#Neuro ARG
1 messages · Page 84 of 1

Ffmpeg has python wrappers. Not sure about other languages.
well i know that it can be used, im just saying that the way i used the decoder it was slower than the binary, an i couldnt figure out encoding because ffmpeg encoder api is very weird
Ah, that's odd. It shouldn't be slower. It's doing the same thing. Maybe it was a downstream process causing it to wait?
yeah i have no idea, but https://github.com/zmwangx/rust-ffmpeg/blob/master/examples/dump-frames.rs for me ran slower than ffmpeg -i video %03d.png
did you build it in release?
yeah of course
it might be to do with that waiting for it to save the file whereas ffmpeg may be offloading that to another thread
i tried passing the frames to another frame
but maybe the memory sync overhead was too big then
I don't know enough about rust to be sure but it looks like this code is not allowing ffmpeg to handle the actual writing. FFmpeg is usually multithreaded in its IO operations.
Guys, sorry, I'm new to this whole topic, but I'm interested
Has anyone tried to disassemble the last video? I'm talking about 360
It seems to me that in the middle, where there is not a single word, there is something.
But I don't understand it. The idea is to open this video in an editor (for 360 videos) and check what's there. It's like there's something lying there
Welcome. If you haven't already feel free to check the docs. You'll find most of the stuff we found and tried in there.
https://drive.google.com/drive/folders/1qm6GSRlstTmjbyOFCuH1OhSmPK4pi4Aa?usp=sharing
I think we mostly looked at the flattened video (opening it in a "normal" 2d player).
Do you have screenshot or timestamp of where you think you see something?
0:20-14:20 
Oh that whole part?
Do you see anything in particular or is it just a gut feeling that there should be something there? Because we've looked around already and haven't found much interesting there.
looked around already is an understatement

of course that doesn't mean we couldn't miss something
maybe
This may seem stupid to you, but has anyone tried to analyze the music playing at Ermcon 2024? I don't understand all this, I just thought I'd ask
It's a fair question. The way ARGs work is that you have an initial point where real life and the alternate reality overlap. That point is the QR code on the stream waiting/countdown screen. After that, the alternate reality game format demands that a previous alternative reality point has pointed to it. The QR points to youtube channel. The youtube channel points to the soundcloud account, the soundcloud account contains a cipher and a funky background image. That way we know the cipher and the image are part of the ARG. Because we have no leads pointing to Ermcon 2024 we can say that it's part of the real reality, not the alternate reality (:
It's possible, but that doesn't rule out the possibility that we might have missed something that referred us to Ermcon. Also, who knows, maybe this is the part that should be used in the distant future, but the “creator” decided to prepare everything in advance so that no one would suspect. There are many different options and perhaps only analysis can reveal the truth to us

And as I just said earlier, I can't understand this. Sorry about it
That may be the case, my here's an open question: If we were to find a clue outside the chain of evidence, would you want it?
and no reason to be sorry, it was a good question!
Oh right, it's a game. I'm already forgot it

Thank you for the clarification 
no problem, always happy to chat about the ARG 

Were there any ideas about high-frequency noise at the end of The Meaning of Life? I mean, except for the obvious clue about the key. https://imgur.com/a/bhD0XlI
What'd you use to make that image? It's very crisp
Nobody's written anything about it, at least that's been documented, just the as you say obvious https://docs.google.com/document/d/1HrzZYlFmxAzI76Ov0aSkjCe6v1hbzHCTaXK7j34Hdi0/edit#heading=h.fngyrl2ajf05
sonic visualizer can do it for sure, idk about other software
anyway most of it is the Numbers song, but the noise at the end might be from Study or something
Loads of ways, sure, i just like that image cause it's crisp (:

chayleaf guessed right, I did it in Sonic Visualizer
That's why I'm asking. Maybe there were thoughts here, but they were not documented
I don't recall it coming up much, no. If I were to speculate I'd wonder if missing parts are to up the difficulty of reading it, or if it's part of the hint. For all of the letters but the Y, the upper half is missing
even the 1 is missing the upper half, i guess
Cut the number in half? Take the lower half / lower register. Could be something like that
without the cut letters/numbers it says "Y = 28". Only the cut is "KE 1 BIT", assuming we've parsed it right
I was thinking about it, that maybe there should be the rest of the text somewhere. But I haven't found it yet
Could be, yeah
But it’s somehow strange that it is “cut” unevenly. For example, I don't see the "t" here at all. Although the lower parts of the remaining letters are present
I figured the T was further over, but yeah, that'd be some wonky kerning
then again, the 1 is also weirdly spaced from the other numbers
could the spacing have meaning?
the kerning in PSV was kinda similar
so it is. Same weird gaps. It's like a left-leaning mono-spaced font.
I guess it's not so bad if we assume monospaced font
Then cool. A little bit funny tbh
Song and Neuro from Numbers, as well as study girl, static and anxiety from Study. And Neuro and study girl were also shown together. Interesting
Question - how did you get the flat "Meaning of Life"?
Images: https://imgur.com/a/ug5d18K
I downloaded the video from YT (image 1), then processed it through ffmpeg with the -vf "v360=eac:equirect" argument and got image 2. But I still have artifacts from unwrapping at the bottom. But in the document the flat version looks fine
no its not using ffmpeg
you have to download the specific video
list the available formats using yt-dlp
and download the largest one, i think it's mp4 too
I didn't know about this project. Thanks, I'll try
i see we have a new lead
or maybe not actually
i was wondering what we were going to do with the QR code of a windows error code once we figured out which one it is, so it not being related makes more sense
SLNDYK. Who is this?

Мне кажется или тут осталось больше русских/украинцев чем иностранцев?
For the first time I saw that in Numbers on 1:08 in subtitles was text like in Hello world "Thank you"
Хз
vlad269139 chayleaf (?) leafeon Static-sama
ага
Alr, quiet, you can't not speak english here ^^
Sure

English only please
Are you fucking nuts?
Ok, I've been gone a year, what with progress?
I didn't find this in the documentation LuL
It doesn't say anything about subtitles.
Nope, I was just enjoying the music, but I had subtitles on
yeah, I don't remember there being automated subtitles on the video although it's been a bit
And the word Foreign in the end of the video
Automatic subtitles on video
I downloaded it here:
wow [Music] I can never say oh it's been so long counting all the days [Music] finding all these numbers [Music] thank you foreign
wow - it's normal for begining
In beginning of Numbers II there is a "Foreign" from the first second
On 0:57 there is word "foreign" on autosubtitles in Public Static Void
1:59
in the same video during the song interruption "Thank you"
So we can say that "thank you" is not specifically for Hello world
Может быть
@golden mirage
I thought it was already somewhere
@serene fractal @daring acorn English please
I can’t help but call while I’m falling asleep
Okey

Not “thank you”
Foreign
I watched the Google drive today, I almost had a roof
A lot of things…
What programs can I watch the video in detail?
personally I just watch the videos in my regular video player (mpv) or look at the individual frames extracted using ffmpeg. for the audio I used audacity before
Study 3:04 there is "thank you" in 'engeener part'
Again in 4:32 'fallen down'
4:59 again in same part
"anxiety" on 6:23 (same part)
6:46 "inside" and 11:55 "inside" may be "it's overdue"
16:27 "foreign" in I'm not like the other girls
Foreign in the end of "Is anyone watching?"
And there is other 2 foreign and 2 thank you in the last I'm fine
"Thank you" never appeared when she was a static static girl
This needs to be checked
An ordinary player will not be able to show what is hidden from ordinary eyes
true. although it's questionable whether or not something so subtly hidden that we can't see it at all with our own eyes could survive the youtube compression
this is a bit of a long shot but I was experimenting with pixel sorting and realised the effect isn't too dissimilar to some of the stuff in filtered #programming message
not very confident though, just kinda resembles parts of it
How phrases appear in automatically generated subtitles is the most interesting question for me
I'm really trying to figure out how this is possible.
Think of it like AI having alien ears that work very different than ours, and they mishear some of the noises and think it's someone speaking
When potato first joined, we did some experiments with youtube subtitles
The conclusion was that it was too schizo to hide/find clues in automatically generated subtitles
It can't be a bug lul
Well it’s the conclusion
Since there’s no way to confirm it
I’ve seen the sound waves of it
There was nothing unfortunately
And google speech recognition couldn’t have been so wrong. Firstly, there are at least two timecodes I recorded that contain the word "foreign", although the music was not noisy and you could clearly hear the absence of any words. Secondly, I analyzed the subtitles because I myself thought that this might just be a misunderstanding, but in every video (except Filtered, where there are no subtitles at all) there is one Thank you (except for study, where there are several of them)
Yes, all the facts indicate that I am schizophrenic, but it is impossible to completely refute the theory, so this is becoming more and more delicious for me
another victim taken by ARG thorycrafting
Hooray! I'm glad to have a new addiction!
If it can be confirmed, then it'll be an interesting clue, that's for sure. But as it stands, reuploading the audio doesn't result in the same subtitling.
Speech recognition is also imperfect by nature. It's hard to draw a cause and effect between a subtitle and the offending audio.
To be convincing, it needs to be systematically replicable
were still on qr?
we're not on one specific thing ever (:
нє
h euro?
It's a broken attempt at saying no in Ukrainian, and really come on dont be childish guys
just so you know, it's part of the server rules
XD, it`s fun
This is why I can call 
And I also followed voice chats sometimes if there were Russian friends there, so that they would speak English more often, so I already took a pill for headaches, it’s just the weather changed so dramatically that I already feel like I have a fever.
And my buns, I always believe in you and will hold you at any time. 
It's all funny
Это It не is not весело funny
I keep calling
as one said, its all shit and giggles until someone actually shits and giggles
I'll probably have to call 
Yep
Do you think they will answer?
Do you think that the moderator will not answer? 
It's just one message, there's no need to be so radical
better call Saul
The main thing is that this does not happen on an ongoing basis.
I agree
we get it bro) why does this trigger you so much
So that you don't get scolded? 
ok, I understand you 
Sorry, misunderstood your previous message
My English is far from perfect
Therefore, please do not violate 2.1 because of bad memory, I read these rules almost every day

Of course...
I'm just keeping you safe 
Thank you, bro
And don’t pay attention that I’m schizo, I always behave strangely 
This is why I like you 
you are a wonderful person❤️ i love you broo
You are actually cool heartheartheart xoxoxoxo
YouTube auto generated subtitles often mistakenly identify the word “foreign” in many videos, especially music
Can you prove it pls?
Also, is there such a bug with Thank you?
This reddit thread talks about it and provides a few explanations https://www.reddit.com/r/youtube/comments/unx23d/why_do_youtubes_auto_generated_captions_say/
I’ve also noticed it happening with thank you from time to time
thx, i'll read it soon
I think you don't need to worry too much right now because we have moved to another place
well im still waiting for the new vid/stream
Has anyone thought to take the piano notes from Hello World and try mapping them to phone dial tones? I don't have the musical knowledge to know if they map to any notes, but I do know phone dial tones are universal. So the sound a #1 button makes on every phone should be the same.
Could be additional buttons to press when calling the phone number
Huh
That’s an interesting idea
So real life phone numbers being involved is still a possibility?
Not sure, this is like the second post I've ever made in NEURO ARG. It just seems weird to me that per the wiki there is no clues at all in Hello World. Like, why would the puppet master even bother posting it then?
Maybe that's a bit meta, but it just seemed weird to me
The biggest issue I see with a theory like this is every other solution has been an encryption or logic puzzle/not obvious. Something like this seems way to easy
I don't believe anyone has done this. You could do it if you make a spectrogram of the sound and map the frequencies to music notes. My biggest question is what you'd do with two notes playing at the same time
Shortly before the video was posted we were trying out a lot of SMS/MMS stuff with the phone number. I can't say if the video is a response to that, but it's the only other theory that makes sense to me. It's either that, or there's a clue in there. So we pretty much have to assume that there may be a clue somewhere in there
I have a gut feeling it might've been a sort of teaser for two next videos. Since for [filtered] Hello world was pretty much the solution. While the whole video looks like beginning of meaning of life.
well your right
While voice could indeed be just related to us spamming the number
exactly. We need to be open to any reasonable explanation. Not closed off on one track, but also not losing track with all the crazy ideas (:
And while response-to-spam is a reasonable explanation, it is not one which leads to anything actionable - it would just dismiss the video rather than investigate it
In principle, it is possible to map a musical octave (1-7 notes) into a telephone number, however, firstly, the number should not use 0 (although what about a pause?), 8, 9. Secondly, close numbers (2-3, 5-6) and some others should sound like dissonance, which I personally did not hear in music

I've broken down the main music into notes:
mi re mi re mi
mi re mi re mi
re to re to re
C A-sharp C A-sharp C
What can be considered in the case of parsing by octave (do - 1, re -2, etc.) as
32323
32323
21212
1?1?1
Question mark - because I don’t know how to parse A-sharp (A is 6, but is A-sharp 6.5?)
I have a musical education and the only thing that came to my mind was to compare the notes as chords in the circle of fifths, but to do this you need to have a musical education. I don't know if the creator has it, but I don't think he would complicate things so much
I don’t know from what other point of view this can be considered
Circle of fifths - Quinta circle
also A-sharp is a A#. What's the same Hb (or Bb (the same))
Sorry for the confusion, the sheet music looks like this:
E D E D E
E D E D E
D C D C D
C Hb C Hb C
If you have any other ideas, I’m glad to hear them, as I can help at least with them
The idea of text on phone buttons is immediately rejected, since there is no consistent repetition of the same notes.
Maybe A# say to us that we shold try to press # button
Or, for Bb *

Wow, well, unfortunate it didn't seem to line up. My other idea was to try and look up the sheet music by it's Parsons code on this site: https://www.musipedia.org/ to see if it's used in anything else
A search engine for tunes and musical themes and a collaborative music encyclopedia. Only the melody needs to be known to search Musipedia or the Web. The melody can be played on a keyboard or whistled or sung to the computer. The Musipedia collection is editable by anybody, just like Wikipedia.
There's also a piano player thing on there to play music and see if it matches to anything else
I have an idea, that ending in Meaning of life is some sort of rebus (or puzzle, idk how call it correctly). Because of Hiyori (probably from Numbers) and sequence which recalls some part of Study (i trying to understand which exactly). I think it tells to use something from Numbers (instruction, number 572943 or something else) in combination with something from Study (for now i'm thinking of phone number). Just "encoded" instruction like line Applying cipher.numbers1(secret_number) to all given numbers from cipher.numbers2.array() in PSV.
Related (rhetorical) question - why static and Anxiety shown without Hiyori?..
About phone number, what if '+' sign is missleading. What if it's sum or concatenating operation? For now i tried to do it with 572943 number, and use result as key to encoded datas, but didn't see any interesting stuff. Maybe it should be used with something else, or in two steps, which i didn't do. Also i have doubts that this is a phone number, because sequence in Meaning of life recalls first 'I think I saw a ghost', but JS code program appears later.
@onyx sparrow see the toolbox https://neuro-arg.github.io/toolbox.html for the numbers algorithm btw
Oh, I forgot to say - the problem with the idea is that I couldn't end up with string of 16 characters long. So yes, I tried removing some characters from the end or in the middle (because Hiyori slightly overlaps study girl) , so yes, it was wrong for me to say that I tried using result as key. And I didn't try all the combinations because I'm little tired today. I'll try to continue tomorrow
thanks
i like how people are going schizo here again 
we never stopped being schizo
but the fact new people came does bring new life, personally the only arg-related project i have ongoing rn is [filtered] spectrogram analysis and i have low motivation about it because i didnt find anything looking like a transposition cipher from simple analysis
yes
I found out about this and I looked at the doc and stuff and it looks sick 

give this man a true
What if Neuro is just depressed.

Her speech made me sad.
New member. Intrigued by the ARG you are?

Yes. Well, I just enjoy the MVs.
Also simulation hypothesis believer.
the music bops fr
crap I accidentally fast forwarded and lost my progress
what

sorry haven't streamed in a while raising my new dog
💖 20
Does this mean the ARG now has moved to reality.
send your real address to the phone number via SMS and Neuro should dispatch a drone to you as soon as possible

schizo delivery 
We’re waiting for him to say again that he loves her and she’s his death
Schizo delivery 

?
Feed detected a change in youtube feed hash
https://neuro-arg.github.io/arg-monitoring/?fromDate=2024-02-09&toDate=2024-02-10
Probably arg master looking at video stats or smth 
From a quick glance, looks like they looked at hello world
If activity is anything to go by, then maybe hopefully new video soon 
😮
hen,los arg people it has been a while
what happened in the last 2 months?
some progress on meaning of life but not much else
i created an arg toolbox tho https://neuro-arg.github.io/toolbox.html
More progress on the meaning of life but none on the ARG.


Feed says YouTube feed hash updated again. Something's cooking 👀
This time it was Numbers II
Oh?
How confident are we that it isn't some background task like youtube reencoding the video?
If it's a re-encoding, we should pick that up via video hash
The last time that happened, YouTube feed didn't change
Surely we arent coping with the hash changing meaning something 
more like trying to monitor if something we should be aware of changes
for what it's worth, someone left a comment on meaning of life 21 hours ago
The channel's RSS feed
Minus likes + number of views
It updates if any descriptions updates, or a new video is published, for example
There is also date modified and date created. Raw feed here: https://www.youtube.com/feeds/videos.xml?channel_id=UCqOK_pl0LS0e8Lp7HMRDZsw
also, i just realized that we're just about to hit the arg anniversary.. has it really been a whole year??
If date modified changed, but no other video hashes / metadata changes, I'm inclined to believe something else changed. A possible update could be opening the video stats page, although I can't confirm because YouTube is elusive about it
is this what the changed hash is a hash of?
yep, minus likes & views
Those are filtered away before hashing
do we have previous versions of that feed stored?
nope. Definitely should start doing that though
I think so yes, so we can tell what's actually changed :p
But I can say that only date modified changed, because we have metadata hahses for everything else
is that the <updated> tag?
Yes
Right, now I get what you're saying :D
<updated>2024-02-13T05:18:43+00:00</updated>
<media:group>
<media:title>Numbers II</media:title>
Then the next question, what does updated really mean. Surely the channel viewing statistics wouldn't change this
Wait what thats insane
We've explored this before at some point last year. Despite tags, thumbnails, descriptions and titles being exactly the same, that timestamp still updates
How do we test this, hmm. I'll check with a test channel later
But iirc we did test this, and couldn't find what exactly updated the timestamp
I don't recall there being subtitles on Numbers II, was it always there? It's autogenerated garbage so it can't be a clue, but I don't remember it being there at all.
I also don't remember this image shaking at the end of the video, but i feel less confident about that
Study also got autogenerated subtitles now
divide 
This sort of stuff would count as the video being updated, I'm guessing
damn. I'll add that to the list of things to track in the metadata hashes then
I remember the shaking. Don't remember the subtitles
nice
@safe monolith i'm pretty sure the jsons on neuro-arg.github.io are 100% full (i.e. contain all the medadata)
so you can reference that
thanks. I think I can also probably encode all of pytube's initial data minus likes, comments and views. Also want to download the subtitle file for hashing too
looks like yt metadata doesnt include subs
so you'd have to fetch it by other means
I don’t have time to translate everything ‘:(

use powerfull AI auto translator
just a quick question, does anyone know if you can selectively update thumbnails? i.e. different thumbnails for different qualities (youtube)
Friends, I've updated the ARG Monitoring Tool slightly. Changelog:
- Subtitles (both automatic and not) are now taken in consideration
- You can now go to https://github.com/neuro-arg/arg-monitoring/tree/publish to check for the source feed XMLs (
youtubefeed.xmlandsoundcloudfeed.xml).
Note 1: I've compared against the full YouTube API JSON, and determined that our current set of metadata should cover most reasonable changes.
Note 2: If you receive anything from the feed today, there has actually been one real change: Numbers updated has changed. Since it happened today, we don't have a previous state to compare it to.
https://i.imgur.com/5HXyGwA.png
The number is active now! 😮

not to my knowledge
anyone feel like calling the phone number again? just to verify it's still the same answering machine. recording would be nice but not required :)
I can't myself (my phone plan does not allow long distance calls)
same, rarely call so i got a dirt cheap plan
(for future searches: call on 15/02/2024 17:46 GMT0)
for real?
-11dB btw, my ears almost died when I played it back
thank you for lowering it!
xdx
i mean i guess we could leave a message 
save me dot dot dot 
Play something back at +11 dB as revenge on the recording
call on two devices 10 seconds apart
Hmm
I mean, we could also try some of the other handshake noises or send SMS/MMS too
yeah sure, if i can get them on loudspeaker
did this
just send all the videos
this is the same voicemail its always been isn't it? why the sudden excitement 
yep same one. it's fine to have renewed excitement
Hey, the youtube feed hash changed again. All metadata stayed constant
https://github.com/neuro-arg/arg-monitoring/commit/21f68d626960704e321bb58d4fc7057eb2ce6322
Commit, which also shows what changed for the youtube feed
This time it was PSV and study
Which both actually had subtitles to begin with
the order is a clue :schizo:
hello world -> numbers 2 -> psv -> study? 
... ya know what, actually isn't that far off, omg
isn't this actually the order? hello world -> house (numbers 2 soundcloud ref) -> psv (door closing scene?) -> study (clear reference towards the end)
Oh i forgor about numbers 1
hello world -> numbers 2 -> numbers -> psv -> study
coincidence? 
so what actually changed?

On the channel where I uploaded test videos for this: #1112082183235047506 message
The feed shows that that the videos were "updated". Keep in mind that we did these tests specifically for the auto-captions. We also have the following facts:
- When a video is re-encoded,
updateddoes not change. This is visible here: #1112082183235047506 message, and can also be found on the monitoring tool: https://neuro-arg.github.io/arg-monitoring/?fromDate=2024-01-22&toDate=2024-01-23 (Numbers 2 video hash changed, but not YouTube feed hash) - I did not "update" any of the videos; in fact, I haven't had this channel opened since the experiment (unless schizo)
- The video was "updated" months after we confirmed the auto captions
So I'm inclined to believe that "updated" is just YouTube shenanigans?
I will now open one of the videos in Video Analytics and see if the feed updates. Will give it a cooling off period of up to 24 hours

where is the comment from staticsama
try sending this order as an SMS to the number
It would make sense. I'd think that the updated field is just to let RSS readers know that there's a new thing to read, not necessarily that there's a change a human would care about.
should I exclude monitoring it then?
or maybe its better to be extra schizo in this case 
I mean, you could have a soft trigger so we know if it changes but it doesn't ring the big bell. Track both with and without?
I think I can still track it as per normal then. I'm the big bell 
literally means I didn't touch the videos in any shape or form since they were uploaded. Am using that channel as a testbed for feed changes
although at this point i'm pretty much convinced it's just youtube doing shenanigans
yeah, maybe. whatever the case is, it's not something we can pick up, at least based on whatever metadata i have access to
it's actually kinda crazy that re-encoding the video doesn't cause a <updated>, but some unseen background magic can
it could be that it's because a RSS reader would still have to download the video from the same place, and youtube doesn't wanna trigger eager-downloading readers to re-download
that's weird though, cuz seemingly nothing happened this time round. that'll still cause downloads 🤔
true, i'm really just speculating
Must be a clue 👀 https://youtu.be/wQJt0WtB3vE?si=RntUBqeYYO8ge-XQ&t=17
Have this vers added on my list 
It turns out that after the new year only “Filtered” was not updated?
Lmao the bar code reads
<3U2SCHIZOS
Yep
Where's this from?
Saddecc’s art from two days ago
oh, i don't think it was ever reposted here

I didint look into this for a while
Is the meaning of life video fully analyzed by now

no
but there hasn't been any new progress either
in the sense that, we think we found every piece there is in meaning of life, but we cant interpret them
and so we're waiting for a clue
can you put custom metadata in youtube thumbnails?
other than steganography?
the thing we found in soundcloud wasn't "hidden" by vedal, it was part of the site
still one track too
we did "find" the track
namely that it was uploaded on oct 31st
but still no url
there was a banner/profile picture with a hash or smt
no, it wasnt that
bro, chill
the image being part of the site
also this
i'm chill bruh, i'm just saying that the picture wasn't uploaded by vedal, but it was still used in the puzzle
artist radios are autogenerated based on the artist's uploaded tracks
hmm
no idea whether private tracks count
so can we find it on a random playlist?
but it wouldnt matter much
i dont think you can add a private track to a playlist
well you can disable it...
soundcloud disables all social features for private tracks
like liking or commenting
if we were to find the hidden track, it woul be like this?
The docs should have the speculated track url formats
reading the docs? 
actually as a third option a private track could also be shared via a (private) playlist
what if one of the incomplete codes is the password to the account 

Stav’ na avu
What 
))
Like, put it on your avatar says
I think this answer is dedicated to that person)
I already like your avatar
300 star ARG forum go
He know
Oh okay
Done
Hiyori's model and Neuro's new model look quite similar. I guess we can expect another clue?
The logic doesn't follow
What? New model?
?
Maybe he'll release another video on the ARG channel once he comes back, he did say he was working on a lot of things
Yeah. How's the model connected to that?
maybe may 15th
Raising dogs is hard xd
Do you think he child proofed a room to prevent neuro from slamming into walls 

I'm not sure about the walls, but it's better to be safe with steps and edges
hi everyone
こんにちは!新人Vtuberの桃瀬ひよりです!
チャットに対して返答していきますのでぜひコメントしてください!!
定期的に配信していきますのでチャンネル登録いただけると幸いです。
いずれゲーム実況もできたらと思っています🎮
AIによるVtuberの完成を目指して頑張っています。
#新人vtuber #vtuber #ai
#ひよりみらいぶ で配信の感想などをツイートしていただけると嬉しいです!
-------------------...
is this shit part of arg?
This stuff is pre-Neuro
I very, very highly doubt it
What if that's what Vedal wants us to think.
neuro was ready back in 2020 as a bot for osu
Other AI projects have used Hiyori's model before, wouldn't be surprised if there were more. Also, it's not part of the ARG until the ARG's chain of evidence points to it.
but wasnt a vtuber and had no arg so the hiyori stream is still irrelevant
Its run by completely different party. So no, it never was and never will be
how do you come to a conclusion a JP channel thats older than neuro herself could be ARG related. 
Hiyori mention = ARG 
duh how have we not realized it yet, the Live2D company is part of the ARG
I'm currently outside of their office building and plan to sneak inside to find a new clue

a bait
ardupilot mission planner my beloved
no

I'm catching up with the ARG, how tf are you all so smart 

Vedal can now make a new game
flappy dog
your tabs might be part of arg 100%
The power of specialization!
real smart ones are lurking 👀

Never underestimate people who watch AI anime girl
as part of monitoring, i'm currently attempting (read: failing hard, mostly skill issue) to create a database of image vectors of the starting soon screen. each image vector is a small segment of the starting soon screen. the idea is this:
- Whenever a stream starts, the script runs and streams the frames.
- The frames get cropped to the part that really matters (the screen to the left of neuro), then broken up into small squares, and then compared against all vectors in the vector database.
- Top
kclosest vectors are selected; I tried MSE and Cosine, almost no difference. Also tried SIFT, but it takes time to calculate; will take too long. Needs more experimenting when I fully implement it - If the distance between the vector and the mean of the top closest vector exceeds some threshold (TBD), it is an alarming square. Else, all is good.
- If the state goes like this: no alarm -> alarm -> no alarm, then ring a bell. Else, it just means the stream has finished the intro sequence.
All this to definitively say QR code is still "dog and giraffe". Ideally, it will also pick up more obvious changes, like the halloween intro.
are the screens random or are they on a loop?
the screens inside the intro
does every intro begin the same
neither. some of the moving parts move to the beat of the current song, while others are on a fixed loop
There are some differences depending on the stream. E.g. in here: https://www.youtube.com/watch?v=-hq32Cs6pGo star doesn't move, in here: https://www.youtube.com/watch?v=Oc1ov3cDCDI star moves
i see
would have been much easier on me if everything was deterministic, i wouldn't have to do this small square vector thingy
keep a hash of each frame 
with enough time surely everything repeats the exact same way 
Database of all possible leg positions?
Surely the next stream won't have a new intro screen without a QR code 
what the heck I just watched a yt vid of this
I thought it was fake 😭

attempting to quantize 100 frames with k-means 

if my code is right (big assumption), even quantizing in a 1:2 ratio destroys enough information to meaningfully infer anything about the QR code using k means quantization. the rest of the intro screen is just too noisy and it drowns out QR code changes
although i'm suspecting it's got to do with me destroying dimensionality since scikit learn's kmeans only takes a 2D vector, of which only one of them is meant for features
5 hours of life dedicated to strangling numpy to work the way i want it to 😔
maybe i need to PCA it first
ok but, given an unlimited sized database, i should be able to pick up only the qr code differences 
i think the only way to proceed is to specialize what this thing is monitoring, but that'll probably be a last resort
minus the leg and star moving around, do all windows appear in exact positions?
i think so
i have two ideas, both exploiting spatial locality, although i probably have to find time to actually implement them. instead of approaching the problem as "given this square, does it match any square anywhere ever?", i'll approach the problem as "given this square, does it match the same square anywhere ever?"
so in other words, I can try solving:
- "have I ever seen this square with color A before in my life?", requiring big database / good quantization / good guess;
- "If current square at current frame is color A, what is the probability that the same square has a color of B in the next frame? (replace color with any metric that can represent a square)" (this will need a time-series forecasting algo, trained per square), raising an alarm bell if the probability is really low. (edit: actually, it might not even need a time-series forecasting algo, just a simple decision tree should also work)
(1) is static and probably unlikely to work well, unless I have unlimited memory. (2) is probably the best bet, but it's probably going to be fairly stochastic
the metric i'm gonna be using is likely sum of magnitude of MSE of all pixel values between frame x and frame x + 1 unless someone else can think of a better way, i'm too basic to think of any creative method
in a nutshell i dunno what i'm doing, it'll mean something one of the days

are you scanning for changes just in the qr or for any changes on the starting screen at all?
If it's just the qr, since you're relying on the qr being in the same part of the screen why not only scan that part ofr a match and if one isn't found within 2 minutes or something it triggers
that'll be my last resort. the primary purpose is still to confidently say "oh that's still dog and giraffe", but ideally i want to detect any other changes as well
doing it this way also future-proofs this detection thingy, i just have to train it on any new intro
fair
the correct word to use would be: 
but also there are so many talented people working on this
actual
s
It has been another month
I’m back for another prediction
I bet there will be a clue within a week or two
Time flies 
Also vedal has been active lately
Should be able to see him in march
Or we’ll have to wait for another month
Two months should be enough for him
Maybe clue was in the music from SoundCloud?
yk hes gonna return with another qr code
its just a matter of time until the next stream
he said high chance before the month ends
our patience will pay off eventually 
or it will finally break
we will cause chaos in the entire server
we will not be contained
When was this?
oh you mean high chance of stream before end of month, I thought QR for a sec 
bread is going insane
Oh, there's no going here, it's already happened
guys
surely
wtf same timing 


Where was it...
It's all connected 

Oo also a translator from DeepL)
looks like tetris yep

if anything, that's the first karaoke stream since singing v3 that lyrics sang in the intro.
Surely a clue
wat the
There's actually something? I wasn't looking, just listening
fuck we cant get vods anymore?
maybe just cool karoke effect 
Time to analyze every frame of a 2h VOD 
was it chat? 
you sure you want all of them 
tbh does look like nonsense
so you're saying it's a clue
check effects before next song
oh
it's the random song selection thingy
on one hand that is cool on the other :reallymad:
i see
now to figure out if it's always the same, and if it means anything
everything has a meaning
surely it must mean something 
everything is a clue
so depressed
I came here to laugh at you all
The background is very cool, depressing, but so cool
(and also because I am curious if you guys found anything yet)
the length of the random string matches the length of the actual song title. so it will definitely be different for each song
Is this the room where the schizophrenic people go?
The crazy long left this room. It got impatient. Only sane people remaining 
did you see a strange code or symbols before the track started playing on the stream?
pain dot exe
yes
oh, i saw it this time. Yeah, just a shuffle thingy
either it's random or I'm schizophrenic
hi arg people
let's see, here's 1: https://clips.twitch.tv/DreamyCoweringHawkRlyTho-RYNai-Zmj6CzgFml
gonna check the next one
someone sent me a bunch of pics to forward here
they dont have embed perms
person in question wanted to know if this qr code is new or already known about
yep this qr
minimally orange piece is known cuz I'm checking specifically for it
cc @daring acorn
likely actually just fully known
i mean, it looks like the same, but i guess we can compare the pieces to the old one
its the same yeah
Sorry about this
hopefully whatever i'm building now can say it for sure eventually
nah dont worry
I'm just new here
its fine
npnp, welcome
yes, it's the same. And no worries (:
Have a look at this if you're curious
https://docs.google.com/document/d/1mttkOpO5Mj_g2rP5DUWcLkP-obYBDfUhFx7oGtE0BKs/edit#heading=h.30qcsc29266k
Sorry
Don't be sorry tho!
did you compare?
suddenly there are 3 pixels taking shape
Today, by the way, is sad
I lazily pasted a crop of one of the image provided into the doc and looked at the other pieces then eyeballed it (then of course deleted the pic i pasted cause gotta clean up!)
It’s terrible how efficient you are
actually
thats not random characters
its actually encoded
ill send a random one
T<%W s yeAN9)G NE^z9>AF
proof?
i split both my clips to frames, i'll check if there's an equivalent one
correct
but it’s worth going through each one
nah
If I were a curator, I would have stuffed something in there
i love frame by frame analysis 
M e l a i
4 _ ) } )
absolutely. One frame could be non-random bs
more than likely yes, but anything's possible, really
They seem to have randomly placed spaces, doesn't look like a well-known encoding like base64 to me
I'm guessing it's something like a Caesar cipher with a random offset and invalid characters replaced with space
Could probably confirm that by just collecting a bunch of strings and doing some statistics
the random thingy gets fast enough to become 1 string per frame
Clue?
after a while. not sure if it's just an encoding issue
Enigma machine /j
base64: T^]4 i donN)'t NE9)M NF%
it's not base64, invalid characters
mhm
fck this random
ah, did you get this from the second clip
si
it's in frame 286, i just wasted time 
I mean, we could transcribe all the frames if we're feeling extra
can i get it in an mp4
the 2nd clip
arigatoue
unless arg master evil, surely if code then it must appear multiple times 
is it always/usually the same number of characters as the "target" text it finally lands on?
the spaces line up
for these 5, yeah. lemme check to see if there are any exceptions
unless we're stripping spaces then this has to be nonsense if the spaces always line up
Bitter choco Decoration - syudou
EC7\x{ ujoVC &}qDc`UZs] E \Mi5T^
$J2V<S Evru4 "!*=cbasXn o yEBfKu
sad
accuracy isnt high
EC7\x{ ujOVC &}qDc'UZs] E \Mi5T*
might be cuz im too zoomed out
what are the resolution of these pics? and size?
they are already 1920x1080
fml typical discord saving images as 100x100
the key is the difference of the binary values between transitions 
proper saving
can (visually) confirm that all of them should be the same length
shitty discord's saving
discord skill issue obviously 
easy, use image thumbnail instead
clue??

ocr results still cringe
get better OCR
shut up
is your ocr better with white text or black text
because iirc they're trained on books which are generally on white paper
tbh i wasnt using black text cuz the background destroys my eyes
and kinda fucks the ocr for some reason
ah nvm, just have to invert treshold levels
guys i dont think this will give anything 
well yeah
clearly this is training for the eventual return
ya'll actually doing it 
we're also enjoying the stream, trust us

btw, i ran the monitoring tool just for fun

"just for fun"
gets a fuckton of results
"so all of the videos changed"
feed hash has changed, but i think it's not a big bell moment

whats a feed hash
more comments or something else?
funny, the date updated changed 2 days ago. for reference this monitoring tool updates every day automatically, yet this change only reflected today.
means nothing to us though, cuz all the metadata and subtitles are the same
anyways, after working on the intro sequence matching thingy for a long time, i've decided to change my approach again, because figuring out whether 841 squares changed per frame will never work with realtime. also my image processing-fu is not good enough
first, we have a database of pre-defined 20x20 squares we wanna compare against. they need to obey spatial locality; i.e. same location. if they don't obey spatial locality, it runs 2 magnitudes slower
second, this database needs to be small. so I can only realistically track a very small number of squares at a time. anything bigger is impractical, at least to me. (0.02x realtime speed, not very useful)
third, i use multi channel SSIM as a metric. this is a little slower than MSE, but it is less sensitive to changes that don't matter
i've tested just a small 20x20 QR square with some streams, and given we set the threshold just right (in my experience, 0.95 is a good number, where 1.0 is a perfect match), we can detect even the smallest of changes
so unfortunately i didn't meet my secondary objective, which is to write a script that can be incrementally updated on any new intros, but i'll make up the skill issue with maintenance work :^)
the above method should allow me to run it on a github runner, and write a twitch stream trigger s.t. the monitoring tool runs on every new stream
other than the QR code and tetris, is there anything else that is interesting enough to track?
Mm?
same tbh
I feel like my prediction is a bit too precise
Is vedal lurking here?
like Vedal announced the stream almost the next day
Anyway
The next clue i believe will drop in march 3 or 13
3 is more promising tho
or maybe it's on the 18th/19th

itll be on the 15th 
clue? I hardly know 'er
watch it actually happening on may 15th
I love May
who is May?
who may be this may person?
May from pokemon?

sliding titles are also info? sounds amazing
Looks like march 3 is not the answer
Unless he decides to stream later
Which i will be asleep 
did we ever look at the music videos from the songs in the arg because they seem not unrelated
currently cooking something
which music videos are we talking here?
the ones from the songs sang by neuro in the halloween video
sorry that was originally in my message but removed it by accident
oh, right. The music video of the original song that neuro sings a cover of.
it's a bit rushed and a little scuffed
looks good to me
I sorta tried to realign some of the major parts where it goes off sync with neuro later
What could we extract from this side by side?
idk I feel like the first song at least slightly matches up with it's video
if it isn't useful enjoy the neuro-miku duet
ok so i kinda had to rush off after sending this but my main point was how close the proper glitching matches up with the start of the stairs falling at 2:45
maybe that is what fallen down means
We wait 
Hi friends, small update for intro deviation detection thing. I've (partially) deployed something rudimentary using SSIM scores and ground truth squares into the monitoring tool. Ideally, on every stream, the monitoring tool gets triggered automatically and looks for squares we deem important, rating it to how similar it is to the ground truths. This has been designed to run as close to real-time as possible, so we get a report as soon as the intro finished rolling. The threshold is defined by comparing the ground truth square with some number of other videos, and then calculating some sort of interval.
If the intro of a stream has scores that falls below that interval, then we say the ARG integrity of the intro is compromised. Otherwise, at least the parts we care about is unlikely to be compromised, but it doesn't mean there are absolutely no changes.
There are still some changes I would like to make (e.g. package cache so we don't get cold boot issues, nearly automatic selection of ground truth squares from a reference, etc), but I think it's ready. We'll only know during the next stream
tl;dr schizo from neuro arg made thing see if dog and giraffe / tetris still dog and giraffe / tetris every stream
^ the above are monitored under neuro_twitch_identifiers and evil_twitch_identifiers. each "true" represents if the squares were similar (or not). currently the intervals used are scuffed ^ 999 because I only considered 2 neuro vods and 1 evil vod in total; currently in the process of including more vods
Weekly report
@safe monolith thought of an idea;
get 3 minutes worth of black&white pixels from the intros, add a black box near the timer (if you want, its better this way tho) and do the pixel average
explaining a little better:
the black & white pixels come from tresholding, its better to only use the white pixels but you can use black or both
a "black box" is just erasing the timer part
and doing the average is simple math
making the image black and white erases too much information imo. Calculating an average over 3 mins also erases too much information, cuz imagine if the qr code changes, it will only minutely affect the average. It would be indistinguishable from compression / artifacting from the encoder.
Even though it looks the same to us, every video is actually slightly different in terms of randomness introduced due to encoding
I had a graph at some point that compares pixel averages to SSIM scores. They spike at the same time, but SSIM scores are less sensitive to noise

random question
whats happening?
havent been here for a while
anythink neww
as always you'll have to check the timeline - we can't remember when you were here last for you.. :)
i now realize this isn't even the original video for rolling girl 
encoding wouldnt matter with m3u8
- treshold can be adjusted
Encoding is an issue from the streamer -> twitch. It introduces noise, which means that strictly speaking, even if the exact same sequence of frames were streamed, they will still be entirely different
The thresholds approach is a good idea, the tool is using it atm. The problem is more turning the frames black & white, then calculating pixel average across 3 minutes. We lose too much information to do a meaningful comparison
Your idea will definitely work if there is no lossy algo anywhere in the pipeline
i wonder if you could posterise or something
Well tbh also varies. In the same 3 minutes, maybe QR code appears twice instead of thrice, which also ruins the average
Hmm, it would be hard to make it work for evil tetris pieces tho
I think the reds/red-like especially. If the structure (the semi-black lines around the red squares) is not preserved, it'll be hard to tell if something changed
but it would somewhat solve encoding noise
Yeah maybe I can run both. The current setup for particular changes, and a new one using posterize + rgb pixel value average as a big change detection
I'll experiment when I'm free (not soon
) and let ya'll know
Our official docs are the google docs
Which page are you using? You might have more success with the docs, yea (:
https://docs.google.com/document/d/17KrbPUAVu4wDTbj3sBtWTnjRsjDTCdh89F_5L1LvX0A/edit#heading=h.j7yktin3xotl
That website is still in pinned messages, proving that mods don't look at this channel.
Uncontrolled schizo
mods are not real
oh, that's where it's from
ye its down
even i forgot about it
since i created it

next time surely…
Has the ARG been solved?

er n yep 

crap, are we gonna get pranked on april fools 
Just so that it doesn’t feel too empty, and this is the first time I’ve done subtitles seriously, or well, in general, I had fun 
And I wish I had cut the video. 
Also
Digging back
I found that this is probably some sort of public source code in anaconda
Idk help me
windows pathname
logs in as root
somehow still localhost
what is happening on that shirt
https://unix.stackexchange.com/questions/398410/what-is-the-file-anaconda-ks-cfg-in-root
this file is mentioned too
This is what I deciphered
where is that art from again
Thank you 
idk how to use the pure word format
There’s a thing looks like p but not really so I wrote @
And underscore are the parts too distorted to understand
(for reference a normal centos login looks something like this)
so i'd suggest your local login line should probably be last login
only place i can think of that'd have arrows is systemd logs, though i don't think that's whats being shown in the text
It looks to me like improperly displayed ANSI escape codes.
https://en.wikipedia.org/wiki/Special:Search?search=ANSI+escape+code
ANSI escape sequences are a standard for in-band signaling to control cursor location, color, font styling, and other options on video text terminals and terminal emulators. Certain sequences of bytes, most starting with an ASCII escape character and a bracket character, are embedded into text. The terminal interprets these sequences as command...
Likely from an older version of conhost that didn't support them.
Yeah it's ANSI escape codes, particularly for color probably
yeah i realise now that systemctl only puts them on the right aswell
Very good

anyone else just feel a slight wave of nostalgia when rewatching the Numbers video 
its been 10 months since it released 
time isn't real

only 2 months for ARG anniversary 
inb4 we get tomfooleried, ARG ends 1 day before anniversary
The arg will never end
im still surprised we havent selected a name for study yet 
i advocate for Iris (even if it is incredibly unlikely to be the actual name) cause of AIris
study is a good name
change my mind
fair enough 
personally i would just stick with "study-sama" and wait for a potential official name from the arg
Why do we need to label them with names? It seems to me like solidifying a highly speculative interpretation of the lore
it would also be confusing if we named her Iris and then the actual AIris became an official character later

ye that is fair
iris was just a suggestion, tho i do feel like we need to decide on what to call her, ig study-sama for now can work.
but everyone calls her by a slightly different name so it can get confusing 
well to be fair there's no reason to not use a name neuro mentioned
not to stereotype, but i can't see study as a blonde 
||dunno if thats offensive
||
||atp anything can be
||

too much of same image

I didn't mention a thing

let's call them bob
it was revealed to you in a dream
too bad i don't remember anything about it (i only remember that it got posted in #neuro-network before this channel for some reason, and it had a weird title)
i'd say you're schizo, but i had a similar dream some days back. Clue was on a stream
we sane frfr
theres gonna
Prophet Vanor!

wait, it has been 3 months since meaning of life? 
we probably have something that allows us to continue, but if we do it's in a hard to reach place and requires so much schizo we find it easier to just wait for the next clue
yeah there is 100% something in there that allows us to continue and i feel like we even have all the parts required to continue but they just don't fit together yet

yea 
it was the same with the two numbers videos, we could have solved it before but weren't schizo enough, so got a clue to solve it with public static void
i feel like this entire merging thing is significant to the solution
imo, the halloween intro may... or may not have something else
youve reminded me to do this
ok what the hell resolve i have two audio tracks and yet you somehow export 4?
i wish i knew
ok there we go

Hi arg people !
I'm just passing here to ask, do we know what are used as the vocal for study ? (what synth-V ? Or what utau ? Or Original ?)
I ask because I heard a "very similar voice" that made me think about the arg from a song using a synth-V vocal
its highly speculated to be synth-V, Eleanor Forte, but we have no confirmations other than "sounds like it
"
some speculations about last shot trained vocals too
there was a schizo theory about a real counterpart's voice used to train the voice for study
Personally I think it sounds too good for synths and AI, based on following the development of both. But in the end it can all be reduced to speculations - we don't know
Thank ! That the synth-V that cover the song I listen to, so I now personally think it may be it ?
if someone want the song, (so they kind of understand what I mean by being similar ?) send me a mp.
I will not post it here as I do not want to pollute here
tbh you can go ahead and just post it here. I don't exactly have the audio setup & sense. Someone else here might be able to match the vocals better than I can
oh ok !, so here it is for reference: https://youtu.be/F67ErmWEN4k
Yes, I'm so glad to finally release a PV for Medicine! Another original song featuring Eleanor Forte, this time I got to collab with my good friend Chevrefee!
Big thanks to Chevrefee for doing such an amazing job with the video, with the art, compositing and storytelling! Such an amazing artist.
Go check out their stuff: https://www.youtube.com...
cast a wide net, never know. Maybe a music genius is lurking 👁️ 👁️
I was checking the other author work cause why not
check the last Eleanor Forte song
and.... the font...
it really look like hello world font
and the background have the same vibe as the arg
that just... a strange felling to have so much similarity

The font was identified in the docs, If I'm not mistaken it's common enough that we can even write in it in the docs
fwiw
That would be funny
If that a common font that explain some of the similarities
and we should write it in the doc its nice it's written in the doc
That's one hell of a coincidence nonetheless
idk i've seen that font around a lot of art places, not sure where though
at most we could prove that one inspired the other, but that wouldn't let us solve anything
True, true
while true, it's kinda nice to just bask in the coincidences
(at best we can meta profile the tones of the arg, but that's stretching it)
fair (: So long we don't start stalking RL people :p
ew 3d people smh
true 2d person enjoyer
haha, i had a listen and heard this in eleanor forte and it completely threw me lol
arg ptsd real
For me, it was this line on the first video I send that triggered me (just missheard cold -> old)
The timbre is very similar-sounding, I'll grant you that, and much more compelling than what I've heard before.
What makes me still question if it's the same is that this song (and the channel's latest song as well) is full of tiny little voice glitches not present in Study.
If we presuppose that it is the same synth, then it would have to be a later version of the synth, operated by someone who knows how to work it better, and/or cleaned up in post.
I find that possibility more believable now
I tried to listen to differences from the video, I referenced and the arg song and to me, it looks like there is strange pronunciation of some consonance (more prononced but less clear ?) I do not find (often ?) in the arg song
arg song "sound more smooth" maybe bc of speed diference ?
if it's minor enough, it probably can tuned / adjusted with synthesizer software. never worked with it before though, but i've seen some videos of people playing around with synth-v to know it might be possible
I think they just use the original voice of neuro and evil for the arg songs tho
I remember it was a similar voice
Vedal accidentally leaked it when he was talking about the future mommy ai vtuber
Can’t find it tho
Isn't it just Ashley without pitch?
interesting, but i don't remember this. gonna need a source if you have one 
leaked or not, v1 and v2 voice was already replicated
well, early v2 voice
theres been so many changes to it...
v1 is ashley with 16 pitch (?)
cant remember well
v2 is more complex tho
based on what you've heard, does it sound anything like study?
azure ashley 1.52 pitch
Lemme search that
v1
No one clipped it 
I’ll have to search that in raw streams
So fun
Neuro archiver my beloved
I’ll put that to the todo list 
Wait, as I was not here for some time, I read the timeline and saw the track & brute forcing the short link.
Excluding potential rate limit, why can't we just lunch hundreds of io thread/process making request to the SoundCloud api ?
because there's literally quadrillions of potential secret keys
that's not bruteforceable
Almost as if it was designed to not be bruteforceable
It's easy to just add more bytes to a random id. And even if it wasn't, this ARG isn't about hacking soundcloud ;)


Yeah true
whoa who is this 



