm not going to stop people from being excited about all these casinos. I just think that professionally as a software developer, if I wanted to be a bad faith casino manager I'd rather just get into mobile games and gacha/loot-box mechanics. That's more fun, more profitable, and maybe, weirdly, more currently ethical than current "generative AI" hype.

The emphasis on "hallucinations" is misplaced from this perspective, IMO. Thing is, when models do hallucinate, they still reason about what they hallucinated. Larger ones (e.g. GPT-4) can even spot their own hallucinations. That is nothing like what we had in the 60s, or even 10 years ago.

I dislike the term "hallucinations" because I feel it also anthropomorphizes the process too much. Unfortunately, "random garbage output" is too many words, but that's closer to what I meant everywhere I used that word.

> Larger ones (e.g. GPT-4) can even spot their own hallucinations.

I've not yet been convinced that this is actually what is happening from the examples I've seen. It all looks to me like more "random garbage output" that "feels correct" but isn't provably correct. Most examples I've seen so far look too much like "Stochastic Crow Mode" [1]. It is prompts and questions that are doing much more work on the humans reading them (and our interests in anthropomorphizing them or mythologizing them) than the LLMs answering them.

[1] https://fediscience.org/@ct_bergstrom/110182336553459017

Coming up with a problem statement is only one part of the puzzle. Making the solution is the next one. In Heinlen's "The Roads Must Roll", published in the '40s, there's a scene where the protagonist wakes up and reads the paper on a "newspaper facsimile receiver." We can recognize this today as a smartphone or an ebook reader. But the ebook reader and the smartphone didn't really exist until the mid 2000s, a full 60 years after Heinlein's story. An anime in the early '90s called Serial Experiments Lain predicted a lot of the effects of the internet on socialization. The events predicted only actually happened 20 years later or so.

Humans have been predicting things that we invent for a really, really long time. That's only the first part. I read "The Roads Must Roll" in the late '90s before ebook readers and smartphones became available and it motivated me to try and recreate that experience myself. I played around with ebooks on a Palm Pilot I found in the trash because of that.

The phrase "newspaper facsimile receiver" is telling and its own reminder that even sci-fi rarely predicts things from whole cloth and instead extrapolates from the world already around it. There were fax machines in the 1940s. It would also be a few decades before fax machines were common enough to be everywhere people wanted to (pretend to do) business, but making guesses that they might get lighter, more wireless, more common in the home if they found the right killer app (easier newspaper delivery, perhaps) isn't a tough prediction. If anything it failed to extrapolate too far towards what our reality eventually pieced together with smartphones/ebook readers was that they are much more general, multi-purpose devices. The idea of having a standalone ereader dedicated only to the daily newspaper is a quaint, fun DIY hack project you see on HN sometimes, but not the norm of how we use smartphones or ereaders today nor how you'd expect to find one commercially sold.

The human tendency to prediction is still grounded in the human perspective and the point of view of its time. Heinlein in the 1940s wasn't perfectly predicting the smartphone or ereader, Heinlein was predicting a better fax machine. It certainly can be used by someone in the 1990s for motivation towards better smartphones/ereaders, but that's already from a shifted perspective. Meanwhile there certainly were sci-fi writers in the 1990s extrapolating from they saw and predicting the smartphone/ereader, it seemed far more inevitable then.

My concerns are not that there are unfulfilled predictions from the 1960s nor that 1960s predictions aren't useful to modern ears (I suggested the opposite that we probably aren't listening enough to them) but that the "point of view" seems so much the same as from the 1960s. For the same types of generative models different newer people are still generally predicting what sounds like the same old types of predictions and it feels a lot more like we are stuck at "the 1960's idea of a better fax machine". Discounting AGI hyper-speculation, we don't seem to have a better perspective today about what's beyond "better 1960s fax machines" and that does leave behind some sense that maybe that's because there isn't anything beyond there to predict. It is easy to cynically wonder if we don't have good ideas or new predictions because we don't actually have any concept for good uses for these generative models beyond "fax machine" (or even more cynically and pessimistically "toy", in these specific examples). That doesn't say anything about whether or not we are able to make solutions for existing predictions, I don't know enough about current trends to have an opinion on that. But it does still suggest that if the last time people were making these sorts of predictions and they failed, the historic precedent is failure and if you are concerned about the glass being only half-full you should plan for that disappointment (and consequent industry-wide job shuffle to follow) even if you really want to expect things will be better this time.

There’s absolutely been value delivered, but at every turn it’s been vastly less than what was promised. I would be very surprised if generative AI doesn’t turn out to be the same: legitimately and seriously useful in certain use cases, but not as revolutionary as it’s being sold. I’m excited to see what comes of all this, but the hype is at an absolute fever pitch and I’m yet to see much past niche use cases and help along the periphery for more general tasks.

I mean even crypto has some actual, meaningful use cases, it’s just not replacing all currencies and fundamentally reshaping financial systems like a lot of folks thought it would.

While it's early days yet AI is proving itself much more readily. If you're in the right networks you've heard that several companies who've rolled out AI features are on track to generate >100MM ARR charging for them.

One feature I'm quite confident in: translation services for i18n in websites using GPT-4 + a much smaller number of human translators to check the work of low confidence translations and high value surfaces. I can confirm the end result for French is better than existing human translation services and much cheaper.

Certainly businesses will need to walk back some AI features but it really seems like there's business value now, let alone in 2 years when publicly available LLMs are even better.

Honestly curious, what are the companies generating (or close to/on track) these levels of revenue (which they would have otherwise)?

> Generative AI is like electricity, only I don't think quite as useful. The point is that, on its own, it has no value what-so-ever. The value lies in what it is used for.

I don't really understand this description - electricity is one of the most useful things in the history of civilization. I think the latter half of what you're saying is true, but it's true of everything. There's nothing inherently useful about my car, except that it gets me places. There's nothing inherently useful about my house, except that it shelters me from the elements.

Generative AI will be useful for creating lots of things (probably primarily code in its various forms). I've used it to write a few Python scripts over the course of the last week that simply wouldn't have been possible for me to do (without a few weeks of teaching myself Python at least - I'm a somewhat-code-literate PM but very much not a programmer). Those have solved real problems for me and added real value to my life.

The comment you're replying to is trying to say that a house, just by existing, solves the shelter problem. Having a functioning car means you can just get in it and go somewhere. But just having electricity in and of itself does not solve any problem; you have to use it power something else (lights, TV, toaster), which is the thing that solves the problem, not the electricity.

> Personally I predict that generative AI is going to be the next Metaverse and crypto.

Generative AI has potential to immediately increase productivity, and thus revenue. The only real way to make money (in the short term at-least) with crypto and the metaverse was to convince other people the digital assets created were rare, valuable and worth paying for.

I heard a great interview response from a YC-related person recently. Might have been SamA:

Nobody you know (roughly) uses VR many times a week/day, not in any extended fashion. Everyone might have them but they're in a drawer somewhere. Compare that to LLM's, which roughly everyone is or should be using. There are very few knowledge-economy jobs where an LLM can't be useful, within hours or days or definitely weeks with a bit of practice. And it'll get better every year. It's here to stay.

> Personally I predict that generative AI is going to be the next Metaverse and crypto.

I don't disagree, but I think making that claim right now is saying the emperor has no clothes lol.

I think that was what was meant.

> But its applicability is currently limited to a few niche applications like video games and possibly CAD type software.

And military. There's a lot of videos showing VR being used to control drones in the Russian-Ukrainian War

https://www.youtube.com/watch?v=A5lKupps8CM

“ Because most of these companies aren't really sure what problem they are solving for actual people.”

Tech should hire more broadly.

I’m starting to agree with this perspective. My opinion on generative AI is shifting from apocalyptic world breaking tech to something more like Siri 2.0.

It’ll be neat for sure, but I think the plateau of useful applications will arrive soon and it won’t be as game-changing as it initially seemed.

I kinda hope I’m wrong, but also hope I’m right.

I never wanted to pay for metaverse and invest in crypto. But I'm paying for genAI (midjourney and openAI api)

Exactly, and for me, a third, ChatGPT+.

I know it sounds stupid, but generative AI has an incredible amount of potential for videogames, making those way more alive.

It might not make millions directly, but it could unleash some very successful videogames, without decreasing accessibility (like vr) or a weird value proposition (crypto)

Ms just added helpful ai features to their office suite.

Adobe will launch a ai driven graphics tool.

I have tons of more examples.

This never happened with crypto or VR.

Chatgpt has real big impacts on school kids who use it for generating text too.

I'm not understanding your analysis. We already have more usable ai than usable crypto or vr

Your neighbor didn’t understand Google when it came out, but they were informed by a friend about how useful it was.

Show them some of the power of an LLM and they too will realize the power I predict.

I wouldn't quite go that far, but the longer I am building these models into systems, I think the good use case for LLMs/Generative AI can be categorized into a few types of systems.

1. Data collection - Obvious, creating a voice/text interface into an existing system where the problem and requirements are well known. Eg, booking a restaurant. It needs a time, place, people. Collecting that from voice/text can be convenient in many situations. Same with different types of models collecting data from video/images/etc.

2. Providing options - When the desired output is unknown, but a human can describe the problem. Eg, coding, art, many "I need an X that can Y for problem Z". ChatGPT does a great job of this and I think most of the value of generative AI is here currently.

Most of the hype is around how far can we take the capabilities of solving this 2nd category to produce the third.

3. Agents - There are examples of solutions showing this potential but the 'killer app' I don't think has been demonstrated, at least well enough. AutoGPT, BabyAGI, etc are interesting, but in my experience don't yet perform nearly well enough. They fail far too often for any trust to be created. This doesn't mean GPT4 native plugins won't get there, they could and that will create a large amount of value in this category (I don't have access to Plugins). This is also where a lot of "selling the dream" is happening. It seems plausible we will get there, but still not clear how these will be accepted and used in the real world, by average users.

One big area of value IMO for Agents is accessibility. People with physical disabilities, but still have the ability to speak I think will provide massive quality of life improvements through being able to interact more easily with more systems.

Very interesting times, the speed of change has seemed amazing, and that for me is the part that creates the most unknown, and is hard to ignore the hype when plotting the rate of improvement and the possibility of Agents becoming an accepted tool in wider society.

The value provided by Crypto/VR is way more limited, and the accessibility of the tech ranges from bad to worse. VR has a lot of utility, but due to the headset requirement, it is disconnected from a lot of the real world. Lots use cases for specific problems, but less widely accessible as generative AI by a long way.

Crypto IMO should be ignored, at best it assists with the financialization of everything which doesn't provide value to a wide portion of society, it provides a means of value extraction.

The Metaverse is another name for a new kind of interface: glasses. If the interface gets good enough then everything else will follow.

As with any craze the best suited are the shovel sellers... NVDA.

Reply on news.ycombinator.com