The Communication Cube


NOYB is asking the EU comission to orderly repeal the EU-US Data Privacy Framework because the US supreme Court just ruled that the FCC can't be independent because of the US Constitution. Since the EU legislation requires an independent data protection authority and the US had nominated the FCC, which can't be independent, the privacy framework won't comply with EU law, just like Privacy Shell and Safe Harbour were shot down in the courts. Well I doubt anyone really believed it was solid, given incompatible legislations, but this moves the courtain we're not supposed to look behind...
 
Null said:
I'm using it to be lazy by having books read to me.

Software: Lector + SherpaTTS
Click to expand...
How's the performance, does your phone get hot/bog down, or are there long pauses between the chunks where it has to compute before the next output?
And how is the quality, does it sound natural/better than the google network voices?

I've been using google network voices in FBReader, but was never really happy with the quality. The offline ones are worse, rather robotic.
Piper-TTS is quite performant running on Rk3588, it runs in real time, but it's about the same quality as the google online stuff.
Kokoro-TTS on the other hand is a big difference, the voices sound very natural. But it's quite a lot slower and fully loads all eight cores all the time. I'll try to convert the ONNX file to RKNN format with the rockchipNPU-toolkit2 to hopefully make it run on the NPU.
 
kuru said:
How's the performance, does your phone get hot/bog down, or are there long pauses between the chunks where it has to compute before the next output?
Click to expand...

There are noticeable pauses. But then I am running the apps on a phone that was released in 2022, and it has never been the most powerful of devices.

www.devicespecifications.com

Oukitel WP18 - Specifications

Specifications of the Oukitel WP18. Dimensions: 83 x 174.8 x 18.5 mm, Weight: 380 g, SoC: MediaTek Helio A22 (MT6761), CPU: 4x 2.0 GHz ARM Cortex-A53, GPU: IMG PowerVR GE8300, 660 MHz, RAM: 4 GB, 1600 MHz, Storage: 32 GB, Display: 5.93 in, IPS, 720 x 1440 pixels, 24 bit, Battery: 12500 mAh...
www.devicespecifications.com

kuru said:
And how is the quality, does it sound natural/better than the google network voices?
Click to expand...

The voices sound pretty natural to me. You can test them out at https://huggingface.co/spaces/willwade/sherpa-onnx-tts
 
I asked GTP-5.6 Sol
Hi. What is J-Pop trying to harden the population for by being more stressfull than Gabber or Death Metal?
and got
If you mean the peculiar phenomenon where some J-Pop feels more neurologically exhausting than music explicitly designed to sound violent, then yes—there is a plausible audio-production explanation. It is probably not a covert resilience program.

A lot of modern J-Pop is extremely information-dense. You can get, simultaneously, fast vocal melodies, frequent chord changes, countermelodies, busy drums, synth arpeggios, guitar fills, backing vocals, sound effects, and very little empty space. The arrangement may change substantially every few bars. Your auditory system therefore gets fewer opportunities to settle into a stable prediction.

Gabber, by comparison, can be brutally loud and fast, but often has comparatively simple perceptual organization: kick kick kick kick, repeating bass/harmonic material, recurring loops. Once your brain has inferred the pattern, much of the apparent extremity becomes predictable.

Death metal has a similar paradox. It may contain blast beats, distortion, growls, rapid guitar work, and complex rhythms, but its sound frequently forms relatively coherent perceptual "blocks": distorted guitars occupy one broad texture, drums another, vocals another. Dense, certainly, but often stylistically consistent over a section.

J-Pop can instead attack you with continuous novelty:

  • very bright mastering and prominent upper mids;
  • aggressive compression/limiting, reducing quiet moments;
  • densely layered instrumentation;
  • vocals kept highly intelligible and forward despite the dense backing;
  • rapid harmonic rhythm and unusually colorful chord progressions;
  • frequent transitions, fills, modulations, and arrangement changes;
  • multiple simultaneous melodic lines competing for attention.
The important distinction is therefore acoustic intensity versus cognitive load. Gabber may punch your cochlea harder. Death metal may sound more threatening. Hyper-dense J-Pop can make your auditory cortex fill out paperwork at 180 BPM.

So, as for what Japan is hardening the population for: based purely on the sonic evidence, apparently processing seventeen independent streams of information while somebody modulates to a new key and a cheerful anime character explains the power of friendship.

I don't know whether I've learned something useful.
 
Djoga'Ro said:
I asked GTP-5.6 Sol
Hi. What is J-Pop trying to harden the population for by being more stressfull than Gabber or Death Metal?
and got
If you mean the peculiar phenomenon where some J-Pop feels more neurologically exhausting than music explicitly designed to sound violent, then yes—there is a plausible audio-production explanation. It is probably not a covert resilience program.

A lot of modern J-Pop is extremely information-dense. You can get, simultaneously, fast vocal melodies, frequent chord changes, countermelodies, busy drums, synth arpeggios, guitar fills, backing vocals, sound effects, and very little empty space. The arrangement may change substantially every few bars. Your auditory system therefore gets fewer opportunities to settle into a stable prediction.

Gabber, by comparison, can be brutally loud and fast, but often has comparatively simple perceptual organization: kick kick kick kick, repeating bass/harmonic material, recurring loops. Once your brain has inferred the pattern, much of the apparent extremity becomes predictable.

Death metal has a similar paradox. It may contain blast beats, distortion, growls, rapid guitar work, and complex rhythms, but its sound frequently forms relatively coherent perceptual "blocks": distorted guitars occupy one broad texture, drums another, vocals another. Dense, certainly, but often stylistically consistent over a section.

J-Pop can instead attack you with continuous novelty:

  • very bright mastering and prominent upper mids;
  • aggressive compression/limiting, reducing quiet moments;
  • densely layered instrumentation;
  • vocals kept highly intelligible and forward despite the dense backing;
  • rapid harmonic rhythm and unusually colorful chord progressions;
  • frequent transitions, fills, modulations, and arrangement changes;
  • multiple simultaneous melodic lines competing for attention.
The important distinction is therefore acoustic intensity versus cognitive load. Gabber may punch your cochlea harder. Death metal may sound more threatening. Hyper-dense J-Pop can make your auditory cortex fill out paperwork at 180 BPM.

So, as for what Japan is hardening the population for: based purely on the sonic evidence, apparently processing seventeen independent streams of information while somebody modulates to a new key and a cheerful anime character explains the power of friendship.

I don't know whether I've learned something useful.
Click to expand...
It could have responded with: Yes, Japan hardens their population with J-Pop.
In comparison with ChatGPT, I noticed that Gemini sometimes does respond with much shorter and brief replies instead.
 
TeDaDeS said:
It could have responded with: Yes, Japan hardens their population with J-Pop.
In comparison with ChatGPT, I noticed that Gemini sometimes does respond with much shorter and brief replies instead.
Click to expand...
But what for? :confused:

@Null I consider you an expert on this, ye ol' Hatsune fanboy. What for?
 
I like these homebrew 8bit computer projects, this channel has an extensions on the breadboard computer design from Ben Eater:


Djoga'Ro said:
But what for? :confused:
Click to expand...
I think it just expresses the feeling of (urban) youth in Japan. If you've ever been in an arcade in Japan the same chaotic experience is expressed in their songs. I guess it hits different for Japanese people compared to foreigners.
 
pyrat said:
I thought the coding agents were trained on practically all public code out there in the net, and then arguably some private code some people may not realize it's training them or something.
Do you think those evil organizations are going to find an even more poisonous corpus ? That'd be a big achievement in its own, if only as an existence proof.

I guess the bad actors (I mean those outside the relevant corporate boards) may find ways to trick the agents to misbehave more or more predictably, but not sure it'd be through training.
Click to expand...
Like, you can hijack some abandoned code or domain, or plant weaponized instructions... Is what this article explains training ? I tend to think it's more in the inference process... Maybe I don't get it ? I guess it could affect both inference and training ?
 
Djoga'Ro said:
Btw, 9Lana > Hatsune Miku. :p
Click to expand...

4HpYoPHw.jpeg
 
Well, I guess it's not specially surprising, there are lots of allegedly illegal inputs and output in LLMs, but I have to find this disgusting beyond all else:. xAI has been sued for training on child abuse stuff.
I just don't know how they pretend to defend from that. Hopefully by proving it didn't happen. Apparenty xAI is regurgitating that shit or similar shit digested from that shit. But it started me thinking...
I mean I know LLMs are too undeterministic to keep any hope to control them, but let's pinky assume you want to train an LLM to never produce, accept, indulge or conflate in CSAM.
How do you do it whitout CSAM material in the training set? AFAIK only the police and judiciary can possess CSAM, and only to prosecute it. A private LLM enterpreneur can never touch any of it.
So how do you label it and get your bot to recognize it and avoid it ? I mean I don't need to have ever seen any of it to recognize it, but a bot ?
 
Possible Russian war escalating to EU/NATO countries (Lithuania, Poland) this year by increasing hybrid attacks or actual hot war:

Quote: "Polish Prime Minister Donald Tusk warned that Russia is preparing for possible escalation andsaid the next seven to eight months will be critical for the security ofUkraine, Poland and the entire region."
Nuance: "Estonia and Latvia said after Ratcliffe’s visit to Moscow that their assessment of the risk of a direct Russian attack on the Baltic states had not changed and remained low. At the same time, officials in Tallinn and Riga stressed the need to remain vigilant because of the threat of hybrid attacks, sabotage and possible Russian provocations."
Source: https://english.nv.ua/nation/tusk-w...-seven-to-eight-months-critical-50636796.html

These risks have been raised since recent news appear to point at Russian current activities in EU/NATO countries:
-News: "The German government will soon announce who is responsible for an attempted drone attack ‌on Leipzig/Halle airport earlier this month"
Source: https://www.reuters.com/business/ae...sives-after-failed-airport-attack-2026-08-25/
-News: "Scores of migrants cross into Poland through hidden Belarus border tunnel"
Source: https://www.reuters.com/world/score...ough-hidden-belarus-border-tunnel-2025-12-12/
-News: "Germany investigates arms cache found in woods near Berlin"
Source: https://www.dw.com/en/germany-weapons-berlin-russia-arms/a-78371750

However, Russia denies the claim they are warned by the US for these plans:
Quote: "The Kremlin on Thursday dismissed U.S. media reports that Russia could attack a NATO member state and that the purpose of CIA Director John Ratcliffe's visit to Moscow this week had been to warn Moscow against such a move."
Source: https://www.cbc.ca/news/world/us-russia-cia-nato-9.7322340


Other news, China and India are pushing for de-dollarization causing problems for the USA, buying massive amounts of gold:
Source: https://www.idnfinancials.com/news/68068/goldman-sachs-monitors-central-banks-gold-buying-spree

pyrat said:
Well, I guess it's not specially surprising, there are lots of allegedly illegal inputs and output in LLMs, but I have to find this disgusting beyond all else:. xAI has been sued for training on child abuse stuff.
I just don't know how they pretend to defend from that. Hopefully by proving it didn't happen. Apparenty xAI is regurgitating that shit or similar shit digested from that shit. But it started me thinking...
I mean I know LLMs are too undeterministic to keep any hope to control them, but let's pinky assume you want to train an LLM to never produce, accept, indulge or conflate in CSAM.
How do you do it whitout CSAM material in the training set? AFAIK only the police and judiciary can possess CSAM, and only to prosecute it. A private LLM enterpreneur can never touch any of it.
So how do you label it and get your bot to recognize it and avoid it ? I mean I don't need to have ever seen any of it to recognize it, but a bot ?
Click to expand...
I think all large tech companies are required to prevent that kind of material to be hosted. I think they usually have a database of hashes of existing material, and algorithms to detect it such pictures (which sometimes blocked Apple accounts, if you have pictures of your kids stored in the Apple cloud). So I can imagine this sounds worse than it is, as it indeed sounds like they would train an AI to accurately detect such pictures. But generating images would indeed be the complete opposite purpose.
I remember in Half-Life they refer to Civil Protections as CP: https://half-life.fandom.com/wiki/Civil_Protection
 
You must log in or register to reply here.

Similar threads

EvilDragon
Welcome to the new OpenPandora / Pyra / GP32X Boards!
2 3 4
Replies
71
Views
38K
NiGame
NiGame
Thorax
Snesmote - The Competitor To The Icontrolpad
Replies
6
Views
6K
Thorax
Thorax
R
The Race Is On!
Replies
17
Views
6K
NssOne
N
Yellow Kitty
Hello, Just Joined The Community
4 5 6
Replies
104
Views
32K
λ the β-Redex Reducer
λ the β-Redex Reducer
Ziz
Some thoughts about the board
2 3 4
Replies
72
Views
21K
renegadechic
R
Back
Top