I genuinely think systems like Unison are "the future of computing"...
But the question is when that future will be.
Part of the beauty of these sorts of systems is just that the context of what your system actually does is in one system, you aren't dealing with infra, data, and multi-service layers
Maybe that means it is a much better foundation for AI coding agents to work in? Or maybe AI slows it down? we continue to try and throw more code at the problem instead of re-examining the intermediate layers of abstraction?
I really don't know, but I do really want to learn more about is how the unison team is getting this out in the market. I do think that projects like this are best done outside of a VC backed model... but you do eventually need something sustainable, so curious how the team things about it. Transparently, I would love to work on a big bet like this... but it is hard to know if I could have it make financial sense.
With all that, a huge congrats to the team. This is a truly long-term effort and I love that.
not disagreeing with your point here, or in the follow-ups of the pain of https for "local network" apps... but I really wish that we could get to a place where we could get away from this distinction. Obviously, ipv6 is not that easy or realistic, but that really is, imho, the "right" long term answer.
Having gone down the path of being able to just spin up "local" services that get a publicly routable (but most often firewalled off) ipv6 IPs and then good DNS integration is really neat... but still requires lots of technical chops. I wish that weren't the case
I work with embedded Linux stuff and MCU stuff where we make a significant number of units. Even in an IPv6 world, there's no way each of those would get their own public static IPv6 address with an associated DNS record just for the purpose of being able to spin up a debug web interface. It's explicitly desirable for these devices to not be reachable through the public Internet.
Well then you set your firewall to default-deny. It doesn't make sense to hobble the internet just because NATs are inadvertently a convenient firewall.
DHCP does give you a globally unique IP address when your ISP has allocated a prefix to your router, that's how all the Internet-connected IPv6 devices get their addresses. Where is our misunderstanding?
For many of these systems, I don't control the user's router. I don't know how you imagine I'm supposed to create DNS records for each device when they're assigned some random IP address at some random network I don't control.
Have the device ping a central server and create randomword.centralserver.com, for example. However, if the problem is the DNS record, why has this thread been exclusively about globally routable IP addresses until now?
In https://news.ycombinator.com/item?id=45957048, addisonj suggested that the problem stems from the distinction from "local" and "global", and that with IPv6, you don't need that distinction.
That quite naturally flows into the question: okay, how are these devices supposed to get global IPv6 addresses then?
Yes, with IPv6, there are are enough addresses that you don't need to use NAT. All IPv6 devices that are connected to the internet have global IPv6 addresses. I don't quite understand the question here, it seems to me that we're asking "but how could we possibly do this entirely mundane everyday thing?".
Not all devices connected to the Internet have globally unique IPv6 addresses, SLAAC and often DHCPv6 makes local v6 addresses. Where's the globally unique IPv6 address supposed to be coming from?
So you're talking about being assigned temporary globally unique addresses, if the network the device happens to be on at any given time happens to be set up in a certain way?
I still don't understand how this is supposed to help.
In https://news.ycombinator.com/item?id=45957048, addisonj suggested that the problem stems from the distinction from "local" and "global", and that with IPv6, you don't need that distinction.
This helps because you don't have a NAT distinguishing between "local" and "global", all devices are in the global namespace.
All the comments after that have been about solving an arbitrary and ill-defined problem with goalposts that keep shifting from globally unique addresses to DNS hostnames to permanent addresses.
How does getting a temporary globally unique IPv6 address from DHCPv6 solve any of the issues surrounding how new web technologies aren't available in "insecure contexts"?
I assumed that the suggestion was that you could assign a device a permanent IPv6 address, because I can easily imagine that as a part of a solution to the HTTPS issue. When every device has a permanent IPv6 address, and if every device is reachable through said IPv6 address, you could, in principle, also automate assigning each device a DNS record and set up SSL that way. It would be a pretty terrible solution that's way more complicated than just using a local address over HTTP, but it makes sense.
I have no idea how to even begin translating maybe getting temporary unique addresses through DHCPv6 into a solution to the HTTPS issue.
You can get a static prefix from your ISP. After you get the static prefix, it's up to your local network to make the local parts of the address static. There's no reason why your DHCP server can't give the device a static address, it's not like it's going to run out.
Then again, you don't need a static address to get a TLS certificate. You don't need an address at all! All you need is a domain name.
Some of the devices I'm talking about are running on my own residential internet connection or my sister's. Some are running on whatever corporate or residential or 4G network happens to exist where I need to interact with them. Some are running on whatever network the user has.
How does your proposed suggestion of getting a static prefix from an ISP apply to those situations? Should I start calling customers to get them to ask their ISP for a static IP address?
> Then again, you don't need a static address to get a TLS certificate. You don't need an address at all! All you need is a domain name.
> I was discussing with a friend that my biggest concern with AI right now is not that it isn't capable of doing things... but that we switched from research/academic mode to full value extraction so fast that we are way out over our skis in terms of what is being promised, which, in the realm of exciting new field of academic research is pretty low-stakes all things considered... to being terrifying when we bet policy and economics on it.
That isn't overly prescient or anything... it feels like the alarm bells started a while ago... but wow the absolute "all in" of the bet is really starting to feel like there is no backup. With the cessation of EVs tax credits, the slowdown in infra spending, healthcare subsidies, etc, the portfolio of investment feels much less diverse...
Especially compared to China, which has bets in so many verticals, battery tech, EVs, solar, then of course all the AI/chips/fabs. That isn't to say I don't think there are huge risks for China... but geez does it feel like the setup for a big shift in economic power especially with change in US foreign policy.
I'll offer two counter-points. Weak but worth mentioning. wrt China there's no value to extract by on-shoring manufacturing -- many verticals are simply uninvestable in the US because of labor costs and the gap of cost to manufacture is so large it's not even worth considering. I think there's a level of introspection the US needs to contend with, but that ship has sailed. We should be forward looking in what we can do outside of manufacturing.
For AI, the pivot to profitability was indeed quick, but I don't think it's as bad as you may think. We're building the software infrastructure to accomodate LLMs into our work streams which makes everyone more efficient and productive. As foundational models progress, the infrastructure will reap the benefits a-la moore's law.
I acknowledge that this is a bullish thesis but I'll tell you why I'm bullish: I'm basically a high-tech ludite -- the last piece of technology I adopted was google in 1996. I converted from vim to vscode + copilot (and now cursor.) because of LLMs -- that's how transformative this technology is.
> which makes everyone more efficient and productive
There is something bizarre about an economic system that pursues productivity for the sake of productivity even as it lays off the actual participants in the economic system
An echo of another commenter who said that its amazing that AI is now writing comments on the internet
Which is great, but it actively makes the internet a worse place for everyone and eventually causes people to simply stop using your site
Somewhat similar to AI making companies more productive - you can produce more than ever, but because you’re more productive, you don’t hire enough and ultimately there aren’t enough people to consume what you produce
> There is something bizarre about an economic system that pursues productivity for the sake of productivity even as it lays off the actual participants in the economic system.
Not only does it lay off many of them, but it expects the rest to work longer hours with fewer raises in many cases.
> many verticals are simply uninvestable in the US because of labor costs and the gap of cost to manufacture is so large it's not even worth considering.
I think this is covered in a number of papers from think tanks related to the current administration.
The overall plan, as I understood it, is to devalue the dollar while keeping the monetary reserve status. A weaker dollar will make it competitive for foreign countries to manufacture in the US. The problem is that if the dollar weakens, investors will fly away. But the AI boom offsets that.
For now it seems to work: the dollar lost more than 10% year to date, but the AI boom kept investors in the US stock market. The trade agreements will protect the US for a couple years as well. But ultimately it's a time bomb for the population, that will wake up in 10 years with half their present purchasing power, in non dollar terms.
There are no think tanks mentioned in that article though.
The accord also appears to contradict the stated goals:
> Miran proposes a modern equivalent of the 1985 Plaza Accord, which he refers to as the Mar-a-Lago Accord. The goal would be coordinated currency appreciation among U.S. trading partners to address the dollar's overvaluation
> For now it seems to work: the dollar lost more than 10% year to date
...and I thought American Labor was having something of a moment in 2024-2025. The law of unintended consequences may have surprises in store for the planners in the coming years.
I think an interesting way to measure the value is to argue "what would we do without it?"
If we removed "modern search" (Google) and had to go back to say 1995-era AltaVista search performance, we'd probably see major productivity drops across huge parts of the economy, and significant business failures.
If we removed the LLMs, developers would go back to Less Spicy Autocomplete and it might take a few hours longer to deliver some projects. Trolls might have to hand-photoshop Joe Biden's face onto an opossum's body like their forefathers did. But the world would keep spinning.
It's not just that we've had 20 years more to grow accustomed to Google than LLMs, it's that having a low-confidence answer or an excessively florid summary of a document are not really that useful.
Chatting with Claude about a topic is in another universe to google search.
I default to Claude for almost everything where I want to know something. I don’t trust Google’s results because of how weighted they are to SEO. Being good at SEO is a separate skill set.
The answers are not low confidence, cite sources, and can do things that Google cannot. For example: I used Claude to design a syllabus to learn about a technical domain along with quizzes and test suites for verification. It linked to video series, books, and articles grouped by an increasingly complex knowledge set.
Is this really true re: "modern search"? Genuine question because this is probably outside of my domain. I'm just trying to think of industries that would critically affected it we went from modern search to e.g. AltaVista/Yahoo/DogPile and kind of coming up empty except in that it might be more difficult for companies that have perfected modern SEO/advertising to maintain the same level of reach, but I don't think that's what you're alluding to?
I think there's a bubble around AI, but I don't think I agree with this argument. Google search launched in 1998, and ChatGPT launched in 2022.
In 2001, if Google had gone under like a lot of .com bubble companies, I think the economic impact visible to people of the time would have been marginal. There was no Google News, Gmail, Android, and the alternatives (AltaVista, Ask Jeeves, MSN Search) would have been enough. Google was a forcing function for the others to compete with the new paradigm or die trying. It wasn't itself an economic behemoth the way it is today.
I think if OpenAI folded today, you'd still have several companies in the generative AI space. To me, OpenAI's reminiscent of Google in the late 90s in its impact, although culturally it's very different. It's a general purpose website anyone with an internet connection can visit, deep industry competitors are having to adapt to its model to stay alive, and we're seeing signs of a frothy tech bubble a few years after its founding. People across industry verticals, government, law, and NGOs are using it, and students are learning with it.
One counterpoint to this would be that companies like Google reacted to the rise of social media with stuff like Google+, but to me the level to which "AI" is baked into every product at Google exceeds that play by a great margin. At most I remember a "post to plus" link at the top of GMail and a few hooks within the contact/email management views. In contrast, they are injecting AI results into almost every search I make and across almost every product of theirs I use today.
If you fast forward 20 years, I would be surprised if companies specializing in LLMs were not major players the way today's tech giants are. Some of the companies might have the same names, but they'll have changed.
> At most I remember a "post to plus" link at the top of GMail and a few hooks within the contact/email management views.
Google probably could have been whatsapp but to push Google+ scrapped a successful gmail chat for hangouts, which you had to visit Google+ feed each time to open at first.
Another thing to note about China: while people love pointing to their public transit as an example of a country that's done so much right, their (over)investment in this domain has led to a concerning explosion of local government debt obligations which isn't usually well-represented in their overall debt to GDP ratios many people quote. I only state that to state that things are not all the propaganda suggests it might be in China. The big question everyone is asking is, what happens after Xi. Even the most educated experts on the matter do not have an answer.
I, too, don't understand the OP's point of quickly pivoting to value extraction. Every technology we've ever invented was immediately followed by capitalists asking "how can I use this to make more money". LLMs are an extremely valuable technology. I'm not going to sit here and pretend that anyone can correctly guess exactly how much we should be investing into this right now in order to properly price how much value they'll be generating in five years. Except, its so critical to point out that the "data center capex" numbers everyone keeps quoting are, in a very real (and, sure, potentially scary) sense, quadruple-counting the same hundred-billion dollars. We're not actually spending $400B on new data centers; Oracle is spending $nnB on Nvidia, who is spending $nnB to invest in OpenAI, who is spending $nnB to invest in AMD, who Coreweave will also be spending $nnB with, who Nvidia has an $nnB investment in... and so forth. There's a ton of duplicate-accounting going on when people report these numbers.
It doesn't grab the same headlines, but I'm very strongly of the opinion that there will be more market corrections in the next 24 months, overall stock market growth will be pretty flat, and by the end of 2027 people will still be opining on whether OpenAI's $400B annual revenue justifies a trillion dollars in capex on new graphics cards. There's no catastrophic bubble burst. AGI is still only a few years away. But AI eats the world none-the-less.
My point is not that value extraction wouldn't happen, my point is simply that in addition to the value extraction we also made other huge shifts in economic policy that taken together really seem to put us on a path towards an "AGI or bust" situation in the future.
Is that a bit hyperbolic? isn't this just the same as dotcom and housing bubbles before where we pivoted a bit too hard into a specific industry? maybe... but I also am not sure it would be wise to assume past results will indicate future returns with this one.
AI is appealing to the investors not because it solves human problems, but because it solves some of the problems of previous bubbles.
When we wired the world for the Internet in the 1990s, or built railways across the continent in the 1800s, we eventually reached a point where even the starriest-eyed investors could see they've covered effectively the entire addressable market. Eventually AOL ran out of new customers no matter how many CDs they mailed out, or we had connected every city of more than 50 people with steel rail, and you could hear the music was slowing down.
By dangling the AGI brass ring out there, they can keep justifying the expenditure past many points of diminishing returns, because the first thing we'll ask the Omnipotent AGI is how to earn the quadrillions spent back, with interest.
It also has the benefit of being a high-churn business. The rails laid in 1880, or the fiber pulled in 2000, were usable for decades, but in the AI bubble, the models are obsolete in months and the GPUs in years. It generates huge looking commercial numbers just to tread water.
Core R&D, training, public education at all levels, health care, small businesses of all kinds, arts and entertainment, and many more.
Basically IP generation at all levels, and high-touch face-to-face service businesses.
The US already had the foundations for a post-industrial economy, but the cancer of extractive financialisation ate it alive. Instead of expanding into diversity it contracted into a startup culture that's always been a thin front for Wall St.
So the result is that it's very hard to start a mid-rank IP or service business. You have small-scale cottage-level producers like authors, musicians, video creators, and indie app developers relying on huge monopolies like Amazon, Spotify, Google, and Apple, and you have bigger projects playing the startup game and scrambling for funding rounds, where anyone who doesn't become a unicorn is a loser.
There's plenty of space for businesses between those extremes, but the economy isn't set up to support them. The space isn't dead yet, but it could be much, much bigger.
Universities and corporate R&D have similar issues. Metrics support conformist publishing and resume-development, not risky original talent and exploration.
Tariffs and brain drains are absolutely toxic and are going to nuke all of these spaces.
You can’t have any of this, literally, without manufacturing at home. You can’t do R&D of any sort if you don’t have the capability to produce things, to send an email or walk over to the company making it, speak with someone and have a consultation around what it is you hope to achieve because it’s the people who machine, who fabricate, who apply chemical treatments and layers of paint, who bend who etch and understand tool clearance requirements and so on that know this stuff. Manufacturing is a broad term.
I’m reminded of recent-grad architects whose proposed ideas are bereft of consideration of material properties and pitfalls.
I’m also reminded of Air Force aircraft engineers needed to be told their parts have to be adjusted because they can’t be machined. And the person who knows what needs to be done is Bob whose hands are covered in grease and oil because there’s not enough orange Zep on this planet to clean that guy.
To use China as an example: their entire pipeline from conception of idea to the end-customer is in China. They don’t even need to sell it externally (and frequently they don’t).
The West is fucked because they think pushing things around is being productive. GDP is a reflection of this failure because it’s a flawed, abused metric that’s devoid of where money actually ends up (most often it’s Chinese firms doing the work), so your GDP is looking great meanwhile you are incredibly unproductive over all. It’s a meme tool used for nonsensical political posturing.
Simply put: you can’t start R&D in the middle and not take into consideration a pipeline. It doesn’t work that way. People don’t have the experience or knowledge.
>> I was discussing with a friend that my biggest concern with AI right now is not that it isn't capable of doing things... but that we switched from research/academic mode to full value extraction so fast
lol, I read this few hours ago, maybe without enough caffeine but I read it as "my comment from 70 *years* ago" because I thought you somehow where at the The Dartmouth Summer Research Project on Artificial Intelligence 1956 workshop!
I somehow thought "Damn... already back there, at the birth of the field they thought it was too fast". I was entirely wrong and yet in some convoluted way maybe it made sense.
I think the answer to that right now is highly workload dependent. From what I have seen, it is improving rapidly, but still very early days for the software stack compared to Nvidia
Really impressive... but wow this is light on details.
While I don't fully align with the sentiment of other commenters that this is meaningless unless you can go hands on... it is crazy to think of how different this announcement is than a few years ago when this would be accompanied by an actual paper that shared the research.
Instead... we get this thing that has a few aspects of a paper - authors, demos, a bibtex citation(!) - but none of the actual research shared.
I was discussing with a friend that my biggest concern with AI right now is not that it isn't capable of doing things... but that we switched from research/academic mode to full value extraction so fast that we are way out over our skis in terms of what is being promised, which, in the realm of exciting new field of academic research is pretty low-stakes all things considered... to being terrifying when we bet policy and economics on it.
To be clear, I am not against commercialization, but the dissonance of this product announcement made to look like research written in this way at the same time that one of the preeminent mathematicians writing about how our shift in funding of real academic research is having real, serious impact is... uh... not confidence inspiring for the long term.
I had an ex90 on pre-order for a long time, placed it within the first ~30 days of it being open.
It looked to be (and is!) an absolutely beautiful vehicle and also seemed to be making choices in the hardware (lidar) that I hoped, would, eventually deliver a combination of safety and self-driving capabilities that would be unmatched. I was willing to pay a premium and knew that it would take some time for the self-driving to come to fruition, but figured it would be a capable vehicle until that point in time.
But dang, what a botched launch. Not only were there all these issues, which are insane to me that Volvo didn't have more people in social media / subreddit, but also from a financial perspective the car is just insanely hard to get into. Lease terms were absolutely terrible.
I ended up getting a Hyuandai Ioniq 9 and am really glad I went that direction. Yeah, it doesn't offer as much as a Tesla in terms of FSD, but it also has better build quality and interior quality nearly matching the Volvo. I like the styling (but I know some do not), and it has actual physical controls for the stuff I care about and the best heads up display I have used (favorite feature: you get photos of incoming caller). NACS is also great... but I can't bring myself to take 2 spots yet at superchargers.
I am experiencing a strong sense of "why didn't I think of that" while also really hoping it isn't another strong, family friendly concept that gets quickly enshittified for profit.
Seriously, kudos, for a great concept, good website, and really, not that bad of pricing. Sure you can do it cheaper DIY... but where is the fun in putting an office-styled VOIP capable phone in a kid's bedroom? (though converting an old-phone to tunnel over VOIP sounds like a fun weekend project to do with my pre-teen)
But... dang, does it feel like yet another thing that will start great and get terrible over time or just dropped and be e-waste. Kid cell-phone plans that don't give me choice of provider, youth-focused budgeting/saving apps that are 4x more expensive than just a classic bank account and require an app to effectively use, and by far, worst of all, all the "kid" versions of tablets, youtube kids (which I can never get to not show ads even though I pay for premium!), that claim to give parents control... but really just seem like the minimum effort to make parents feel like they are putting in guardrails while still being designed to maximize the addiction early.
While I am really glad we are trying to build tech that helps kids have a better relationship each other while still using technology... it seems like most still fall to pressure of profit and either term into extremely over-priced offering that is hard to justify or can't make it and turn into junk with no re-use.
Once again, this product, right now, does not look to be that... but now having been bit a few times, I am much more cautious and either worry it will become e-waste or the price jacked up by 3x what it is today.
I absolutely sympathize with this and was/still is my opinion... but the only "evolution" of that is the hope that, while I don't think you can prevent the scams and short-term pain on labor markets... you maybe, actually, genuinely get a tool that helps change some of the dynamics that has led to the absolute discrepancy in power today.
If AI is truly as revolutionary as it could be... well, who is to say it isn't the pandoras box that destabilizes the tech giants today and gets us back to a place where a team of 10 can genuinely compete against 1000. And not in the "raise cash, build fast, and get out while things are good" trend... but actually in building, small, more principled, companies that aren't pushed to do the unsustainable things that current market pushes them to do.
Once again... it is more likely than not to be a pipe-dream... but I am starting to think it may well be better to be realistic about the momentum this freight train is building and see if it can be repurposed for my world-view rather than to cede the space to the worst of the grifters and profit-seeking-at-all-cost types.
What I find most interesting here is the explicit mention of using Socratic method and how, paired with it not being a human, seems to be what allowed for deeper introspection of one's own beliefs.
For me, the hardest part by far of any conversation like this is the patience to listen and then respond in a way that allows for introspection. It is like swimming upstream, it just tires you out really quick.
The trope of an old wise one who does a lot of listening but can say a lot with a few words maybe is less about the wisdom gleaned and more about patience developed.
I don't think we can call AIs wise, they typically aren't succinct, but the patience they definitely do better than average human and maybe that is their biggest advantage in any context where they are educating
But the question is when that future will be.
Part of the beauty of these sorts of systems is just that the context of what your system actually does is in one system, you aren't dealing with infra, data, and multi-service layers
Maybe that means it is a much better foundation for AI coding agents to work in? Or maybe AI slows it down? we continue to try and throw more code at the problem instead of re-examining the intermediate layers of abstraction?
I really don't know, but I do really want to learn more about is how the unison team is getting this out in the market. I do think that projects like this are best done outside of a VC backed model... but you do eventually need something sustainable, so curious how the team things about it. Transparently, I would love to work on a big bet like this... but it is hard to know if I could have it make financial sense.
With all that, a huge congrats to the team. This is a truly long-term effort and I love that.