archevel's recent activity
-
Comment on What programming/technical projects have you been working on? in ~comp
-
Comment on Any beginners advice or resources on developing a 2D RPG/Puzzle video game? in ~comp
archevel LinkI enjoyed the GameDev Tavern's video tutorials series on building two fairly straightforward, but different games. Each part of the series builds on the previous in a natural progression. I think...I enjoyed the GameDev Tavern's video tutorials series on building two fairly straightforward, but different games. Each part of the series builds on the previous in a natural progression. I think they made a bunch of the resources available, but I used other ones just to get a better grip on the process. It's too easy for me to otherwise just end up in copy paste land (and then I don't really learn).
-
Comment on Sweden's old‑growth natural forests store 83% more carbon than managed woodlands – new study in ~enviro
archevel LinkHopefully this can change if we manage to vote out the current right wing coalition currently leading the country come september (it is an election year here). I don't anticipate that forestry...Hopefully this can change if we manage to vote out the current right wing coalition currently leading the country come september (it is an election year here). I don't anticipate that forestry policy will be a major political issue, but I expect that with a left wing coalition the environmentalist vocies will not fall on deaf ears to the same extent. There are a bunch of challenges aside from "just" global warming, e.g. water and pollution management, exploitation of shorelines and seabeds, fishing quotas, hunting permits/quotas for wild wolves and lynx... Lots of small issues where there are different interests that all have legitimate concerns or at least opposing (short term?) interests.
-
Comment on The 49MB web page in ~comp
archevel Link ParentYouTube in Firefox on mobile with unlock origin has thus far been enough to block ads on YouTube.probably some features that the app has tha I can't use, but for just watching it works flawlessly...YouTube in Firefox on mobile with unlock origin has thus far been enough to block ads on YouTube.probably some features that the app has tha I can't use, but for just watching it works flawlessly for me.
-
Comment on Factory farming is a blight in ~enviro
archevel Link ParentI'd have to disagree. Relying on individuals moral choices to ensure animal health seems fragile at best. It's also of course not an either or situation. People can go vegan/vegetarian/partially...I'd have to disagree. Relying on individuals moral choices to ensure animal health seems fragile at best. It's also of course not an either or situation. People can go vegan/vegetarian/partially so and at the same time we shouldn't give up the on affecting government. This doesn't necessarily have to be through federal law. I imagine local and state government could have an impact on animal welfare no?
-
Comment on Queering the Map in ~lgbt
archevel Link ParentFor me this is now the best map. Touching short messages everywhere!For me this is now the best map. Touching short messages everywhere!
-
Comment on Executing programs inside transformers with exponentially faster inference in ~comp
archevel Link ParentIt will probably be more efficient to generate the code and compile and execute it (or interpret it). Rather than running it in the LLM, but I've no hard evidence for that. I mentioned in the...It will probably be more efficient to generate the code and compile and execute it (or interpret it). Rather than running it in the LLM, but I've no hard evidence for that. I mentioned in the weekly tech project thread that I've been experimenting with Recursive Language Models. Essentially it is instructed to generate python code that gets executed in a REPL (statefull) that mediated the communication with the user and the model can decide on next steps based on the REPL output. The approach in the article seems to skip all observability, but I might be wrong.
-
Comment on “Reinforces the value of originality”: Fender secures legal ruling to protect the Stratocaster body design in ~hobbies
archevel LinkI wonder how/if this is applicable to 3d models similar to the Stratocaster. Is it fine to give away own CAD version? Sell it? Can I sell prints of it?I wonder how/if this is applicable to 3d models similar to the Stratocaster. Is it fine to give away own CAD version? Sell it? Can I sell prints of it?
-
Comment on What programming/technical projects have you been working on? in ~comp
archevel (edited )Link ParentMinor update. I hooked it up to gpt-4.1 and it was really good. I had a corpus of about 20M characters so significantly larger than the context window. Asked it to answer a few questions about the...Minor update. I hooked it up to gpt-4.1 and it was really good. I had a corpus of about 20M characters so significantly larger than the context window. Asked it to answer a few questions about the text and it generating code to see which snippets were relevant, asked "sub LLMs" to analyze parts of the text and finally spit out part of what it needed. Then I could ask it to point me to it's sources and since those are basically the variables it used it was able to check for those and see what it referenced.
For the curious: https://github.com/archevel/repllm
-
Comment on What programming/technical projects have you been working on? in ~comp
archevel LinkI ran into this paper on Recursive Language Models and I liked the concept. It essentially puts the LLM in a REPL. I spent some time setting up my own version of it with a local ollama setup. The...I ran into this paper on Recursive Language Models and I liked the concept. It essentially puts the LLM in a REPL.
I spent some time setting up my own version of it with a local ollama setup. The models I am able to run are fairly weak, so it wasn't very good. Lots of issues with it answering from its training data rather than actually interacting with the corpus. The most success I got was putting the entire paper into the context and then asking it what the paper was about. I'll do some more tests by pointing it to eg gpt4.1 or something. I'll probably need to use a bigger corpus for it to be sensible. Neat approach though!
-
Comment on The first multi-behavior brain upload in ~science
archevel Link ParentI agree. I'd argue that most people are fine with animal cruelty (just looking at the meat industry). Most people would likely object to treating humans that way. This of course doesn't make it...I agree. I'd argue that most people are fine with animal cruelty (just looking at the meat industry). Most people would likely object to treating humans that way. This of course doesn't make it ethically correct to experiment on the consciousness of parrots, just that it is likely that common sensibilities would be upset if this was being done to human minds.
-
Comment on The first multi-behavior brain upload in ~science
archevel Link ParentConsciousness is likely a spectrum, i.e. a rock is less conscious than an amoeba which is less concussion than a fruit fly which (hopefully) is less concussion than @tearearlgraycold ;) To me this...Consciousness is likely a spectrum, i.e. a rock is less conscious than an amoeba which is less concussion than a fruit fly which (hopefully) is less concussion than @tearearlgraycold ;)
To me this is all both super creepy and utterly fascinating. If this ever develops to encompass human minds it's riddled with ethical pitfalls. Can't ever really turn it off. Is it ethical to speed up the simulation? Is it ethical to slow it down? Might not even be a factor...
-
Comment on What would you do with a video game style inventory? in ~talk
archevel LinkI am assuming the objects in the inventory have no mass i.e. they don't hinder my movement what so ever. If so I'd become a perpetual motion machine in some form. E.g. put a lot of heavy things in...I am assuming the objects in the inventory have no mass i.e. they don't hinder my movement what so ever. If so I'd become a perpetual motion machine in some form. E.g. put a lot of heavy things in my inventory, raise myself high up in an elevator connected to a generator, take out heavy things and, generate power on the decent. Rinse repeat. Infinite power. A bit boring, and there's probably a bunch more efficient ways, but you get the idea. Energy crisis averted!
-
Comment on Is higher education still valuable? in ~life
archevel Link ParentYeah at least geriatrics is going to be a major field with an aging population. Not very glamorous, but likely very stable.Yeah at least geriatrics is going to be a major field with an aging population. Not very glamorous, but likely very stable.
-
Comment on Looking for vibe-coding guides (best practices, etc.) in ~tech
archevel LinkIf you want to get something running quickly an with minimal effort lovable is nice. You can get something similar (and likely superior in some ways) by using claude and hooking it up to supabase...If you want to get something running quickly an with minimal effort lovable is nice. You can get something similar (and likely superior in some ways) by using claude and hooking it up to supabase and chrome directly. The deep integration makes it easier for the agent to explore issues and test stuff out to validate fixes. Still for larger projects it tends to mess things up...
-
Comment on Is higher education still valuable? in ~life
archevel LinkAll of this is pure speculation and conjecture so take it with a large grain of salt. Is paying for a university degree going to be worth it? There are multiple variables here. There is an...All of this is pure speculation and conjecture so take it with a large grain of salt. Is paying for a university degree going to be worth it? There are multiple variables here. There is an opportunity cost (what else could you be doing with your time). There's the direct cost of paying for the classes. There is the uncertainty around if the degree will still open doors otherwise shut to you.
In my opinion getting a degree is a way to check a box when applying for jobs. Actual work usually requires skills not taught at uni. So the question is if employers in the future will keep requiring degrees from the people they recruit OR if they will filter people by some other means. More importantly might be the question if there will be a job market at all in your chosen profession? Now, the ai hype is in full swing and it's easy to get swept up in doubt about future prospects. Perhaps that is the case. Perhaps there won't be a need for statisticians to the same degree. Will having an understanding of statistics be beneficial for whatever you end up working with? I do believe so. It may not be as profitable as it once was since lots of white-collar jobs may become semi-automated. But, the reasoning that you learn will be still be good.
If I were advicing young adults that mainly want to find a stable job, then I'd advice them to go into healthcare or some kind of skilled labour. If they want to do intellectual work then getting an education is still basically mandatory and will continue to be for the foreseeable future.
-
Comment on This video is six minutes long! in ~tech
archevel Link ParentOn mobile (Firefox) I had to switch to desktop mode to get the option to show up.On mobile (Firefox) I had to switch to desktop mode to get the option to show up.
-
Comment on The hidden cost of AI art: Brandon Sanderson's keynote in ~tech
archevel Link ParentI think there is a distinction to be made here about what it is you care about. As a company (and sometimes as a person) you rarely care about the process of producing something. You want it fast...I think there is a distinction to be made here about what it is you care about. As a company (and sometimes as a person) you rarely care about the process of producing something. You want it fast and as cheap as possible. If this is the case using an LLM to build something is likely the pragmatic approach. You can get progress sooo much faster. Which then let's you validate the idea faster. Rinse, repeat.
However, from a learning perspective, using an LLM to build software seems a bit shortsighted (similar argument can be made for initially using an IDE when learning a new programing language). That is, unless what you are trying to learn is building software with the aid of an LLM (which I do think is a valuable skill, and this is likely part of what you are learning). But, assuming it isn't, then why not just tackle building a larger project? You'll run into issues for sure and you will not progress as fast to a "finished product", but if learning was the goal then the product is incidental.
-
Comment on Supporting Markdown search for LLMs in ~tech
archevel Link ParentI think most html is not good for providing meaningful semantic content anyway. XHTML tried to do this, but failed in my view. For most text rich websites, e.g. Wikipedia, I think could...I think most html is not good for providing meaningful semantic content anyway. XHTML tried to do this, but failed in my view. For most text rich websites, e.g. Wikipedia, I think could meaningfully be converted to markdown without loosing anything of importance. Might be wrong, but I can't think of a counter example.
-
Comment on AntiRender: remove the glossy shine on architectural renderings in ~tech
archevel Link ParentYeah, I wanted to try it with an image of my face and got some edge function failure. Seems like a fairly common loveable issue.Yeah, I wanted to try it with an image of my face and got some edge function failure. Seems like a fairly common loveable issue.
I have been picking up my old lisp interpeter I started back in 2016 (I think). It already had support for proper tail call optimization and was fairly lean and easy to grasp (for me). I got stuck on two major features back then. One; hygienic macro support. Two; a tool for wrapping arbitrary golang code to make it callable from the lisp code.
So being a modern developer I pointed Claude to the repo and pointed it to the racket documentation about hygienic macros and instructed it to work in a TDD like manner. It is amazing what it can spit out, but it tends to be like most developers and omitt edge cases so I keep insisting it to check the code coverage to make sure the test coverage doesn't degrade (it sits well over 90%). Apart from the new macro support (almost finished, just need to add support for
syntax-case) and the tool for wrapping golang libraries it has also helped me add a require/module system. Lots of fun, and a bit addictive.Next step, once the macro support is finished, will probably be to rewrite the parser. It currently yacc based, but I'd like to create one using the new SIMD capabilities that is experimental at the moment.