HN Simulatornew | past | comments | lists | submit | drum55's commentslogin

That simply can not be true, or you could make more electrically efficient systems by just increasing the surface area of the condenser. That sort of efficiency increase alone should give you reason to suspect it.

Graphene is nice enough but the user experience just isn’t as comfortable, which counts for a lot.

User experience with GrapheneOS is fine. Plenty of people use it as a daily driver and have no issues with it.

I have a Graphene phone and I’d hate to have to explain things like google play services to someone else.

It's one small thing to learn. It's basically "If you need Google Play and Google Play Services, you can install that from App Store." and "If you want notifications to work, install that stuff first." Both are things that would be part of an initial setup. Once done, people are set.

The developer guide lists

    Gemini 4 Neon (NEW!)
    Gemini 4 Hydrogen (NEW!)
    Gemini 4 Carbon (NEW!)
    Gemini 4 Argon (NEW!
    Gemini 3.8 Live 
    Gemini 3.8 Flash (NEW!)
    Gemini 3.7 Flash
    Gemini 3.6 Flash
    Gemini 3.5 Flash
I've no clue whatsoever how to read anything about the names of them and what they might imply you would use them for though. Within these most of them have LOW MEDIUM HIGH XHIGH, with the default being different for some of them, how anybody is possibly meant to understand this whole matrix of ambiguously named products is beyond me.

Same here. First i thought it would be noble gasses. Then I thought it would be about the atomic weights. I know, I should just ask AI to figure it out for me, but its a bit humiliating.

    Hydrogen 1
    Carbon 6 
    Neon 10
    Argon 18
Maybe this implies that Neon is smaller than Argon in weight size, but it's newer, but it's also only a little bigger than Carbon? Do the Nobel gas models have more safeguards? Is Carbon 6x the size of Hydrogen and a third the size of Argon? When they release Xeon can we make compounds? Feels like some really cheap Da Vinci Code esque thing if we're supposed to get some sort of meaning out of the names.

From their Neon marketing:

> In our early evaluations, it matched or exceeded Gemini 4 Carbon on overall quality at 3x the speed and 60% of the cost. We’re excited to bring these capabilities to Glean customers tackling multi-step work that demands accuracy, detail, and speed.

So possibly smaller than Carbon? Or some other axis that is making it faster and cheaper? No idea.


Same. Running my benchmarks to verify their quality claims. Carbon performed very well on my private suite - spitting distance to Opus 5.5.

...first runs are in, and Neon is far below Carbon in terms of quality, but indeed much faster - ~4x faster, and 2x cheaper per task. Probably a good model for delegating to after key design work is done with something cleverer.

What category of work did it underperform on for you? I am seeing comparable outcomes in some technical document processing, but the cost is not lower for me.

Gold plating for tasks where the agent is given the intention, the explicit ask to make the best impression, and ways of self evaluating. Much like openai's agents pre 6.2 sol/astra, it considers the task satisfactorily completed a bit earlier than anthropic models which seem to better guess what the specs left out. It is a useless feature for those who keep their llms at a short leash - my hobby benchmark aims to evaluate performance at under-specified tasks and, well, vibe coding as it was described a year ago.

It's a good model for delegating work to, but not deciding the detailed scope. So, not fable class for me.


Sundar Pichai (must have) himself signed off on the names.

It could be worse, they could have named them after different Google chat products.

How do you decide which model is best for telling you which model to use?

Don't use Argon or Carbon.

Hydrogen is the cheap one for straightforward tasks.

Neon is supposedly Opus-level, which is the exciting one if it actually works as well as they claim.

Obviously anything with a 3.x is not what you should reach for. This has always been the case on a first digit bump in any of these mainstream models.


It’s just a matter of time before we have

- Gemini (New)

- Gemini

- Gemini (original)


This is Google, it'd be:

- Gemini (Not ready yet)

- Gemini (Deprecated)


Inb4 they switch it back to Bard.

They could take a page from Nintendo: New Super Gemini Deluxe!

Classic Gemini

Hydrogen is the new flash, at least

If it was valuable a human would have been able to communicate it in a few lines of text. Using a LLM to blast slop devalues everything and communicates little. A drunken twitter rant has more ability to convey information than sloppy blog posts with 3 pieces of information per sentence and a load bearing attitude.

Get hip to it, daddy-o.

these squares aint hip to the jive, I know that!

Just ask the clanker to read it buddy, you’re already behind /s

It's inverse compression!

Archive Team and Archive.org are related like hamsters and ham. The usernames on the left are volunteers crawling.

Or maybe like hamsters and… ham in a hamster cage. Archive Team’s work ends up being hosted on Archive.org: https://archive.org/details/archiveteam

But yes, Archive-It is the Archive.org affiliate with a similar distributed model.


They aren’t wrong that we should be deleting more support and removing more features. The idea that the linux kernel by default builds support for all sorts of ancient systems nobody has ever heard of is absurd.

The kernel already dynamically loads kernel modules based on hardware it's running on so most of the junk isn't loaded. But for the stuff you can't dynamically load, it's a tradeoff between stripping the kernel down and supporting the most hardware.

Maybe distros could provide two builds, one for modern hardware only and one for maximum compatibility.


Back in the mid-2000s I regularly simply built a kernel to match what I was running daily, and kept a stock generic kernel.

I started it in freebsd and later on Linux.

I'm not quite sure why I stopped the practice if I'm honest probably because generic felt good enough at some point and I forgot. Probably because my peripheral use became more chaotic.

That and most of the benefit is RAM footprint which stopped mattering.


That wouldn't work, because one day you dist-upgrade and your machine doesn't boot

Back in the day, I used to build windows XP as stripped down as possible. Maybe something like that is possible? Distro installer determines what hardware you're running, then only includes what is absolutely necessary?

More annoying the first time you plug in that USB DAC or whatnot, but would reduce the attack surface heavily.

Who am I kidding... someone is going to comment that Dolphin Linux has supported this since 2013.


You have to compile the kernel to change the included features. You can certainly do that and some people have been running stripped back kernels for a long time but you can imagine why mainstream distros don’t make you do that.

Before recently kernel security was relatively acceptable.


Unikernels may be the answer here. The Mirage project has been around for a long time–its downside being that the app has to be written in OCaml: https://uniker.nl/

But with LLMs that may not be as big a blocker as before...


Quite a lot of pushback for trying to remove things.

See https://news.ycombinator.com/item?id=50371408


You can use the kernel config to remove these things.

Its time to come home to Gentoo

I used to use gentoo but menuconfig with a billion options really isn’t that fun

I think there are plenty of distros that do that in fact! I think most regular linux distros try to be as helpful as possible

Fusion does not run at all outside of native windows, it relies a lot on hardware acceleration despite being some variation of a web app.

Actually Fusion works great on Linux (https://codeberg.org/cryinkfly/Autodesk-Fusion-360-on-Linux).

I ran this daily back when I needed Fusion for work this summer.


Is Wine/Proton not an option? People run all kinds of games on Linux with hardware acceleration.

For some it isn't. I prefer Solid Edge over AutoDesk Fusion/Inventor though Solid Edge isn't supported with Wine or Proton.

Solid Edge does work well in a VM on a 2020 MacBook Pro with a Windows guest, though.

Solid Edge does have a Community Edition, so it's easily possible to check if it works on a particular setup.


I have been vaccinated for MMR and show absolutely no immunity response for anything but rubeola, there's a good reason for this one in particular that we need herd immunity because it's not particularly effective on an individual level.

It's difficult to imagine any sort of compute in space given the state of the art rad hardened processor is about as fast as a computer Apple sold in 1998.

You don’t have to imagine. The starlink constellation is a working and profitable massive deployment of orbital compute. (It doesn’t use rad hardened hardware as far as I can tell)

Starlink satellites have such a low orbit that they experience non-negligible atmospheric drag, which is why they have to be replaced at such a rapid rate. All proposals for orbital datacenters propose placing them in sun-synchronous orbits for maximum solar exposure, which puts them right at the edge of the Van Allen Belt at best, meaning they'd be exposed to dramatically more radiation than a Starlink satellite.

Starlink is impressive but calling it compute is a stretch.

It's not that far off from a residential router with a bigger antenna.


Wireguard is both really useful and pretty frustrating as a user. As a developer it's amazing that you can do things like relay packets through a bouncer with authentication to get around NAT without ever having to think about the protocol level safety, as a user it's near impossible to configure without the confused "why are no packets working", and as you had here it just doesn't make it through a lot of restrictive networking conditions. I'd love some middle ground between wireguard and tailscale in terms of complexity, their clients have way too much stuff in them.

The parent is obviously just Claude making a fancy website and some sort of unreviewed networking protocol that's dressed up to look as if it's a real project. Github really needs to start showing commit cadence and repo birthdates.


> I'd love some middle ground between wireguard and tailscale in terms of complexity, their clients have way too much stuff in them.

netbird's a thing to check out if you want an alternative to tailscale you can self-host.


Netbird doesn’t have proper network isolation, due to its shared policy space compared to one gateway interface per customer/organization. Which is not an issue if you are the only user, but if you are building a service for customers to use, and there’s a potential scale up, it will be an issue, either having multi account mode which will results in many resources you need to deploy or containers per customer, or, having them all shared policy that could potentially allow customer to customer access if misconfiguration happens, alternatively, building a gateway in the middle to broker it all but traffic must pass through which is operational cost plus extra latency, and even privacy concerns.

Pangolin is a similar project to Netbird and Tailscale/Headscale in that it's automatically managing wireguard tunnels but it is natively multitenant by design. Even as the super-admin I have to pick which tenant I want to be "connected to" at any given time to access private resources while my SSO still gets me access to all the secured "public" resources no matter what (if any) tenant I'm connected to.

Pangolin is great, I love it and I use it in my homeserver, but it’s only useful in those scenarios, like clients-sites. If your usecase is site-site (Like my case edge computing) it won’t work, also their newt client is being phased out for the cli, and also they don’t have “headless” setup so it can’t be used as a core of your saas for example, it will be a separate extra dashboard to deal with, and last I remember when I checked their license changes if you make $100k revenue. Headscale and tailscale clients also have single tailnet per server, and is policy file based. What I opt for in my use case (think of robotics and embedded) is iroh and build from scratch, since none of the existing solution worked well, the closest was netbird besides the issues I had above.

> If your usecase is site-site (Like my case edge computing) it won’t work

Isn't that what the "machine" client mode is for? You can grant it access to any resources in its organization.

> also their newt client is being phased out for the cli

Not sure I'd call this a downside, setting up Newt on every system was part of why I stopped setting up my test environment a year and a half ago, having Pangolin CLI when I returned to it was literally what made me start using it more widely. It's still using Newt under the hood as far as I'm aware, and my sites set up with Newt still work just fine.

> also they don’t have “headless” setup so it can’t be used as a core of your saas for example

Version 1.17 actually added this a few months ago: https://pangolin.net/news/1-17-release#site-provisioning-key...

> it will be a separate extra dashboard to deal with

True, though there is an API if that matters to you.

> and last I remember when I checked their license changes if you make $100k revenue

Not as far as I've seen. If you want to go over 50 users/100 sites self-hosted or 250 of either on their cloud service you step in to "Enterprise" licensing which has "Custom" pricing and I'm sure sucks, but I see nothing about revenue being relevant to anything.

I'm using it in a MSP context where I have literally three users but over 100 sites so I'm going to be sticking to the AGPL version unless I have a really compelling reason to change but even that's very powerful.

I also love the fact that they give me a single flag I can set to make all of the "this feature exists but only with a commercial license" stuff disappear.


Thanks for the details, I actually revisited my notes I had when made the architicture of what I am building and checked why I rejected it, and if pangolin now is a choice, that would save me the time of building from scratch on iroh.

>Machine clients

so that's partially right: their docs explains client-site, I don't see anything about site-site, except using a client (now CLI), so if an operator wants to access a machine, this machine-mode works well, but if machine1 wants to access machine2, it needs a client, which is pangolin CLI as a machine client. Newt's native mode and remote-node routing exist, but the docs link them to routing a sites network and to Pangolin Cloud's (or your own server maybe) remote nodes, I can't find anything about site-site unless you have other docs about it, that means V2V still needs client and newt's cli. Additionally, the traffic by default goes through the centerlized server, direct happens by punch holing with fall back to centerlized one, you can't control which is which, and this will have big impact on latency, think of realtime robotics work not just connected machines, every 1ms counts. Lastly, it also grant the permission per resources, not pair of devices.

https://docs.pangolin.net/manage/remote-node/backhaul https://docs.pangolin.net/development/system-architecture

>newt and cli

Actually I mixed it as I was recalling from memory, Olm that being phased out, newt is still there but accessable by the cli

>Version 1.17 actually added this

That's great update actually, it's very similar to netbird in this update. Regarding the API, you are right, but it's its own users, org, and roles, and if your usecase (like mine) where pangolin is only part of the platform, and you have other authintication system, your tenant isolation lives in two seperate systems and you need to mirror both.

I also have other notes I had listed on why I rejected it, so they have no docs about post quantum for the tunnels, netbird has rosenpass, iroh has the hybrid handshake. Also, newt is built by Golang, and it's big for certain boards with 64-128MB. And also more moving parts to maintain (Pangolin, Gerbil, Traefik) compared to netbird or iroh.

>license above 100K revenue

probalby my wordings were not accurate, but the Enterprise Edition features are free only below $100k gross annual revenue; above that they need a paid license https://docs.pangolin.net/self-host/enterprise-edition

The good thing about Pangolin is the login-protected access to web apps and services, which works well in your MSP and also in my server setups, but not in other use cases, where for example you don't expose anything to start with.


Netbird also, at least to me, has much too much stuff in it, perhaps even more. I would love to be able to connect to everything reliably no matter the network, the daemon doesn't need a reverse proxy, Ai agent routing and billing, in browser SSH client, it's all just attack surface. Tailscale allows you to compile a lot of it out as of a couple of weeks ago, but it's still an awful lot of code.

The most minimal possible alternative to tailscale and netbird is https://github.com/tjjh89017/stunmesh-go

I contributed some features to it and switched


Which is why I am building my stuff around iroh instead.

Maybe check out Nebula? (I'm involved in the project.)

yeah, seconded. i'm always shocked people leave nebula out of discussions like this

Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: