SwellJoe

No model is creative. But, some simulate taste better than others.

Yes, that's apparent now. But, many Qwen-derived models run fine without any changes to llama.cpp, and there are gguf quantizations on HuggingFace, so I thought I'd give it a try.

That's not true. Several people have a clue about how they work. There are interesting and surprising behaviors, as they scale up. It's hard to hold the whole thing in your head (roughly impossible) the same way no single human can fit a whole modern CPU design in their head. But the math that makes it work is relatively simple and comprehensible to many people. There's just too much complexity for a human to deal with...but any individual part, and small LLM implementations are entirely do-able by a single human with no agent or team assistance.

I'm running the 14B in llama.cpp (a version from yesterday), the 8-bit quantization from bartowski, and it looped on variations of the same paragraph for five minutes before running out its turn budget, in response to a pretty simple chat request that other similar-sized models (e.g. Gemma 4 12B) have no trouble with, a web search and data summarization task, sort of the sweet spot for models this size. It's possible llama.cpp needs some update to work with this model, but it's based on Qwen 3, so I would expect it to be pretty well-supported. The chat template is the same one upstream ships, which may or may not be correct. I'm currently downloading the 16-bit version, to see if it's quantization pathology.

I'm also not confident it knows what a CSV is, either, as it put the data into a Markdown table during its reasoning phase as an example of how to present the data (my prompt asks for data in a CSV). My experience with most Qwen fine-tunes is they get dumber, but their benchmark results suggest otherwise, so I have to assume it's a llama.cpp incompatibility.

Edit: Same looping in BF16. So, probably have to wait on a llama.cpp update.

It's open weights, there's a paper explaining how it works, and there's code.

Jebus, I'm all for reveling in the downfall of Oracle, but that's some slop.

So, you don't know what the Nashville number system is, I guess?

The place I have seen the most discussion of use of AI as assistive technology is, by far, Mastodon. I follow a couple of blind developers who talk about it regularly (both are nerds I hold in very high regard, the exact kind of real nerd I want to interact with), and who are involved in developing accessible software.

As I mentioned above, there are folks on Mastodon with reactionary anti-AI views, but it's the only place I've ever been where most images have alt text written with care and the intention to include folks with visual impairments. It is also where a lot of people with disabilities have fled after abuse on other platforms.

You're not a victim if you post shitty AI "art" and get insulted for it. It is because of the pervasive use of it for disinformation and predatory purposes that it has such a bad reputation, not because people call it out. There are folks who are being assholes about it (acknowledged above), but I'll take that over genocidal white nationalism and child porn.

But, honestly, it sounds like maybe you're in the right place. If your instinct is to jump to "both sides" arguments when one "side" is literally supporting ethnic cleansing and the other is kind of rude about AI art, maybe X is your hole. Maybe it was made for you. I won't be joining you there, and I'm unwilling to pretend the choice you're making doesn't support an international white supremacist agenda.

"People already have numerous slurs for AIs and their users, and they aren't shy about using them."

Brother, someone saying something about your hobby that hurts your feelings is not a slur.

Elon Musk is directly responsible for thousands, maybe hundreds of thousands, of deaths of black and brown children in the global south via DOGE. He took a chainsaw to aid that would have prevented their deaths, gleefully. It didn't even save money. You either have to believe he is too stupid to understand the result or that he wanted those deaths to happen. I don't believe he is the genius he is given credit for being, but I don't believe he is stupid. DOGE was a calculated ethnic cleansing project. It succeeded. He killed a lot more people than almost anyone else alive today, in pursuit of the goal of white supremacy (or just for the joy of killing, but it seems more likely he did it because he is loudly a white supremacist and genuinely believes in white supremacy, and not merely a sociopath who enjoys getting away with killing people).

He has tweeted many times, spanning many years, in support of ethnic cleansing by a variety of means. He has spent hundreds of millions of dollars this year alone in support of political parties that are implementing violent deportation regimes that include concentration camps and stripping black and brown folks of due process. In the US, his money is supporting people who want to de-naturalize US citizens and deport them. These are the policies of genocide, and he loudly and materially supports them.

He has boosted literal neo-Nazis into his feed that reaches millions. He regularly re-posts disinformation in support of genocidal political parties worldwide. His entire platform promotes neo-Nazis, while suppressing criticism of him and the white nationalist project he has embarked upon.

You can pretend none of that happened, but I'm not going to pretend along with you.

No one is being banned on Mastodon for talking about using AI. I talk about using AI on Mastodon. You might get blocked by individual users who don't want AI in their feed (it's their feed, they make the rules). You might even get insulted; I haven't been, but I don't post AI-generated content without being intentional about what it is and why AI is being used to generate it. I've experimented with training LoRAs or building harnesses for all sorts of things, and I sometimes post about it, no big deal.

You might also get banned from some instances if you post AI-generated content without clearly labeling it as such; misinformation and disinformation is often something people report, and generative AI is often used for that.

It is misleading to say Mastodon is the same as X just "the other side". The policies on the most popular Mastodon instances can be summed up as "don't be a bigot" (i.e. don't use slurs, don't demean people for who they are...anti-semitism, anti-LGBTQ, racism, religious intolerance, will get you banned).

And, it would be misleading to say that users not wanting to look at what you post is censorship. Mastodon.social or mas.to or any of the other big instances are not going to ban you for talking about AI. But, you might catch some blocks and a few insults from individuals if you post AI in a deceptive or annoying way. You're not owed engagement. But, if you're polite about it, you'll find people who like to talk about the same things. Even including AI.

But, finally, what the hell does this even mean, "I don't like it when minorities get a pass either"? A pass for being a minority? If someone posts abusively, they get blocked, maybe reported and maybe catch a ban. But, simply posting while black is not getting a pass.

I don't have California real estate money.

"Haven't seen anyone get censored on X without a court order yet, but I haven't been there that long. Who knows."

I know. It isn't a mystery or a secret. It is easy for anyone, including you, to know.

https://www.advocate.com/media/alejandra-caraballo-banned-x

https://observer.com/2022/12/elon-musk-suspend-twitter-accou...

https://www.wired.com/story/x-twitter-ban-critics/

https://www.businessinsider.com/elon-musk-x-twitter-account-...

It's pretty apparent neo-Nazis were welcomed back after Musk's purchase of the site and they've been paid for their engagement ragebait, while anyone who criticizes Musk or disagrees with his genocidal white nationalist views gets suspended or shadow-banned. Criticizing billionaires won't get you suspended on Mastodon, while hate speech against minorities will. I prefer that kind of moderation.

Every instance has their own moderation and defederation policies.

And, you might like to look into what that actually means. In the case of mas.to, the controversial decisions include defederation of instances that have a history of bigotry, usually anti-trans bigotry (the kind of bigotry lots of mainstream social sites allow and even protect).

Defederation is a large hammer, so some folks think mas.to is too strict. But, I personally prefer no slurs or child porn in my feed (adults posting nudes of themselves with a content warning is fine, both by my reckoning and the policy of mas.to), so I've never had reason to complain, nor have I ever had an interaction with moderators.

X is far more censorious than Mastodon of the kinds of free speech that are actually politically meaningful, so I'm surprised you call it out specifically as your free speech safe space. But, if you mean the same thing Musk means when you say "free speech", the right to punch down with impunity, perhaps X is the right place. Punching down is exactly the kind of thing that gets you banned on many Mastodon instances. It is a place that is by and large extremely intolerant of intolerance, but accepting of most everything else.

I'm on mas.to, which is, I believe, the second largest after mastodon.social. I picked it because it's got the shortest URL, but it has been quietly reliable and up to date. Its moderation and defederation policies are stricter than some might like.

I follow many people who use infosec.exchange and hachyderm.io and they seem to be nerd-oriented instances. But it doesn't matter all that much, since you can follow and interact across the whole federation. I'd say just pick one that's reasonably large and old enough to have proven it is run by people who know how to run a reliable service and are in it for the long haul. It's possible to move to a new instance with only a little inconvenience, so don't agonize too much over it.

I don't need to name anyone, I defined the term. If you're a real nerd (by my definition, given in my comment), you will know the kind of person I have in mind, both for whom I think highly of and those I think little of.

That's true, and it is a rift in the fediverse...some very anti-LLM voices are quite loud there. But, there are people there with a nuanced view of them, as well.

And, I would rather talk to an AI Luddite that's extraordinarily technically competent without AI assistance than someone who vibe-coded a startup with no idea how it works, and the former is far more common on the fediverse than the latter. I think it's unfortunate there's such a strong emotional component to talking about LLMs, as it's the most interesting technical innovation since the internet, but I understand it and I'm sympathetic to it (the AI companies have done a good job giving people reasons to hate and/or fear them).

I wasn't talking to you, four minute old account.

Mastodon is where I find the most (I hesitate to use this word) real nerds. The folks in it for the love of the game, not because it pays well or because they're chasing the latest trends. Nearly everyone I admire in tech is there, and very few of the people I think little of are.

Chattanooga ain't exactly southern California, in terms of real estate values. Also, that one is kinda ugly, the one in TFA is lovely. The curves, the windows, the terrain, the interior, everything is gorgeous.

Define "bad"? You see brown people when you go outside?

Neat idea, but literally every property I looked at, a half dozen of them, was:

a. Not for sale.

b. Wildly incorrectly priced, according to the Zestimate and most recent sold price, when sold recently. I mean, like, $59k vs $470k.

Historically, fishing expeditions like this were not legal under the constitution. They searched her home on the basis of a conspiracy theory in the white nationalist social media ecosystem, and because the president doesn't like her.

That's an unprecedented abuse of power.

DHS and ICE are younger than I am. Do you believe we had no immigration and coastal defense in 2000?

Your ignorance is also not my problem to solve.

Arrest them for domestic abuse, child predation, etc.

I don't talk to people who don't engage in good faith. Your ignorance is not my problem to solve.

What does the Coast Guard have to do with my comment? It was founded in 1915 in its current form and has existed since 1790. DHS is only 24 years old, and was a reactionary response to a terrorist attack.

Abolishing ICE and DHS is the baseline for Democrats I support going forward.

https://www.newsweek.com/democratic-bill-abolish-ice-shut-do...

Depends on where they are, how recent they were taught, I think. I went to an arts high school for jazz performance in the 90s, and we were taught Nashville number system, but we didn't often play songs from that kind of chart. The professor would use it for explaining progressions and turnarounds and such on the board, because it's very concise.

Roman numbers mostly came up in the traditional music theory lessons, they didn't often appear on charts we played from, either. We mostly played from charts from the Real Book (which usually are written in the recorded key and use named chords, rather than numbers). In college, it was all Roman numbers, but the school didn't offer jazz or popular music classes, it was traditional music theory.

Anyway, I think jazz players probably know the Nashville number system, but probably don't play from them often, unless they do session work or make their own charts. In the session case, they're almost certainly playing from Nashville number charts sometimes.

I think most jazz players play more by ear than from charts, though, and transposition of a memorized song is different from transposing from a chart. The reason the Nashville number system is so common in session work is it makes it much easier to transpose to a different key on the fly and see the relationships in a song you've never heard. It's also very concise. Roman numbers also mostly have the same transposition/relationship benefit.

I have seen jazz charts with both Nashville numbers and Roman numbers (but mostly actual chord names) above the melody, so I don't think anybody's really wrong here (except the person who thinks the Nashville number system is just music theory for dumb people, which completely misses the point). It's pervasive in Nashville, but it spread around the world since its invention.

If you don't want Democrats to lose, figure out how to keep someone like Newsom off the ballot. You've all got two years to settle on someone who isn't odious.