The Technology Vent and Rant Thread

[Renate]

I'm fairly hostile to the idea of abandoning my gray cells (such as they are) to AI.
Still, the thin end of the wedge comes in even without full trust in AI.
Just in the last year significant strides have been made and it will only get worse better?

I can't say that I envy anyone starting a career in this era.
OTOH, it's like all the stories of old employees trying to glide through to retirement while avoiding the "new stuff".

I recently googled a question and got an answer back that was germane, correct and not an angle that I had been thinking of.
OTOH, a bit of thought came up with a better answer.

[Quoth]

Originally Posted by Solitaire1

Why can't metadata tagging be more consistent? I just spent hours trying to do something as simple as putting the total number of tracks in each album, yet it works with some programs and not with others, and also not with some digital audio players. Worse, different audio formats have different tags for the same metadata. As an example, the tag for dates in MP3 files is not the same tag as the one for FLAC files.
…
Why does this have to be a problem?

…
Why should this be so difficult?

See XKCD Standards

[ratinox]

Originally Posted by Renate

Just in the last year significant strides have been made and it will only get worse better?

Most of the actual significant advances were made 40+ years ago. What the media and tech bros call "AI" today is... okay, brief overview:

Neural networks are software simulations of how living brain cells work. Neural networks are really good at building collections of patterns and identifying aberrations in those patterns. Credit card companies went all in with "AI" in the 1980s because when software can easily detect aberrations aka fraud? That saves a lot of money. Many of the most significant advances in machine learning came out of those neural networks.

When you feed images instead of credit card transactions to a neural network you get image recognition software.

When you feed words -- books, magazines, web scraping, etc., -- to an image recognition system you get a large language model or LLM.

And when you run an LLM in reverse you get what the tech bros call "generative AI".

It's not generative and it's not AI, and it's definitely not new. It's 40 year old tech run in reverse.

[Renate]

Originally Posted by ratinox

Most of the actual significant advances were made 40+ years ago.

Eh, whatever.

It's just that the scale of vacuumed-up human knowledge (sentence patterns) has increased to where it could be useful answering questions for grade school homework.

[Quoth]

Originally Posted by ratinox

Most of the actual significant advances were made 40+ years ago.

Back in the mid to 1980s 100% curated data from experts was used to build AI called "Expert Systems".

In the late 1990s ALICE was not much different to Eliza (1960s) as a chatbot, except was on the Internet. The flaw of those was the data was in the code, not separate. The so called "neural network" isn't new and dates to before WWII, when an array of sensors and lamps etc was demoed doing pattern matching. The current LLMs add storage outside of the program code. That's the only big change.
It takes a lot of processing, so much water and electricity is wasted.

Originally Posted by ratinox

What the media and tech bros call "AI" today is... okay, brief overview:

Neural networks are software simulations of how living brain cells work. Neural networks are really good at building collections of patterns and identifying aberrations in those patterns.
It's not generative and it's not AI, and it's definitely not new. It's 40 year old tech run in reverse.

Except it's not how biological brain cells work. The word "neural" is marketing. It's more akin to a dataflow diagram with storage at the nodes. It's certainly pattern matching.

[Quoth]

Originally Posted by Renate

Eh, whatever.

It's just that the scale of vacuumed-up human knowledge (sentence patterns) has increased to where it could be useful answering questions for grade school homework.

But never reliably.

[Renate]

Originally Posted by Quoth

But never reliably.

Correct.

But sometimes when you're flailing you can ask an unreliable source for an answer if you accept the caveat that it may be wrong.

A somewhat parallel:
You'd have to be a real expert to certify something like the Titan submarine was safe.
But a high school student could tell you that carbon fiber under compression is not the strongest thing in the world.
It all goes back to "The Emperor's Clothes".
Even an idiot can give sage advice sometimes.

OTOH, I keep seeing brainless things like this:

[Cactus Chef]

Originally Posted by Solitaire1

Okay, I'll try ranting.

Why can't metadata tagging be more consistent? I just spent hours trying to do something as simple as putting the total number of tracks in each album, yet it works with some programs and not with others, and also not with some digital audio players. Worse, different audio formats have different tags for the same metadata. As an example, the tag for dates in MP3 files is not the same tag as the one for FLAC files.

Why does this have to be a problem? Just enter the total number of tracks and it should work. I'm at the point I'm going to just give up on the whole things because it seems pointless.

*Rants to the Heavens*: Why should this be so difficult?

Can't speak for FLAC specifically, but in general, many of these things (particularly in the FOSS realm) started off as someone's pet project to scratch a personal itch, which means they're opinionated. So saying, "Why can't you do it the way MP3 does it?" isn't likely to gain traction because the author might say, "I don't like the way MP3 does it, that's why I wrote FLAC!"

Broad consensus usually comes later, once the original maintainer gets past his "pet project" phase and the tool starts to get traction.

I looked at the official FLAC spec, but oddly I don't see any mention of the date field in the metadata:

https://datatracker.ietf.org/doc/htm...tf-cellar-flac

[pdurrant]

Originally Posted by Cactus Chef

I looked at the official FLAC spec, but oddly I don't see any mention of the date field in the metadata:

https://datatracker.ietf.org/doc/htm...tf-cellar-flac

"For a more comprehensive list of possible field names suited for describing a single work of music in various genres, the list of tags used in the MusicBrainz project, see [MusicBrainz], is suggested."

So there's only a suggested list of tags.

[Cactus Chef]

Originally Posted by pdurrant

"For a more comprehensive list of possible field names suited for describing a single work of music in various genres, the list of tags used in the MusicBrainz project, see [MusicBrainz], is suggested."

So there's only a suggested list of tags.

Well, there's your answer then...it seems like it simply wasn't a priority to the FLAC developers.

Anecdotally, even though a large portion of my own collection is FLACs, I tend to go for separated tracks. But I feel like many of the FLAC enthusiasts will only distribute their rip as a single, monolithic FLAC with a CUE file to handle the track separation and pregaps. I guess they feel like they're better preserving the integrity of the original rip that way by altering it as little as possible. I wonder if perhaps that might influence the lack of attention to the metadata fields.

[Quoth]

And
https://www.thejournal.ie/chatgpt-ai...84432-Aug2025/

AI CHATBOT CHATGPT has begun sending users “gentle reminders” to take breaks during long sessions.

It comes as OpenAI, the operator of ChatGPT, has announced a rollout of behavioural changes in the bot – including how agreeable it is with users when making “high stakes personal decisions”.

OpenAI added that it seeks to improve how it measures “real-world usefulness” over the long term, as opposed to whether users “liked the answer in the moment”.

[Quoth]

Originally Posted by Renate

OTOH, I keep seeing brainless things like this:

Now I need brain bleach.

[Renate]

Originally Posted by Quoth

Now I need brain bleach.

And well you can say that!
They have the probes backward on the DVM.

[ratinox]

Originally Posted by Quoth

Except it's not how biological brain cells work. The word "neural" is marketing. It's more akin to a dataflow diagram with storage at the nodes. It's certainly pattern matching.

I think we're quibbling over semantics: neural networks are simulations of the structure and function of living brains, nodes and edges, neurons and synapses. It might be marketing today -- but that's what tech bros do: market things.

[Quoth]

Originally Posted by ratinox

I think we're quibbling over semantics: neural networks are simulations of the structure and function of living brains, nodes and edges, neurons and synapses. It might be marketing today -- but that's what tech bros do: market things.

A computer Neural Network and biological brains have almost nothing in common. It's a label applied to make it sound more "AI"ish. So pure marketing nonsense.