bravura

@nikisweeting ArchiveBox is awesome and we'd really love it to be more awesome. And sustainable!

I've posted issues and PRs for showstopper issues that took months to get merged in: https://github.com/ArchiveBox/ArchiveBox/issues/991 https://github.com/ArchiveBox/ArchiveBox/pull/1026

You have the opportunity for the community to lean in on ArchiveBox. I understand it's hard to do everything as a solo dev, we've seen many cases in the community where solo devs get burned out or have personal challenges that take priority etc.

It's hard for us users to lean in on ArchiveBox when after a happy month of archiving, things start break and you're left with maintaining a branch of your own fixes that aren't in main. Meanwhile, your solution of soliciting one time donations just makes the whole project feel more rickety and fly-by-night. How about thinking bigger?

We NEED ArchiveBox to be a real thing. Decentralized tooling for archiving is SO IMPORTANT. I care about it and I suspect many people do. I'm posting this so other people who care about it can also comment and chime in and suggest how it can become something we can rely on. Because archiving isn't just about the past, it's about the future.

Maybe it needs to be a dev org of three committed part-time maintainers, and a small foundation that people recurrently support is what grants it? IDK. I'm not an expert at how to make open source resilient. There have been discussions about this in the past, but I think it's worth a serious look because ArchiveBox is IMPORTANT and I want it to work any month I decide to re-activate my interest in it. I invite people to discuss ways to make this valuable project more sustianable and resilient.

Zendesk Competitor. That's actually the name of my AI-generated death metal band.

How is transcription done?

Introducing Lacan: A Graphic Guide, by Darian Leader and illustrated by Judy Groves.

Freudian theory really was just a way to psychoanalyze Freud and his complexes.

Lacan jettisoned the weirdly specific Freudian stuff and had a more general template, with a focus on the relationship between language and the subconcious.

Thinking this through more deeply, I agree and see your position. It is creepy to surveil audio and possibly send in full to Shazam. [edit: And post the original audio recordings online.] The ethical way to do this would be to use your own code to decimate the audio signal to extremely low dimensionality.

How do I join the party? Is there a quickstart so you can hear the vibes of Berlin?

[edit: It would be awesome if others could collaborate on this and had a guide on how to do it!]

I've also been using the nougat models from meta, which are trained to turn PDF into md using the donut architecture

Alternately, we could amplify the gene in the entire population, so that allism becomes a disorder.

From DSM-2300:

https://www.fysh.org/~zefram/allism/allism_intro.txt

edit: "The underlying trait that makes people allistic is a dysfunction of the parts of the brain dealing with emotion. Allistic people lack the capacity to independently experience emotions. That is not to say they lack emotions: far from it, the allistic mind experiences emotions just like any other. The dysfunction is that the allistic person's emotional state is not determined by eir own thought processes but instead is borrowed from other people that are expressing emotion nearby. Emotional cues in tone of voice, posture, facial expression, and so on, cause the allistic person to automatically and unavoidably experience the same emotion being expressed."

As someone who has had success at turning people you wouldn't expect to like crocs, on to crocs, do you mind sharing: Who are the groups you wouldn't initially imagine to be against crocs who are?

I only have experience proslethyzing, I have little understand of who has fallen.

I'm just curious if anyone is using unpoly instead of HTMX?

I know that HTMX has a much larger userbase, its main dev is well known and beloved, and the landing page doesn't look web 1.0

With that said, unpoly appears to have a similar feature set and one feature that HTMX doesn't: The ability to create modals without loading a new page.

How loud is it? Was special electrical needed?

With ICLR paper deadline coming up, I guess it's worth wargaming how GPT4 would review my submission.

Didn't your mother teach you not to prematurely optimize? You'll grow hair on your hands.

Do all variations of the model need to have the same architecture?

Or can they be different types of models with different number of layers, etc?

Is there a way to get "realistic" reverb using GPU computation?

This is fascinating. I realized that a model I have for evaluating relationships is basically another take on Albertus Magnus's model of friendship.

I'll start with Magnus:

Albertus, following Aristotle, distinguished between different types of friendship. He identified three main types:

Friendship of Utility: Based on mutual benefit, where individuals associate because they gain something from the relationship.

Friendship of Pleasure: Based on shared enjoyment or pleasure, where friends take delight in each other's company.

Friendship of Virtue: The highest form of friendship, where the bond is based on mutual respect and a shared commitment to moral goodness and virtue. This type of friendship is selfless and enduring.

Now, I'll run through my model, which I have applied to dating. I call the three dimensions: friendship, sex, and romance. You can have varying compatibility on all three dimensions, but in a sense it basically works out to Magnus's version above.

Friendship means shared interests. You like doing to same stuff and spending time together.

Sex is the pleasure compatibility vector in Magnus's formulation. i.e. how much fun do you have when you're together. Don't think of it just as sex, but also just like "wow this is fun and pleasurable"

Romance, people find this word tricky. I also don't think Magnus's idea of "virtue" hits it correctly either. It's basically a sense of tenderness and the deep-rooted feeling that you and the other person should have shared outcomes.

You can set up a Schwab brokerage account internationally and then use it to buy money market funds.

The same reason that after Jobs died, it was decided that iPhones should be large enough to fit in people's pockets. Because: many engineers in tech wear cargo shorts.

Can we get a 2-d visualization of topics, and drill into topics?

This is a really broad topic. I began studying it about 5 years ago.

Can you start by suggesting what you task you want to do? I'll throw out some suggestions, but you can say something different. Also you are welcome to email me (email in HN profile):

* Voice conversion / singing voice conversion

* Transcription of audio to MIDI

* Classification / tagging of audio scene

* Applying some effect / cleanup to audio

* Separating audio into different instruments

etc

The really quick summary of audio ML as a topic is:

* Often people treat it audio ML as vision ML, by using spectrogram representations of audio. Nonetheless, 1D models are sometimes just as good if not better, but they require very specific familiarity with the audio domain.

* Audio distance measures (loss functions) are pretty crappy and not well-correlated with human perception. You can say the same thing about vision distance measures, but a lot more research has gone into vision models so we have better heuristics around vision stuff. With that said, multi-scale log mel spectrogram isn't that terrible.

* Audio has a handful of little gotches around padding, windowing, etc.

* DSP is a black art and DSP knowledge has high ROI versus just being dumb and black boxy about everything.

I have had mixed luck with this model, which is supposedly state-of-the-art: https://github.com/magenta/mt3

What kind of music are you trying to transcribe?

Feel free to email me.

Backpropping filter coefficients is clever, but it hasn't really caught on much. Google also tried with LEAF (https://github.com/google-research/leaf-audio) to have a learnable audio filterbank.

Anyway, in audio ML what is very common is:

a) Futzing with the way you do feature extraction on the input. (Oh, maybe I want CQT for this task or a different scale Mel etc)

b) Doing feature extraction on generated audio output, and constructing loss functions from generated audio features.

So, as I said, I don't exactly see the utility of this library for deep learning.

With that said, it is definitely nice to have really high speed low latency audio algorithms in C++. I just wouldn't market it as "useful for deep learning" because

a) during training, you need more flexibility than non-GPU methods without backprop

b) if you are doing "deep learning" then your inferred model will presumably be quite large, and there will be a million other things you'll need to optimize to get real-time inference or inference on CPUs to work well.

Is just my gut reaction. It seems like a solid project, I just question the one selling point of "useful for deep learning" that's all.

For instance:

https://github.com/descriptinc/descript-audio-codec/blob/mai...

https://github.com/NVIDIA/BigVGAN/blob/main/loss.py#L23

https://arxiv.org/pdf/2210.13438 (the github repo doesn't include training, just inference)

It is INCREDIBLY common to use multi-scale spectral loss as the audio distance / objective measure in audio generation. They have some issues (i.e. they aren't always well correlated with human perception) but they are the known-current-best.

Given what I've seen in audio ML research:

1) Tuning hyperparameters of your audio preprocessing is a pain if it's a preprocessed CPU step. You have to redo preprocessing every time you want to tune your audio feature hyperparams

2) It's quite common to use torchaudio spectrograms, etc. purely because they are faster (I can link to a handful of recent high-impact audio ML github repos if you like)

3) If you use nnAudio, you can actually backprop the STFT or mel filters and tune them if you like. With that said, this is not so commonplace.

4) Sometimes the audio is GENERATED by a GPU. For example, in a neural vocoder, you decode the audio from a mel to a waveform. Then, you compute the loss over the true versus predict audio mel spectrograms. You can't do this with these C++ features. (Again, I can link a handful of recent high-impact audio ML github repos if you like.)

Again, I just don't get it.

If this is supposed to be used for deep-learning, shouldn't all the transforms be GPU-accelerated torch functions?

I prefer the term caffè corretto, since the coffee clearly needs to be corrected by applying booze.

Curious: Once the fonts are made, how are they marketed and sold?

Where do the people who use them become of aware of them and how?

Philosophy time:

Does this mean in 400 years it's possible we no longer disagree about how to evaluate things? i.e. we converge on one totalitarian utility function that everyone basically accept answers every possible trolley dilemma?

In 1600, people just took the world as that: measurements are sloppy, and vary culturally and based upon location etc. But we eventually came upon tools and techniques that are broadly accepted as repeatable and standard.

Would this sort of shift be possible? Or desirable?

It's the concept of a pfand system that involves a token.

The token is ridiculous and makes no sense. I've actively interrogated GPT4 on it and it becomes increasingly nonsensical trying to explain it.

Possession of the cup itself is proof in itself that I possess the item of value worth the pfand. The token itself is pointless.

Then can someone explain why the obscene and completely ludicrous "pfand" token system is so common still in Eastern Germany?