Back to the channels

#java

15:07 dreamreal Maybe one more deployment for nevet - I'm setting up an MCP server :D
15:29 dreamreal https://bytecode.news/posts/2026/03/bytecode-news-now-has-an-mcp-server
15:29 javabot dreamreal's title: "ByteCode.news now has an MCP server | bytecode.news"
19:39 jbosmans dreamreal, so given nevet is here, what does it do (here) exactly?
19:40 jbosmans well, "here" without parentheses i guess :)
19:42 Para It is a learning computer.
19:43 jbosmans :-D
19:44 Para It kinda is, it's got per-user Markov chains for imitation.
19:44 Para nevet | * channeling jbosmans: sun ceo but what's his name
19:44 dreamreal It's muted HERE because of javabot; two infobots in the same channel is just asking for noise.
19:45 jbosmans first the bots get replaced, then ...
19:45 Para Back in...whenever people did run bot channels with the whole idea that they'd be allowed to chat together to see if something interesting would spawn from the noise.
19:45 jbosmans :P
19:45 dreamreal it has controls set so that it will never emit content in some environments (like this channel)
19:45 jbosmans dreamreal, what does it ingest?
19:45 Para If anything, those channels taught us that Markov chains do not Artificial Intelligence a make.
19:46 jbosmans dreamreal, sorry, what does it ingest from here if anything?
19:46 dreamreal primarily logs and some factoid content, although the factoid trigger is different because the *syntax* is different; nevet's factoid model is a lot more indepth than javabot's.
19:46 dreamreal karma as well.
19:46 Para Does the article poster actually post yet?
19:46 dreamreal It also gives me a chance to create content in the bot without having to change contexts.
19:46 dreamreal Para: what do you mean?
19:47 Para On a bit of a tangent, this is apparently still getting a few updates now and then https://kingjamesprogramming.tumblr.com/
19:47 nevet King James Programming
19:47 dreamreal I've used !article in *this channel* to create content on bytecode.news, yes
19:47 Para dreamreal: Just iterating in my head what commands were available.
19:47 dreamreal https://bytecode.news/posts/2026/03/diataxis was written that way
19:47 nevet Diataxis
19:47 javabot dreamreal's title: "Diataxis | bytecode.news"
19:48 dreamreal https://github.com/jottinger/bytecode.news <-- the actual repo
19:48 javabot dreamreal's title: "GitHub - jottinger/bytecode.news: The nevet codebase, generally for bytecode.news · GitHub"
19:49 dreamreal nevet is running the app module, so it's got everything enabled that it's in the codebase (you can disable features by ignoring their modules in the classpath)
19:49 dreamreal I think the only thing that ISN'T supported that the repo has is OIDC, because I just haven't tested it
19:49 Para Ah, the nifty self-registration which people still underuse IMO.
19:49 dreamreal yeah, I don't store passwords if I can help it
19:49 jbosmans dreamreal, i guess i want to understand what it's doing with things people post here?
19:49 dreamreal jbosmans: very little
19:50 dreamreal logging, more than anything else: it's not doing ingestion of anything more detailed than javabot is
19:50 Para The Markov chain is probably the most involved thing, and it's a curiosity more than anything.
19:50 dreamreal it's tracking karma from #java, I guess, but most other interactions are specific and via privmsg
19:50 jbosmans logging like, writing things to a file and that's it ?
19:50 dreamreal even markov is just ingesting from logs
19:50 dreamreal database, not file
19:50 dreamreal but yes
19:50 jbosmans okay, cool
19:51 dreamreal those logs ARE publicly available, just like javabot's
19:51 jbosmans right, yeah
19:52 dreamreal I think it's better than javabot for most things (there's like ONE feature javabot has that nevet does not) but javabot's not something I'm trying to replace: I want to be ABLE to but I'm not seeking to replace it
19:52 dreamreal (the feature: API search.)
19:52 jbosmans don't make us redundant eh
19:52 jbosmans ;-)
19:53 dreamreal Well, if you know anything about markov chains, you'll know that's unlikely. The markov chain in nevet uses three-token sequences, so it's recognizable but not, like, sane :D
19:54 dreamreal I may bring that up to four tokens, but that requires a lot of inputs. Maybe when nevet's been running a year.
19:54 dreamreal (at four and five tokens, you're starting to see it just straight up quoting people.)
19:54 Para And it'll take time to get the chain to produce any meaningful variation anyway.
19:55 Para But it's a kinda fun toy.
19:55 dreamreal Para: well, it depends: the markov chain is *really* fast
19:55 dreamreal like, retrieving the logs is the most expensive part of the whole thing
19:55 dreamreal and that's around 4ms
19:56 jbosmans having no memory i'm probabilistically (?) reducing you're making us redundant
19:56 dreamreal it's a fun toy and it's related to the whole reason nevet exists
19:56 jbosmans it's nice to have a dataset to do things with
19:57 dreamreal (the first *real* iteration of nevet that ran on #java was actually not *connected* - it ran over #java logs and generated markov data from various users and ran sentiment analysis over it.)
19:57 dreamreal Speaking of...
19:57 dreamreal 15:57 <nevet> Sentiment for irc://libera/%23java: Sentiment +3/10 | Intensity low | Themes: bot features (nevet), markov chains, logging/data ingestion | Summary: Calm technical discussion about nevet bot capabilities; jbosmans curious about data handling, dreamreal explaining architecture—tone informally positive, collaborative
19:58 jbosmans nice :-)
19:58 dreamreal You could theoretically set up a timer operations and have the bot monitor sentiment for various ingress channels
19:59 dreamreal and go "hey op you might want to pay attention to channel #x, sentiment is at -7 thanks to user foobaricus"
19:59 jbosmans yeah that would be bad
19:59 dreamreal I dunno, -7
19:59 dreamreal is pretty aggressive but it's not THAT hard to trip
19:59 jbosmans yeah, exactly :-)
20:00 jbosmans let's reduce interaction to an integer?
20:00 jbosmans but i understand what you mean
20:00 dreamreal nah, it's actually a floating point, you normally measure sentiment in -1/+1 terms
20:00 jbosmans just trying to prove a counterpoint i guess
20:00 dreamreal but you can hit -1 pretty easily, 1 is kinda tough
20:02 jbosmans i'm wondering, what does that mean?
20:02 dreamreal what, sentiment scores?
20:02 jbosmans yeah
20:02 jbosmans more like <dreamreal> but you can hit -1 pretty easily, 1 is kinda tough
20:03 dreamreal Oh, it's really easy to be "extremely negative" but being "extremely positive" is difficult
20:03 GreenResponse from huffington post to full 4chan I guess
20:03 Tenchi heh
20:03 dreamreal like, issue a lot of death threats and be all shouty and you can get -1 fairly easily
20:03 dreamreal but be all shouty and *positive* and it ends up coming off as sarcasm and flattery
20:04 Para The sentiment measures are always funny.
20:04 jbosmans yeah, it's interesting stuff to play around with for sure
20:05 jbosmans but given where we're at (& going i guess), i just think it matters, not to dramatize :-)
20:05 dreamreal nevet uses an LLM to gauge sentiment: it's not doing heuristic. I wouldn't MIND doing heuristics for it but I was specifically trying to get an LLM bridge in place.
20:05 Para Years ago one place I worked at had autosentiment analysis for customer feedback and one message they got which had like three non-swearwords (agglunative languages are very flexible for verbally creative) had a sentiment of -0.68 on scale of -1 to 1. Which was the worst score ever received, so obviously it was printed and put on dev team's wall with the sentiment analysis.
20:05 dreamreal Para: hahahahahahaha
20:06 Para Which of course begs the question, if basically a collection of swearwords is just -0,68 then what can be worse than that.
20:06 dreamreal creative use, maybe
20:06 jbosmans i guess it'd depend on the model
20:06 dreamreal That's part of why I didn't want heuristics
20:06 Tenchi beyond that threshold is sarcasm and passive agressiveness
20:06 Para This was over a decade ago, so who knows.
20:06 jbosmans or heuristics
20:07 Para Maybe the analyzer had general issues with said agglunation and saw it just random nonsense :)
20:07 Para +as
20:07 dreamreal I have a summary generator in bytecode.news that does heuristics, it's... good but not anywhere near great, the LLM summaries are much gooder
20:07 dreamreal BTW, I'm considering offering a reward for people to write for bytecode.news - I'm trying to get it to have enough critical mass to replace TSS
20:08 dreamreal so far, I think *I* have written everything on it but it's meant to be open contributions
20:08 * dreamreal waits for someone to say "oh really, how much"
20:09 dreamreal and the answer is "not THAT much, my money tree ain't bore no fruit yet" but it'd be something and you'd have an actual editor help you
20:10 dreamreal for it to be in the TSS role it needs more voices and more people propagating the URL
20:18 jbosmans TSS == Therapeutic Support Staff ? :)
20:18 jbosmans i'm guessing not ^
20:19 jbosmans also wouldn't go reward for posting
20:19 jbosmans ends in clickbait, see (alas a large part of) #journalism
20:24 kcomhnall dreamreal: interesting. I'll take a look. what's the tech stack and where is the source repo?
20:24 * kcomhnall not concerned about a reward other than learning experience.
20:32 dreamreal kcomhnall: https://github.com/jottinger/bytecode.news
20:32 javabot dreamreal's title: "GitHub - jottinger/bytecode.news: The nevet codebase, generally for bytecode.news · GitHub"
20:32 dreamreal TSS = theserverside.com
20:33 dreamreal and if someone posted something I could use, I don't mind compensating them for it, you make the iron hot by striking it
20:40 kcomhnall dreamreal: cool, I'm bookmarking the repo now. right now I'm trying to figure wth is going on with my own JWT_SECRET environment variable (Windows) and why it's not being picked up.
21:31 jbosmans finished moving away from thymeleaf layout dialect \o/
21:34 jbosmans no more groovy = jar file 8mb smaller
21:34 jbosmans now i can install doom
23:26 johnjay speaking of bytecode. are there any good sites about internals and guts of the jvm?
23:54 dreamreal Yes, they're publicly available, the vm spec is useful
23:54 dreamreal it's kinda deep though