archevel's recent activity
-
Comment on Yarn, React, and Udemy. Help requested. in ~comp
-
Comment on Let's build a playlist! in ~music
archevel LinkSuper fun! I picked an eclectic mix: Sage Francis - Make Em Purr Adagio for Strings Niccokick - I want you back Sahara Hotnights - Cheek to Cheek Visa från UtanmyraSuper fun! I picked an eclectic mix:
-
Comment on Can I have some advice on the neural net I've been working on? in ~comp
archevel Link ParentGood luck! I've done this a couple of times now (in Java and octave/mathlab). It is a great way to get an intuition about neural networks for sure.Good luck! I've done this a couple of times now (in Java and octave/mathlab). It is a great way to get an intuition about neural networks for sure.
-
Comment on Can I have some advice on the neural net I've been working on? in ~comp
archevel LinkI had a quick peek at the code and didn't spot any obvious errors. Given a 10 minute review is probably not enough. Some general advice to narrow down the problem: Use a simpler problem to verify...I had a quick peek at the code and didn't spot any obvious errors. Given a 10 minute review is probably not enough. Some general advice to narrow down the problem:
- Use a simpler problem to verify your training algorithm. E.g. train an xor network or a simple categorization network.
- Remove the parallelism. You likely could do this without a lot of effort by reducing the number of parallel batches to 1.
Finally, consider using a preexisting .net library for training NNs or consider using a non .net library like pytorch, keras or some such more widely used training tool and find a compatible .net library that can load the weights after training. Writing your own backprop is only really useful to understand the algorithm, so of that's your goal then go at it! If you goal is to use a NN to solve your actual problem, then it's likely a waste to implement the training algo (unless you enjoy it of course).
-
Comment on Tildes Game Giveaway: June/July 2024 in ~games
archevel LinkNot really a gift, but I recently discovered Moonring on steam. It is an homage to late 80s cRPGs like Ultima, but with a lot of QoL improvements. It is also available for free which is nice so if...Not really a gift, but I recently discovered Moonring on steam. It is an homage to late 80s cRPGs like Ultima, but with a lot of QoL improvements. It is also available for free which is nice so if that type of game is your jam give it a spin!
-
Comment on Microsoft CEO of AI claims online content is 'freeware' [and can be used to train LLMs in the absence of a specific directives from the author against this] in ~tech
archevel Link ParentJust because something can regurgitate a text if properly prompted doesn't mean the thing by existing is violating copyright. Learning that a particular pattern is common just influences the...Just because something can regurgitate a text if properly prompted doesn't mean the thing by existing is violating copyright. Learning that a particular pattern is common just influences the weights of the likely next word, but actually getting the word for word copy of a text does depend on a few of other parameters too. Also I am not arguing whether or not an LLM or it's output violates copyright law as is. I am arguing that it shouldn't be considered to violate it, i.e. if things were how I want them to be.
-
Comment on Microsoft CEO of AI claims online content is 'freeware' [and can be used to train LLMs in the absence of a specific directives from the author against this] in ~tech
-
Comment on Microsoft CEO of AI claims online content is 'freeware' [and can be used to train LLMs in the absence of a specific directives from the author against this] in ~tech
archevel LinkI dislike Microsoft in general and I think their CEO of AI is wrong w.r.t. the content being "fair use" because it is online. That AFAICT has nothing to do with fair use. However, should they be...I dislike Microsoft in general and I think their CEO of AI is wrong w.r.t. the content being "fair use" because it is online. That AFAICT has nothing to do with fair use.
However, should they be allowed to use online content for training? In isolation, I think, the answer ought to be yes. The argument for this is that training an AI is essentially IIUC counting the occurances of combination of words and build a model of how likely other words are given the previous ones. Then when we use an LLM we essentially roll a dice and use that to pick the word sequence to follow a given input. So should the result of counting word combinations be considered a derivative work of the text examined? I don't think it is obvious either way.
Complicating things further if I use a prompt like:
Write the first chapter of the sequel to [book title] by author [author name]
I think the output should be considered derivative if I use "A Game of Thrones" and "George R.R. Martin" even if his exact works were not part of the training set. If I'd instead used "A Play of Benches" and "Gabriel B.N. Sinclair" (both made up) then no matter if Martin's work is part of the training set I wouldn't consider the output derivative of his particular work.
The real complication comes when we start to consider the effect of this stance. Large entities would be perfectly free to train LLMs on any corpus (you would too, but you can't because you don't have access to enough compute power). Creative work will likely be further devalued. The quality of content will likely regress to the mean (especially if corporations start training on texts generated by AI).
For me the preferred solution (however unlikely) would be to abolish property rights (both physical and intellectual). If people want to generate AI content, they should be allowed to (it does little harm). A handwritten message from someone will always be more valuable than something mass-produced, be it by AI or an AD-agency/copywriter.
-
Comment on Tildes Minecraft Survival - Returning blockier than ever on Friday July 12th (1.21 update) in ~games
archevel LinkExcellent initiative! I was never on the previous server, but now I'd like to be added to the whitelist if possible. It was quite a while since I last played so I figure there's lots of new stuff...Excellent initiative! I was never on the previous server, but now I'd like to be added to the whitelist if possible. It was quite a while since I last played so I figure there's lots of new stuff I haven't tried. I briefly joined a model server, but there was just too many systems in place for my tester. Vanilla is wat more accessible to me (though I did like the ulti-mine? mod for the convenience of clearing out areas).
-
Comment on What programming/technical projects have you been working on? in ~comp
archevel Link ParentCool! Not at all familiar with the area, but searched for "five way junction" and got this. Some of the boxes seem right, but others look like regular three way junction. Nifty either way!Cool! Not at all familiar with the area, but searched for "five way junction" and got this. Some of the boxes seem right, but others look like regular three way junction.
Nifty either way!
-
Comment on How generative AI could reinvent what it means to play in ~games
archevel Link ParentI've been toying with the idea of creating a GM AI hooked up to a voice chat since chatgpt 3. The main issue in my very limited trials was finding a prompt that didn't cause it to direct the...I've been toying with the idea of creating a GM AI hooked up to a voice chat since chatgpt 3. The main issue in my very limited trials was finding a prompt that didn't cause it to direct the actions of the players too much. I.e. it usually went
Me: I try to open the door.
GM: You succeed at opening the door. You walk in. There is a troll there which you engage in combat. Finally it flees down a dank tunnel....And it would go on describing the outcomes and essentially rob the player of agency. This part is a bit of a balancing act and I believe with time and sufficient promot engineering a better more dynamic output could be achieved. It seems however that current LLMs are bad at short answers preferring longer expositions (might just be my prompts).
The second part is something i call the "yes and" disease. LLMs are too accommodating. If I tell it I open a portal to the realm of ferries. It won't protest even if the player character has no magical abilities or portal in sight. Nor if there is no realm of ferries in the setting. I'd want it to be able to deny such requests
Ideally it would even be able to handle false information in the setting, I.e. there is a widespread belief in the setting about dragons that is actually false. This however leads to the main problem of LLMs: Consistency.Going through a door and then going back will likely lead to completely different descriptions of the room. Those could be awesome in a magical labyrinth, but in regular circumstances it doesn't fly. I think this could be solved by using an assembly of specialised LLMs in a chat; a chronicler that mostly observes and just outputs facts about the world as has been previously stated ny others, a rules laywer that interjects when different tests are required, faction and major npc representatives who would output potential actions these undertak, and finally a mouthpiece that is responsible for narrating the actual play and drives the dialogue with the players. There could of course be more of these specialised agents to handle some element of the game...
-
Comment on What programming/technical projects have you been working on? in ~comp
archevel LinkI started looking at migrating an web app written in asp.net with ef core to use postgres instead of sql server. I quickly ran into a weird issue that updates failed (expected 1 row affected but 0...I started looking at migrating an web app written in asp.net with ef core to use postgres instead of sql server. I quickly ran into a weird issue that updates failed (expected 1 row affected but 0 rows affected). Still haven't figured out why. I thought ef core would be largely agnostic when it came to this type of basi operation! The updates essentially maps a domain object to an entity, then calls something like:
var entry = _context.Entry(entity); if(entry != null) { entry.CurrentValues.SetValues(entity); entry.State = EntityState.Modified; } await _context.SaveChangesAsync();It's weird to me that the updates fail now, but worked when the db was sqlserver... have to look over the migrations to see if there's something weird there that is causing issues... inserts seem to work 😕 🤷
-
Comment on What are you reading these days? in ~books
archevel Link ParentOh, interesting! Didn't know there was a book club going. Will have to read the others impressions and be on the lookout for the next book!Oh, interesting! Didn't know there was a book club going. Will have to read the others impressions and be on the lookout for the next book!
-
Comment on Science fiction or fantasy recommendations for children in ~books
archevel LinkSome of Sandersons books might be nice: Tress and the Emerald Sea Mistborn series (might be too young) My oldest daughter (13 now) got into the Warriors series of books which is essentially about...Some of Sandersons books might be nice:
- Tress and the Emerald Sea
- Mistborn series (might be too young)
My oldest daughter (13 now) got into the Warriors series of books which is essentially about wild cats (there are a lot of books in the series)!
-
Comment on What are you reading these days? in ~books
archevel LinkFinished listening to The Dispossed by Ursula K. Le Guin while painting our house. It is a lot more slow paced and pensive than my usual books, but I would recommend it! It made me think about a...Finished listening to The Dispossed by Ursula K. Le Guin while painting our house. It is a lot more slow paced and pensive than my usual books, but I would recommend it!
It made me think about a few things. How language shapes our morals and ethics by vurtue of how we express concepts. Secondly, why we strive for material wealth.
The anarchist society described in the book seems like a anarcho-syndicalist ideal. While the society has problems it is depicted as fairly benign and actually quite inspiring! There are a few mysteries that are hinted at, i.e. the Hanish's role. It's fun that Earth is part of the story, but only really aluded to in the background, with a few exceptions.
Recommended it to my 15 year old. Super interested in their opinion of the book... it might be too slow for their tastes though.
-
Comment on Your favorite deeply unpopular music in ~music
archevel LinkI recently discovered The Tiger Lillies - Banging in the Nails. Not sure if they count as deeply unpopular since in my head they are too obscure.I recently discovered The Tiger Lillies - Banging in the Nails. Not sure if they count as deeply unpopular since in my head they are too obscure.
-
Comment on What games have you been playing, and what's your opinion on them? in ~games
archevel LinkV Rising Player it a bunch when it was in early access. Now that it has been officially released I'm giving it another go. I remember it as more difficult, but now it is a bit quicker to get...V Rising
Player it a bunch when it was in early access. Now that it has been officially released I'm giving it another go. I remember it as more difficult, but now it is a bit quicker to get through the different bosses. I like the introduction of trader camps, nice when missing some particular resource.
My main gripe isn't with the actual game, but with the lack of a Linux based server option. I'll look into setting it up with wine in a docker container, but it's a bummer they've made it more difficult to host a server. Seems like a nobrainer to compile a Linux option for the server part at least... should try going to their office to complain ;) or just ask them for the source and build it myself! I did try for a few minutes to decompile the binary and figured I could rebuild it, but couldn't get it to decompile with dotPeek unfortunately.
-
Comment on AI: The decade ahead in ~tech
archevel Link ParentUltimately this is true. In an unequal economic system however education becomes a means to an end. A way of improving the odds of increasing your material wellbeing. A way to gatekeep and filter...Education is its own reward.
Ultimately this is true. In an unequal economic system however education becomes a means to an end. A way of improving the odds of increasing your material wellbeing. A way to gatekeep and filter people. If we could get away from this I think we'd be more fulfilled in general. Then education would be a its own reward. Today it can be... if you are privileged enough.
I like to think that also work, in and of itself, is its own reward. That tend to be a bit more controversial, but a job well done is satisfying no matter if there are economic incentives in place.
-
Comment on How the US is destroying young people’s future | Scott Galloway in ~life
archevel (edited )Link ParentI tend to agree. I believe it is much more important for a functioning democracy that people are taking part in formulating the choices that are being consideredrather than the actual choosing...I tend to agree. I believe it is much more important for a functioning democracy that people are taking part in formulating the choices that are being consideredrather than the actual choosing between options. There is much more power in deciding what the question on a ballot is rather picking between the options, e.g. "should women have bodily autonomy?" vs "should we allow the murder of unborn babies?". While both questions pertain to the same underlying issue; which you choose to ask will influence the answer people give.
So in general, get organized! Start championing issues you care about within the existing political system. Don't wait and allow others to formulate the choices for you!
-
Comment on V Rising 1.0 out now in ~games
Seems like the
ReportHandlerwas deprecated quite a while back. Here is a GitHub issue with some more info (and a workaround that might be applicable):https://github.com/GoogleChrome/web-vitals/issues/482