Back to the channels

#java

00:20 * Square joined #java
00:37 * kusanagi joined #java
00:58 * dob1 joined #java
01:04 * dob1 joined #java
01:10 * x1bncwn joined #java
01:38 * droid3 joined #java
01:39 droid3 I got a huge issue with java programming language. Now i love the language along with many others and its a very powerful language.
01:41 droid3 However as memory keeps increasing for home computing and definition for HPC supercomputers...etc. aka 16GB , 32GB , 128 , 1TB much ,...100TB DDR...etc
01:41 * ChaiTRex joined #java
01:41 droid3 The language is fundamentally stuck at creating only allowing the creation of arrays of 2GB entries
01:43 droid3 we are fundamentally stuck with java when creating an array[int] because we can only index an array by signed in we only can create 2GB entries at most.
01:43 droid3 This is not a limit on the size of the array but a limit on the amount of entries an array can have.
01:44 droid3 I find this very restrictive. For example BigInteger use arrays internally for the size of numbers so technically the size of an integer is restricted to 2GB digits if each entry stands for one digit.
01:46 droid3 No for most this aint to much of a problem but for say c/c++ you normally can create arrays index by longs so in this case you can create an array technically with 18,446,744,073,709,551,616 amount of entries
01:47 droid3 18,446,744,073,709,551,616 far exceeds the amount of memory even a supercomputer will have available to it for quite sometime.
01:48 droid3 So can sun/oracle beef up the array[int] to array[long] or create some indexing datatype that isnt going to suffer from array barriors i call it.
01:50 droid3 even if long is signed we still would have 9,223,372,036,854,775,808 entries which still be more then enough for quite sometime
01:52 droid3 I mean alot of my math i do i have to switch back and forth from c/c++ to java when i am dealing with really huge stuff. If only java had bigger array entry creation
01:53 droid3 This is fundamentally the problem which also restricts all the java Collections to being 2GB entries as well.
01:53 droid3 However one can piece arrays together and classes can be in theory any size you could construct an array of objects or a class of many arrays.
01:54 droid3 To exceed the 2GB barrior that java has... however its still not clean one should just add the ability to create larger arrays
01:55 * pengu1nx2 joined #java
01:56 droid3 This is not only for large million and billion digit number computing or math research or fun.
01:56 * ForeverDreaming joined #java
01:56 droid3 There are benefits for nonmath based programs to beable to create large huge bigger then 2GB entry arrays.
01:58 * stfstfm joined #java
01:58 droid3 Before this time not so much because memory was like 4GB or 8GB but now as memory exceeds and is getting ridiculously larger even for home computers. To me it just make sense to account for this in the java programming language. And not only restricted to java but all languages... native languages like c/c++ are easy to keep upgrading/uping the data type size. I imagine VM based languages are more of a challenge
02:00 droid3 And possiblely can break compatibility with with previous compiled to bytecode programs ...so i understand the reason it may be a hold off feature. But its definitely a feature that needs to be added if java is ever going to survive the future of HPC or bigger data/ more memory.
02:02 droid3 PM me if you want to talk on the subject its a bother to me. Sure i can use JNI to just call c/c++ do huge array computation in that world but its just better to not have to split shit up and have it all in terms of one java huge array.
02:03 droid3 At that point i usually would just switch to another language to achieve what i want but i feel it not complete until java has the same abilities
02:12 droid3 Another issue is if/when you do upgrade java to bigger arrays. I would imagine you also need a feature to shut off garbage collection for certain huge data structures or the garabage collector would chew up most of the computing /processing time searching for null references
02:13 * ferdna joined #java
02:14 droid3 And also i do like garbage collection features in languages but it also would be nice to have the ability to shut them completely off for the whole program or parts of the program or for certain data type/structures you create.
02:15 droid3 So as to save performance when dealing with huge data structures or huge memory allocation ...etc. We never really cared or need this aspect with the language but now that huge memory and big data/object/data structures are becoming a thing...its going to get important for the performance of the language
02:22 * ChaiTRex joined #java
02:27 dreamreal droid3: wait, what
02:27 dreamreal droid3: I don't know what you're doing, I just sat down, but WHAT
02:27 dreamreal Java's going to be FINE
02:27 dreamreal for your specific application, who knows
02:32 dreamreal Valhalla may address some of it, but fixing it isn't trivial, and the number of applications it would address is relatively small, for what it's worth: that doesn't make it better for you, but there ARE efficiency reasons for array indexes being integral.
02:33 dreamreal It would be NICE to have some flexibility here but for 99.9999% of the applications out there - being conservative - not really. (And note that I work in a problem space where this IS a liability.)
02:48 droid3 So no matter how bigger are memory gets where always going to be stuck at 2GB entry arrays size. I just find that in future is going to be restrictive maybe not for most right now but eventually
02:50 droid3 So BigInteger will basically be stuck at 2billion digit size representation so wont be like computing the next biggest mersenne prime in this language :)
02:51 droid3 But joking aside i still think there we are going to want to create sometype of builtin datastructure for java if we cannt fundementally upgrade the int to a long
02:51 droid3 for arrays
02:52 droid3 I mean i imagine its far easier to add a bigarray data structure to java then to change the underlying fundamental language of java.
02:53 droid3 That way new versions of java will be compatible and there wont be any issues with old compiled bytecode...etc
02:54 droid3 I like java to have a datatype to have unlimited arrays entry sizes... as not only for math but for other huge stuff.
02:55 droid3 I get changing the builtin data types would break stuff but adding a new datatype wouldnt
02:56 droid3 Heck newer versions of java added lambda functions abilities and before that generics ,...etc it make sense to have unlimited sizes.
02:57 droid3 I mean you got BigInteger /BigDecimal in theory these are arbitrary precision crude libraries. I just think you have a way to scale arrays which is the fundemental datatype that allows for scaling everything else.
02:58 droid3 Really builtin nonarray based data types are when you dont need the precision that exceeds the hardware specific registers/sizes.
02:59 droid3 when you do you switch to BigNum based libraries. But c/c++ scales arrays entries to any size quite easy with the native language.
03:00 droid3 java if we just scaled unlimited array entry creations we be perfect.
03:01 droid3 Aka we have no restrictions on the BigInteger or arbitrary precision libraries as well as no restriction on array entry sizes.
03:02 droid3 Basically we be independent only restricted /confined to the RAM and storage sizes we have available to us when writting a program.
03:02 droid3 in java
03:03 droid3 Also it wouldnt really break bytecode or old JVM /java runtime versions. It just mean you have to run the code on a more modern JVM going forward.
03:04 droid3 Which usually is the standard typically as you use newer JDK you eventually upgrade the JRE in sync with them.
03:05 droid3 So it doesnt break portability with existing code running or have to be recompiled unless of course you running it on the new JVM...
03:05 droid3 So not sure why you say would be to difficult to do so over the benifit.
03:06 droid3 Bit is not a huge workaround for me it just an inconvenience when doing mathematic computations in this language that all. At the moment.
03:07 droid3 All languages have there strengths and weakness though it be good to strengthen this aspect for java
03:08 * ChaiTRex joined #java
04:53 * NeXeN joined #java
05:35 jreicher droid3: is there any particular reason you don't like Vector?
05:36 jreicher Or ArrayList?
05:36 jreicher I'm not entirely sure what you're after.
05:37 * Aedil joined #java
05:49 Bombe I mean, nobody likes Vector…
05:50 Bombe But yeah, sounds like a lot of non-pointiness.
05:58 * smlckz joined #java
05:58 jreicher I can't remember the last time I used Vector, but it's always the name I think of first. Maybe because of other languages.
06:06 * smlckz left #java (WeeChat 4.6.3)
06:09 * stfstfm_ joined #java
06:15 * agnivn joined #java
06:24 * stewi joined #java
06:26 * jreicher joined #java
06:35 * tabmow joined #java
07:01 * lostlazy joined #java
07:56 droid3 no but vectors are restricted to the same size limit as arrays they just are dynamic in how you can size them where as arrays are fixed.
07:58 droid3 The issue jreicher is that any java Collection has this max 2GB restriction on the number of entries. I use all kinds of different Collections for different things when i am coding depending on what the problem is/ what i want to use for it.
07:59 droid3 That not the point its the fundemental restriction on max number of entries dictated by the index being a signed int.
08:00 droid3 It isnt even the size i came make with objects and arrays of object i can create well over any memory i have on any computer system in the java language like most others as well.
08:01 droid3 But the issue is for java there is a small limit 2GB on the amount of entries you can create.
08:01 droid3 Not how big a particular entry is
08:48 * Afroboy joined #java
08:57 * TomyWork joined #java
09:03 * kathadris joined #java
09:17 * TomyWork joined #java
09:24 deebo definite "limitation", but if you're hitting an array size limit, you might want to think about what you're doing, there's probably something much mroe suitable
09:26 Maldivia eh yeah, having any collection in memory with more than 2 billion entries sounds like "this needs to be done differently"
09:30 * MikeBux joined #java
09:30 Bombe While it’s certain an issue that you can’t create arrays with more than two billion entries, I’m quite certain it’s not the issue for a very, very large amount of people.
09:33 Maldivia and there is a way to get around it also, if you really like
09:34 Maldivia MemorySegment.allocateNative(MemoryLayout.sequenceLayout(1_000_000_000_000, MemoryLayouts.JAVA_INT), globalScope()) -- there you go, an "int array" with 1 trillion entries
09:35 Maldivia hope you have enough ram
09:39 Maldivia or simply slice your data int[1024*1024][1024*1024] -- an in array indexing 2^40 ints -- get(long index) { return array{index / (1024*1024)][index % (1024*1024)]; }
09:39 Maldivia so it's a non-issue
09:41 * jbosmans joined #java
09:56 kcomhnall Maldivia: sooo, you have MemoryLayouts.JAVA_INT for a 1 trillion value? does that work?
09:57 Maldivia kcomhnall: no, I have a segment of a trillion entries, each of type JAVA_INT
09:57 * kcomhnall would've thought long if anything
09:59 Maldivia kcomhnall: on memory level, the above MemorySegment command is equivalent to new int[1_000_000_000_000]
10:06 kcomhnall ah,.. and the only restriction then is the hardware / hard drive capacity?
10:06 Maldivia the context was that native java arrays are limited to 2^31-1 entries
10:06 kcomhnall 2billion
10:06 Maldivia (technicaly, 2^31-8)
10:09 * kcomhnall thinks they were talkin about FFM in here before
10:17 Para that'd be in context of dmlloyd's smallrye-ffm probably
10:19 Maldivia droid3: as for your BigInteger issue - the spec says they support values up to 2^(2^31)
10:20 Maldivia so that's "only" ~ 650 million digits :D
11:16 * TomyLobo2 joined #java
11:21 * x1bncwn joined #java
11:23 * agnivn joined #java
11:38 dreamreal yo Maldivia
11:47 Maldivia yo yo
11:52 * Nemu64 joined #java
11:52 dreamreal Maldivia: I spun up a second UI for bytecode.news yesterday: the model works! (Both UIs suuuuuck but that's not the point)
11:57 Maldivia dreamreal: so I should be submiting a AI generated UI, you're saying? :D
12:07 dreamreal Maldivia: hell, an AI generated both of those! But the key isn't "the UI" - the REST model behind it is working and it's pretty damn fast. I've got an older UI here written by someone else against an old version of the services - the reference ui looks like crap but actually validates the back end well (I found a lot of holes in the services while writing it)
12:07 Maldivia :D
12:07 dreamreal so now we have a stable API and a good UI can actually keep up. But that's the model I wanted for TSS back in 2007
12:08 dreamreal multiple ingress - one article draft was written with !suggest and another with !article here on IRC, and there're "anonymous posts" (submitted while not logged in) and I can spin up other ingress models for the system really quickly, I think
12:10 dreamreal The UIs really do suck: basic literally does nothing except read and echo, and the reference is an SPA that accounts for some popular bots but otherwise renders nothing up front, which is bad
12:10 dreamreal but the model's working, the reference UI *is* an actual MVP (emphasis on 'M'), and if I get a decent-looking UI up that doesn't rely so heavily on SPA, plus some actual content...
12:11 dreamreal I'm gonna see if I can kill TSS
12:12 dreamreal The site isn't anything really "new" in terms of content, TSS was always better at commuting value through curation rather than possession; I don't know of anyone else intentionally designing the multiple egress part nor do I know any other site besides the LLMs who're so aggressive about ingress
12:16 dreamreal The idea is that ... like... okay, today, right? We have two discussions: one is droid3's thing about memory limitations, and then you have the REAL gold in my blathering, of course. But both of these can now easily be *captured* for bytecode.news - in the ideal, I can capture the interactions with droid3 and have them summarized automatically (and included for verification, because those summaries
12:16 dreamreal are NOT intended for publication directly). But they're sort of pre-baked, so we don't see the discussion just fade into IRC logs if it's worth preserving.
12:24 Para MVP, as in Most Viable Product.
12:24 dreamreal hah
12:24 dreamreal in this case, "minimum"
12:24 Para I just ran cloc on my sideslop, I mean my website infra.
12:25 Para 21k rows and there's like three Spring handlers in this :D Of course I'm doing this The Hard Way on purpose and this runs tons of things one wouldn't need if all I wanted was a blog, but still...
12:25 * dreamreal shrugs. I get it.
12:25 dreamreal I mean, technically all *I* wanted was "a blog"...
12:26 Para And now we can play hangman over IRC.
12:26 dreamreal and it's hooked up to github, discord, slack, irc, the web, polls RSS, has factoids, games, utilities, can recite poetry at you (and evaluate it)... not great poetry, but still!
12:27 dreamreal and now we can play hangman over IRC. Wait until I get star traders hooked up. The economic model's actually pretty stable.
12:31 dreamreal One of the things I need to do for hangman is hook up an external word list; it only has 990 words and they're all pretty restricted in length. But hangman's a silly toy, so it's... pretty low priority.
12:32 dreamreal I want it to use aspell or dict for words, but it needs to be able to run on multiple systems, so it'd have to scan
12:50 * agnivn joined #java
12:51 * zorone joined #java
12:59 jbosmans upgrading to spring boot 4 && jackson 3 && hibernate 7 && fun
12:59 dreamreal Why jackson 3?
12:59 dreamreal I've done that but haven't seen the benefits yet
12:59 jbosmans afaik 2 is or will be deprecated
13:00 jbosmans no more/less than that
13:00 dreamreal sure. but it's going to be a while, and 3 has very "point release" energy to me
13:01 dreamreal I look at 3 and I get it, but I'm also not going "oh thank god they fixed _that_", except in ONE area, and that area was so low-concern that I had to think about it to find it
13:02 jbosmans yeah i see what you mean, but even so the immutable JsonMapper etc does bring some value and allows me to construct fewer instances
13:02 jbosmans yeah, it's json mapping
13:02 dreamreal (specifically: ObjectMapper -> JsonMapper as a specific migration; you probably want to use the "proper form" now. But if you use ObjectMapper you'll be fine.)
13:02 jbosmans so it's not important
13:02 jbosmans yet it's uber important
13:02 jbosmans :-)
13:02 jbosmans yeah i'm halfway there @ ObjectMapper -> JsonMapper
13:02 dreamreal yeah. I just feel like it's ... a move that I haven't been able to say "OMG MUST DO" yet
13:03 jbosmans [x] construction through JsonMapper.builder()
13:03 dreamreal I mean, spring!
13:03 dreamreal you do that once and you're done
13:03 Para Just Slop It
13:03 dreamreal but then you want to change the reference types
13:03 dreamreal I'm of the opinion that "AI slop" is the fault of the person driving the AI
13:03 jbosmans also added a bunch of JPA annotations, since apparently some naming changed again going hibernate 6 -> 7
13:04 dreamreal jbosmans: yes, but that's pretty minor
13:04 * tabmow joined #java
13:04 jbosmans after the joy of the changed id generation with hibernate 5 -> 6 :')
13:04 dreamreal fun migration, though: I've done it myself recently and enjoyed it almost 1%
13:04 jbosmans sounds about right
13:04 dreamreal (including moving to jackson 3.)
13:05 jbosmans claude helped out with a few Q&A i have to say
13:05 jbosmans and did a pretty good job
13:05 dreamreal That's another recent shift: I've been using codex over claude, and it's been pretty effective
13:06 jbosmans yeah, openai is out here
13:06 dreamreal claude's context window has been ineffective: I have to explicitly remind it to do basic triage things, even though the directives SAY "never ignore these processes, I will bitch about them endlessly" and I still found myself bitching about them endlessly
13:06 jbosmans right, good that there's plenty of competition
13:06 dreamreal like, constant "Sure, claude, go ahead, fix main, why not OH YEAH I TOLD YOU NOT TO"
13:06 jbosmans hopefully it stays that way
13:07 jbosmans gemini 3 is apparently default in jetbrains junie
13:07 dreamreal and "sure, go ahead and consider that feature done without mutating src/test in any way"
13:07 Para I've been running claude succesfully mostly by not allowing it direct access at all.
13:07 jbosmans got some pretty good results for (pretty simple) react stuff
13:07 dreamreal for UI the AIs are pretty good
13:07 jbosmans yeah i'm pretty restrictive too
13:08 jbosmans mostly hobby stuff for now
13:08 dreamreal I don't mind AI code - but I use it pretty aggressively and I direct it a lot
13:08 dreamreal very few "oh I want this go code it" sessions without a *lot* of lead-up design discussions, which helps
13:08 dreamreal and demanding test coverage helps a lot, too
13:09 jbosmans yeah, "agentic development" && "i'm getting there"
13:09 jbosmans maybe "i'll be getting there"
13:19 * kcomhnall joined #java
13:25 * vobar joined #java
13:25 * GreenResponse joined #java
13:52 * jamezp joined #java
14:18 jbosmans grrr my jackson streaming reading code is breaking
14:21 dreamreal I wonder if they changed it!
14:21 * dreamreal giggles.
14:22 * tabmow joined #java
14:24 * sa02irc joined #java
14:32 * x1bncwn joined #java
14:40 jbosmans yeah, they did and i guess it's better
14:40 dreamreal It's slightly concerning that this is a guess. :D
14:40 jbosmans haha
14:41 jbosmans fewer to no checked exceptions is a win
14:41 dreamreal Is it?
14:41 jbosmans the immutable mappers same
14:41 jbosmans imho yes
14:41 dreamreal the mappers, I'd agree with
14:41 jbosmans i could still catch
14:41 dreamreal the checked exceptions, I'm neutral on
14:41 jbosmans i either removed the try/catch, or changed to catch Exception
14:42 jbosmans depending on whether there was any handling
14:42 jbosmans 7/10 -> no handling
14:42 * vitaliy joined #java
14:43 * tronexte joined #java
14:51 * Square joined #java
15:02 * deepy joined #java
15:07 * sbalmos joined #java
15:16 * sbalmos joined #java
15:20 * skinkitten joined #java
15:42 * vitaliy joined #java
16:12 * stfstfm joined #java
16:15 * stfstfm__4788 joined #java
16:18 * stfstfm_ joined #java
16:32 * magla joined #java
16:46 * dinomug joined #java
16:57 * cation joined #java
16:58 * MikeBux joined #java
17:04 * stfstfm joined #java
17:15 * gas51627 joined #java
17:30 * kcomhnall joined #java
17:56 * magla joined #java
18:11 * tomaw joined #java
18:29 * lex_v joined #java
18:49 * nevet joined #java
18:49 * CodeGeek!~codegeek@about/java/CodeGeek changed the topic to: Welcome! || Read Channel Rules at https://javachannel.org/ before participating. || Paste limit is two lines; ~pastebin lists options. || No applets, please. || Minecraft, Android, and Javascript all have their own channel. || You are being logged.
18:56 * ferdna joined #java
18:57 * ernimril joined #java
19:12 * Shell joined #java
19:22 * ernimril joined #java
19:24 * ForeverDreaming joined #java
19:28 * Everything joined #java
19:30 dreamreal jbosmans: thanks, BTW. I felt bad about nevet still being on jackson 2, so I'm doing the migration myself. Great fun all around. Thaaaaaaaaanks.
19:32 jbosmans haha, with pleasure + don't give me that crap, you were somehow tempted ! :-D On a more serious note, suggest you make sure you tested it all :-)
19:33 dreamreal oh, I have tons of tests. It's a 20 second build if I don't use tests... and a 3 minute build if I do.
19:33 jbosmans sounds relatively efficient overall
19:35 dreamreal We'll see. When I'm done, I'll be able to describe how painful it was; overall the build's pretty modular, so it shouldn't be too bad.
19:35 jbosmans regarding jackson, the immutability is pretty nice, i had a bunch of placed where defensive copies were made "just to be sure"
19:35 jbosmans because if those placed failed, hellfire would've rained down (or something)
19:37 jbosmans also, project nearly never need explicit "do" or "don't" fail on unknown properties, or serialize/deserialize null properties whatnot
19:37 jbosmans with mutable ObjectMapper i basically never trusted myself
19:37 jbosmans "just to be safe"
19:37 * agnivn joined #java
19:37 dreamreal heh. I'm pretty careful about those myself. But I just realized... I use json everywhere, partially because I use jsonb fields... and I just thought "hey, what about toon"
19:38 jbosmans hah same @ jsonb
19:38 dreamreal Most of my storage is pretty light, so I don't think it'd save much, but that's a nasty thought to land out of the blue
19:38 dreamreal article Comparing JSON and toon
19:38 nevet Idea session started: "Comparing JSON and toon". Use 'content <text>' to add body paragraphs, 'includeai' to enable AI summary/tags, 'done' to save, or 'cancel' to discard.
19:39 dreamreal content look up definition and spec for toon
19:39 nevet Content block #1 added to "Comparing JSON and toon". Use 'content <text>' to add more, or 'done' to save.
19:39 dreamreal content size differentials, data scale
19:39 nevet Content block #2 added to "Comparing JSON and toon". Use 'content <text>' to add more, or 'done' to save.
19:39 dreamreal logs 20m
19:39 nevet Added 24 log messages (last 20m) as content block #3.
19:39 dreamreal aisummary
19:39 dreamreal done
19:39 nevet Idea saved as draft: "Comparing JSON and toon" (3 content blocks).
19:42 jbosmans i'm guessing no jpa ?
19:42 jbosmans @jpa i don't feel like i need to be an advocate for it, and it's bitten more than once
19:42 Para On that tangent, Spring 6.1's JdbcClient is actually kinda nice for us raw-SQL-enjoyers.
19:42 dreamreal *definitely* JPA
19:43 dreamreal spring's on 7 now, lamer!
19:43 jbosmans :D
19:43 jbosmans haha
19:43 Para dreamreal: And Boot is in four yet I'm doing 3.5
19:43 dreamreal jbosmans: one remaining section to fix, in openjpa
19:43 dreamreal sorry, openapi
19:43 jbosmans yeah well spring 7 + jpa + hibernate + jsonb field needs an extra annotation from where i'm sitting
19:43 dreamreal you guys got me thinking jpa :/
19:43 dreamreal jbosmans: I was already there, jackson 2 was a decision I made out of familiarity
19:44 dreamreal this is me being dragged kicking and screaming into the new modern
19:44 jbosmans dreamreal, it's hibernate 7 behavior related
19:44 dreamreal jbosmans: yes, but like I said, I was already there :D
19:44 jbosmans fwiw @ColumnTransformer
19:45 jbosmans dreamreal, what, you've complited the migration spring boot 4 and/or spring framework 7 etc ?
19:45 dreamreal spring boot 4 was already done, as was hibernate 7
19:46 dreamreal it was spring boot 4 + hibernate 7 + jackson 2, this is changing that last
19:46 jbosmans ah
19:46 jbosmans yeah makes sense
19:46 jbosmans afaik most impactful one
19:47 jbosmans well, hibernate is always impactful
19:47 dreamreal The deprecation of @Temporal annoys me. I get it, but still...
19:47 jbosmans i have to do these kinds of migrations "in one go" to preserve my sanity
19:47 jbosmans oh yeah
19:48 jbosmans i think i got bitten by that
19:48 dreamreal well, yeah, migrating to spring 4, it's probably BEST to do all three in one fell painful swoop
19:48 jbosmans basically workaround code to parse a (previously) java.sql.Timestamp to a LocalDateTime blew up
19:48 jbosmans yeah
19:49 jbosmans major spring boot version upgrades have been rough experiences for myself tho
19:49 dreamreal I have all green tests but the code isn't idiomatic, working on that now
19:49 jbosmans tho it makes sense, major dep upgrades are what they are
19:50 jbosmans dreamreal, kudos for the tests
19:50 dreamreal They're working: migrating to idiomatic code is failing!
19:50 dreamreal Which is exactly what they're supposed to do
19:51 jbosmans "tis good to fail"
19:51 dreamreal much gooder to fail in tests than in production
19:51 jbosmans codex?/claude code?
19:51 Para I'm kinda amazed by how well the testing works in this ...thing of mine.
19:51 dreamreal jbosmans: between the two: codex
19:52 Para I mean obviously it works because I spent two days prompting and rewriting it until I was happy, but still it's kinda pretty.
19:52 jbosmans oh right, you said before, sorry
19:53 dreamreal I like claude but I'm thinking lately it's not been as good as codex, a little surprising considering the lead it had
19:53 Para I haven't tested codex; Claude at least seems to be a good baseline for things. Anthropic's business strategy seems to be on promoting overspend and Ouroboros-like setups which I don't like.
19:54 dreamreal claude has been struggling with context windows. codex has done a better job lately of respecting my edicts, which is nice.
19:54 cheeser the new codex model is supposed to be pretty nice.
19:54 Para Claude is curious that it seems to work better when you give it the minimum possible context and lots of extra "focus on this", "ignore that" type of instructions.
19:54 dreamreal (I have strict instructions to try to resist stroking my ego: my ego will be inflated best when MY CODE EFFING WORKS, LLMS.)
19:55 Para My default prompt is basically five sentences of "treat me like an adult".
19:55 * Everything left #java
19:55 dreamreal cheeser: I haven't tried the new CODING model, because it emphasizes speed of response and I don't care about that: I want the code to be good, not to have it generated quickly
19:56 dreamreal Para: Yeah, mine's more like 30 lines of treat me like an adult and these are the things I value, you stupid machine (like: "tests" and "make it work, make it pretty, make it fast in that order" and stuff like that.)
19:56 Para +:tää:
19:56 jbosmans so i've had an openai subscription for a pretty long time, but then the DoD thing happened, and i felt like i should re-investigate things?
19:57 dreamreal DoW, you mean? The DoW thing is against anthropic, not openai
19:57 * dreamreal despises "DoW" as a moniker
19:57 jbosmans really .. thanks for addendum :)à
19:57 jbosmans those things matter i think
19:57 dreamreal I've had both openai and anthropic subscriptions; I work for a company for whom this is all important
19:57 jbosmans LLM decided kills is just ......... wrong
19:58 dreamreal oh, anyone handing such decisions to an LLM is off the rez
19:58 dreamreal but that's not what's happening
19:58 dreamreal okay, idiomatic usage is ... "in" and all tests green
19:59 jbosmans sounds good
19:59 dreamreal biggest problem was null handling
19:59 dreamreal hmm, hmm, i wonder if I trust this enough to deploy
19:59 jbosmans biggest ethical problem is null handling? :-D
19:59 dreamreal jbosmans: anyone handing ethics to an LLM is a problem, at any level
20:00 jbosmans dreamreal, agree, i guess that's what anthropic was trying to prevent
20:00 dreamreal It's not difficult to get an LLM to decide that SURE, it's fine to shtup the babysitter
20:00 * jbosmans && defending bn$ company :')
20:00 dreamreal no, that wasn't the problem
20:00 dreamreal anthropic and the DoD both had legit points
20:01 jbosmans i read the anthropic points, somehow missed the DoD points?
20:01 jbosmans sorry, as in "legit DoD points"?
20:01 jbosmans imho they'll win their lawsuit
20:02 jbosmans just becuase somewhere someone, unexpectedly, still has common sense
20:03 jbosmans dreamreal, i just noticed, from jackson 2->3 migration to this :-)
20:04 Para I'm old enough that "DoD" to me means Day of Defeat.
20:04 jbosmans \o/
20:04 jbosmans those were the days
20:05 jbosmans i always was the worst @ knowing the maps, including anzio
20:07 jbosmans nothing a browning couldn't fix ;)
20:07 cheeser dreamreal: then you are neither moving fast enough or breaking enough things
20:08 dreamreal oh, anthropic definitely has legit points, so does the DoD
20:09 jbosmans cheeser, hasn't move fast and break things been sorta deported since?
20:09 dreamreal I'm not willing to take sides
20:09 jbosmans :o
20:10 cheeser jbosmans: promoted actually to focus on global efforts
20:11 cheeser "And Zuckexander wept for there were no more worlds to break."
20:11 jbosmans cheeser, haha perfect :-))))
20:12 dreamreal I feel back for zuckerberg, actually
20:13 dreamreal nobody else shares the burden of being him. The downside of that is that we still have to worry about how to pay bills, but still...
20:13 jbosmans yeah :-(
20:14 jbosmans luckily he jumped in with the other billionaires when the time was right
20:15 jbosmans it must've been tough for him tho
20:15 dreamreal Hey, it's gotta be hard to be a lizard person in a world full of mammals
20:15 Para I wouldn't be surprised at this point if Zuckerberg turned out to be an android/lizard hybrid.
20:16 dreamreal all right, haha, we've had our fun, let's be kind to even soulless cretins
20:16 jbosmans haha, those were some good meme/vids back in the day
20:16 jbosmans actually, pretty recently
20:17 dreamreal William Gibson had a point that's kinda hard to disagree with: the very very rich are very difficult to consider as being actual humans any more, their perspectives are... warped
20:17 dreamreal this doesn't mean they're not actual humans, it's just difficult to actually understand them, and it's difficult for them to understand us muggles
20:18 dreamreal "oh man, I need to figure out how to pay my rent" or "oh man, I broke my arm, the ER's gonna be $4k" gets met with "why don't you just... you know, not have that 17th mansion, 16 is enough" or "just have them replace the arm, my good man."
20:19 jbosmans yeah
20:19 dreamreal "Would you like a small country? I have a spare."
20:19 jbosmans fwiw that stuff is spreading
20:19 dreamreal well, the internet was supposed to bring us together, and instead it's split us apart
20:19 jbosmans second that
20:19 Para The split point is somewhere around 2011, ironically.
20:19 jbosmans it didn't at first tho !
20:20 dreamreal Oh, I thought it was GREAT - in 1994 I was talking to ACTUAL GENERALS on usenet
20:20 dreamreal I still do, but back then it was very egalitarian
20:21 Para That'd be Arab Spring events and how information flowed freely thanks to in no small part, Twitter. Not the event itself, but the showing that _this_ is what Internet can do. And that made the old money/power _very_ interested/worried.
20:21 jbosmans i remember installing quicktime and then waiting like half an hour or more to watch a very brief star trek first contact trailer :-)
20:21 Para But unfortunately this is so political stuff that even I have to now point out that this doesn't compile with javac so...
20:21 dreamreal I think the inflection point you're seeing is incorrect, but I get it
20:22 GreenResponse there are forces that are vehemently against bringing humans together: like Asian autocracies Russia, China and Arabic countries
20:22 jbosmans i think it's good to talk about "things", whenever possible
20:22 GreenResponse that’s why Russian trolls try to destroy decent communication
20:22 dreamreal yep. But let's stay on topic before we drift back into politics.
20:23 dreamreal topic, please.
20:23 dreamreal I am not saying I disagree with you, at all: but let's stay on topic.
20:23 jbosmans GreenResponse, just for my info, where do you live?
20:23 dreamreal topic.
20:23 dreamreal Take that to PM.
20:23 Para What's the state of non-LLM AI things?
20:23 * waznot joined #java
20:23 GreenResponse sorry, got carried away
20:23 dreamreal Para: still great! I use AI for markov chain stuff in nevet :D
20:24 Para People focus on those but I'm actually more interested in image processing things.
20:24 dreamreal GreenResponse: it's easy to do, no shade, I do it myself, but let's apply discipline
20:24 * vitaliy joined #java
20:24 dreamreal Para: that's not really AI, though, is it? It's ML, but not really AI
20:24 dreamreal i hate the conflation of terms
20:25 Para I've been thinking of a pipeline which takes crap quality video -> convert to frames -> mask away noise parts from the video -> run the remaining through gaussian splatting.
20:25 Para It just seems a lot harder to find the bits for this kind of thing.
20:26 dreamreal yeah, well, you're supposed to hand it to an LLM now!
20:26 dreamreal those data centers ain't gonna pay for themselves, you know!
20:26 Para Maybe I need to create a darn startup for this.
20:38 * sa02irc joined #java
20:55 dreamreal okay, I found a gap for jackson3: springdoc and openapi. It's easy to FIX - add the jackson kotlin module back for the openapi generation phase, with no other code changes, but without it the openapi spec loses the ability to specify required fields
21:18 * monkeyPlus joined #java
21:23 * nevet joined #java
21:23 * CodeGeek!~codegeek@about/java/CodeGeek changed the topic to: Welcome! || Read Channel Rules at https://javachannel.org/ before participating. || Paste limit is two lines; ~pastebin lists options. || No applets, please. || Minecraft, Android, and Javascript all have their own channel. || You are being logged.
21:27 * Ragnor joined #java
21:36 * waz joined #java
21:39 * jreicher joined #java
21:40 * stfstfm_ joined #java
21:43 * jamezp joined #java
21:46 * zorone joined #java
21:58 * dinomug joined #java
22:16 * talbot joined #java
22:23 * MikeBux joined #java
22:30 * chris64 joined #java
22:38 * x1bncwn joined #java
22:53 * kcomhnall joined #java
23:22 * metalmaniac joined #java
23:29 * cation joined #java
23:40 * Square joined #java