archevel's recent activity

  1. Comment on Why normal people aren’t using AI agents in ~tech

    archevel
    Link Parent
    That has been my general experience with diffusion models. In my cas with the porch I used a base photo and it fairly consistently was ble to add a fence to it in various styles. So I think this...

    That has been my general experience with diffusion models. In my cas with the porch I used a base photo and it fairly consistently was ble to add a fence to it in various styles. So I think this depends a bit on the model you use and probably on Gemini's surrounding tooling.

    3 votes

  2. Comment on Why normal people aren’t using AI agents in ~tech

    archevel
    Link Parent
    Tangentially related, as I actually found a nice use for stable diffusion models the other week. Generally, I think they've saturated the web with bland imagery. In short, I need to build a fence...

    Tangentially related, as I actually found a nice use for stable diffusion models the other week. Generally, I think they've saturated the web with bland imagery. In short, I need to build a fence for my porch. So I took a picture and asked Gemini to add a fence. I did this a few times with various descriptions of the fence to get an idea what it would look like. Some were really bad, e.g. making the fence where the current stair is thus blocking the entrance, but some were nice examples. I could show them to the rest of the family and we could pick a design that seem nice. Doing it by hand would be impossible for me at least and hiring someone to draw them (photorealistically) would be prohibitively expensive.

    Other than that in the case of image generation I'm at a loss.

    8 votes

  3. Comment on Browser fingerprinting demonstration in ~comp

    archevel
    Link Parent
    I wonder if I should just build a "be an iphone" extension for Firefox with a whitelist button to disable it if I absolutely need to. Better to hide as something very common rather than become...

    I wonder if I should just build a "be an iphone" extension for Firefox with a whitelist button to disable it if I absolutely need to. Better to hide as something very common rather than become fairly unique based on running various blockers etc.

    7 votes

  4. Comment on Browser fingerprinting demonstration in ~comp

    archevel
    Link
    No one likes tracking technology. No one likes registering an account at various sites. No one likes the forced age verification being foisted on various websites. What if we combine all these...

    No one likes tracking technology. No one likes registering an account at various sites. No one likes the forced age verification being foisted on various websites. What if we combine all these things? Something horrible or something potentially amazing?

    Imagine a site where your browser fingerprint along with your site interactions become your identity score for the things you post? It could be presented as the likelyhood of the author being the same as the author of other posts. Essentially you'd get a network of disparate posts with a degree of confidence that they were made by the same person/entity. You could weave in multiple fingerprinting techniques, various browser based JS checks, text analysis for style, interaction history (tracked client side and sever side with timings ought to be very unique).

    I imagine this probably wouldn't be immune to age verification laws, but an accountless service seems like a nice possible use of this tech. I imagine stopping spam messages might be the hard problem here though.

    2 votes

  5. Comment on Dark Sun and World of Warcraft in DnD in ~games.tabletop

    archevel
    Link
    So apparently Dark Sun is making a return as a mature rated product for DnD. I never played in the original setting, but it does seem to have some interesting twists on e.g. species like the...

    So apparently Dark Sun is making a return as a mature rated product for DnD. I never played in the original setting, but it does seem to have some interesting twists on e.g. species like the halflings. It is a rather twisted setting, so I get why they'd go with a mature rating on it.

    Also interesting to see them doing a WoW setting. WoW does have some fairly deep lore and pretty much everyone who plays DnD will at least be familiar with the world at least a little bit.

    My DnD group might do a one shot in these settings, but we do tend to get stuck on philosophical things in the world like debating if one of the PCs is still the same person after coming back from the dead due to wild magic + rebirth (he went from gnome to orc). RPGs truly make the best stories (no matter the system used).

    2 votes

  7. Comment on The magic is in the language in ~tech

    archevel
    Link Parent
    Just a couple of days ago I used to think that the models are basically "just" predicting the next token. I.e. if I feed it "Happy birthday to" it is highly likely to output "you" as the next...

    Just a couple of days ago I used to think that the models are basically "just" predicting the next token. I.e. if I feed it
    "Happy birthday to" it is highly likely to output "you" as the next token. However I saw a short YouTube clip that made a persuasive argument that models are infact doing at least something more advanced. If I feed an LLM:

    Add these numbers. Only output the answer.

3 
Five
One hundred
25
999
3 000 006

    It will likely output: 3 001 138. The argument is that the LLM hasn't learned this is the likely answer to the sequence of input tokens. If that was the case they'd often produce garbage answers for this type of input (older and/or smaller models do struggle with this). But, since newer models often seem to get it right, it is in my view now more likely that they have some form of concept of addition embedded in their weights. So they don't "just" predict the next token, they do seem capable of mental arithmetic.

    Perhaps there's some other explanation. Perhaps this breaks down on larger input sequences. I do however find it intriguing.

    2 votes

  8. Comment on What programming/technical projects have you been working on? in ~comp

    archevel
    Link
    My first contribution to hermit os was just merged! I've improved the vsock support which is something I'm using for my security oriented webbapp hosting project. I still have some additional...

    My first contribution to hermit os was just merged! I've improved the vsock support which is something I'm using for my security oriented webbapp hosting project. I still have some additional patches that fixes issues with running hermit os based applications on firecracker; kvmclock support and some other stuff.

    Next focus is to figure out a solid path for postgres replication to backup nodes. I have backups based on the WAL working (all encrypted outside the guest vm), but I think a replication solution and filesystem base backups might be nicer. Needs more work either way.

    1 vote

  9. Comment on A friend showed me another "learned a new language" site in ~comp

    archevel
    Link Parent
    Many years back at university we had a semester where we learned programming in Erlang/OTP. To this day I think Erlang and the BEAM has the best ergonomics w.r.t. its actor system. The functional...

    Many years back at university we had a semester where we learned programming in Erlang/OTP. To this day I think Erlang and the BEAM has the best ergonomics w.r.t. its actor system. The functional nature of Erlang with the pattern matching just meshed well my mind. Writing recursive functions starting with a base case and then adding the other cases was great! I think Elixir borrows a bit of this due to it running on top of the BEAM (but I never liked Ruby for some reason so never got into Elixir). Anyway, main point if you want to get a solid grasp of actor systems I'd recommend actually working with Erlang/OTP. Setting up a supervision tree and using gen_server and gen_fsm based actors is great. Understanding how this works under the hood by building a smaller version of ORP; creating processes and linking them and trapping their exit messages; could be a good exercise. The language's constructs generally makes it hard to mess up completely and the "let it fail" philosophy let's you avoid defensive programming and focus on more important things.

    1 vote

  10. Comment on I wonder if years from now hand written code will be antique in ~tech

    archevel
    Link Parent
    Not entirely. My goal when making software haven't been overly focused on the craft. I focus on solving problems. The analogy with the craftsman never sat entirely right with me. A furniture...

    Not entirely. My goal when making software haven't been overly focused on the craft. I focus on solving problems. The analogy with the craftsman never sat entirely right with me. A furniture carpenter finishes a table and then that table is done. For a software developer the work is continual. Maybe more like a gardener where the garden is in a continuous flux. So a gardener that goes from tending the garden to sitting in a tower far away with binoculars and a megaphone screaming at the robots to "prune the trees and make no mistakes"... I don't know. Not great at analogies.

    2 votes

  11. Comment on I wonder if years from now hand written code will be antique in ~tech

    archevel
    Link Parent
    I enjoyed coding and solving issues with code. I've been doing this professionally for about 20 years soon. With an LLM I am certainly able to churn out more, both in terms of lines of code and...

    I enjoyed coding and solving issues with code. I've been doing this professionally for about 20 years soon. With an LLM I am certainly able to churn out more, both in terms of lines of code and actual features. I've mostly settled into validating the output of the code; not the code itself. Design and QA is taking a larger portion of my time at work. Downside is that, while it is still satisfying to ship features and improve the product, it isn't as enjoyable a craft anymore. I no longer have a deep understanding of the underlying code.

    I know that, in general, I am still able to produce code at a higher level of quality than an LLM, but that doesn't matter that much to me. Something ineffable is missing now. I tend to think this is similar to how artist feel about gen ai in general. They can see the flaws produced by e.g stable diffusion, but does it diminish the sense of creating? Is the process tarnished in some way? I think it is for me.

    3 votes

  12. Comment on What programming/technical projects have you been working on? in ~comp

    archevel
    Link
    I've started to build a secure hosting platform. The goal is to provide an environment that helps devs build secure systems by locking down what's actually accessible and make foot guns hard....

    I've started to build a secure hosting platform. The goal is to provide an environment that helps devs build secure systems by locking down what's actually accessible and make foot guns hard. Currently it provides a postgres instance running in a vm without a shell and without network access, a way to host apps that handle https requests, and apps for background processing. The postgres instance is running on an encrypted filesystem decrypted with an ephemeral key that's only held in memory (derived from a master key + tenant secret held in a control plane). Daily backups are also encrypted so nothing is ever stored as clear text. Additionally, there's also a key rotation system inplace in case a key or a disk becomes compromised.

    On top of the db sits an arbitrary number of the tenants' apps. These can talk to postgres over a vsock channel relayed by the host. The apps are all based on hermit os which boot in microseconds when a request comes in. The host doesn't decrypt the traffic, instead it peeks at the incomming encrypted tcp stream to determine the sni and use that to route the request to the correct app. The guest app hold the certificate and decrypts the request (not the host). In addition the apps, like the postgres instance, has no network access by default. It is possible to green light specific domains for an app if it needs some external access. But, even this is fairly locked down. All outbound connections are mediated by the host which blocks connection attempts to non green listed domains, but crucially the host only sees the domain the guest tries to connect to. Apart from that it just relays the bytes (likely encrypted) to the IP resolved by the domain name.

    Why would anyone want this? Well, the limited capabilities severely reduces the attack surface. Even if someone managed to get arbitrary code execution in one of the client apps (unlikely unless you royally screw up) they are still running under a hypervisor and would need to perform some vm escape in order to be able to set up a permanent foothold. Since the apps are unikernels there is no shell to escape to, there is no disk drive to access and no network devices attached to the guests.

    The host manages the tenants' VMs and suspends the DB instance when nothing is running against it (after a ~5min delay). When a request comes in it is handed off to a running app instance. If there isn't one the VM is booted. At most, on my local laptop, this takes a second or so if the tenant is completely cold with the DB never having been started. If the DB is running, but the app is not, it takes a few microseconds to boot. Understandably one of the aims is to keep resource utilization and startup times to a minimum.

    I'm preparing to open source the setup and possibly run it as a service here in Sweden. Sovereign hosting solutions out of reach of American interests seem to be talked about more and more. I need to sort out a few final issues; PTR for postgres is needed essentially to handle graceful failover and tenant load balancing. I've sketched out a way to handle the PTR, but I'm still weighing the options for how to do proper tenant assignment (and failover) to physically disparate host nodes.

    3 votes

  13. Comment on No WebGL, no AI, no bytes to spare. My new game is a 3D 90s fever dream in 1024 bytes! in ~comp

    archevel
    Link Parent
    Neat! Way to hard for me :)

    Neat! Way to hard for me :)

    1 vote

  14. Comment on Burnout, A(u)DHD, and what next in my life & career? in ~life

    archevel
    Link Parent
    I don't know how it works in the US, but I would guess it is similar to here in Sweden. The main benefit of an LLC is really in the name. Limited liability. I.e. typically me personally is not on...

    I don't know how it works in the US, but I would guess it is similar to here in Sweden. The main benefit of an LLC is really in the name. Limited liability. I.e. typically me personally is not on the hook for e.g. debt that the business has (I think there might be a few exceptions here). If I was just self employed my personal finances (eg. my house) would be on the line if I have a debt of some kind. I see it mainly as a way to separate my business's financials from my own. Apart from that, here there are tax benefits when paying out dividends compared to paying out regular salary, but this might vary a lot depending on where you are!

    2 votes

  15. Comment on Burnout, A(u)DHD, and what next in my life & career? in ~life

    archevel
    Link
    In 2023(?) I startedy own consulting company as a software developer. I had been employed as a dev full time since about 2008 for various companies and wanted to try something new. By contracting...

    In 2023(?) I startedy own consulting company as a software developer. I had been employed as a dev full time since about 2008 for various companies and wanted to try something new. By contracting through a kind of broker I got a contract with a bank at a quite good hourly rate filling a role as a lead dev for a small team. It was nothing exciting tech wise but I enjoyed it. The main benefit looking back at it was the sense of freedom. Even though I was essentially working like a full-time employee it felt like I had more freedom. After a few months I had managed to generate enough income that even if I walked away I could keep paying out my salary for a few additional months. To me, building that capital reserve in the company was great. Not all businesses will have that kind of margin and it isn't something you get actually rich doing (unless you start hireing, which bra ngs it's own headaches), but if you manage to land enough contracts and generate some additional income it can be very fulfilling. You end up working for yourself... Downside is that you can always work more. It is easy to see that if you work more hours in this setup you make more money. So taking a step back can be a challenge. An alternate way would be to charge a larger fixed fee for a delivery, but that has other risks associated with it.

    What I would recommend is to:
    a) see if you can try it out in a limited fashion while still employed. That way you can fall back on your employment if things don't necessarily work out.
    b) if things seem to fly, incorporate. Not sure how things work where you are. But, here it is much more beneficial to run a business as a limited liability company than to "just" be self employed. YMMV. This will probably involve having to do a bunch of extra work (eg. accounting, salary payments, taxes etc). For me that was worth it though.

    Good luck whichever way you choose to go. There is no wrong way to live a life :)

    5 votes

  16. Comment on Sweden may oppose Tesla's supervised self-driving tech in Europe over speeding concerns in ~transport

    archevel
    Link Parent
    Who's behavior should it encode? Even looking at the messages here there isn't a singular behavior that everyone would agree on. Having the vehicle follow the local laws seem like the least...

    Computer driving systems should encode human behavior first and foremost.

    Who's behavior should it encode? Even looking at the messages here there isn't a singular behavior that everyone would agree on. Having the vehicle follow the local laws seem like the least arbitrary and most easily predictable behavior... Besides, you know, it is the law.

    6 votes

  17. Comment on Sweden may oppose Tesla's supervised self-driving tech in Europe over speeding concerns in ~transport

    archevel
    Link Parent
    I mostly tend to speed if I am passing someone. It's more of a judgement call, but strictly speaking this is still speeding.

    I mostly tend to speed if I am passing someone. It's more of a judgement call, but strictly speaking this is still speeding.

    3 votes

  18. Comment on Sweden may oppose Tesla's supervised self-driving tech in Europe over speeding concerns in ~transport

    archevel
    Link
    To me it seems a bit strange to have "break the law" as a feature, but maybe that's just me. I can see there might be a point in some emergencies, e.g. someone is bleeding out in the car. So MAYBE...

    To me it seems a bit strange to have "break the law" as a feature, but maybe that's just me. I can see there might be a point in some emergencies, e.g. someone is bleeding out in the car. So MAYBE have an emergency mode that also alerts the authorities? A general go 10kph above speed limit seems bad (even if that's what the general populace does).

    13 votes

  20. Comment on What programming/technical projects have you been working on? in ~comp

    archevel
    Link Parent
    Got me thinking... If you express the shape of a slime as a function that operates on a line then you get an arbitrary wiggly line. Then you turn that line into a circle by connecting the ends. If...

    Got me thinking... If you express the shape of a slime as a function that operates on a line then you get an arbitrary wiggly line. Then you turn that line into a circle by connecting the ends. If the line ever crosses itself when it is bent then you could treat that as the slime splitting at that point into two slimes. Two slimes colliding at a point would merge the slimes functions in some way. Slimes could naturally then grow by combining the length of their circumference line. Anyway, fun project!

    1 vote