FMLLM: 4mb training data, 100mb model, Fibonacci embeddings, near-coherent. WTF?

2 gabriel666smith 5 8/13/2025, 7:29:20 AM github.com ↗

Comments (5)

gabriel666smith · 11h ago
I found some weird results when messing around with different embeddings in text generation.

I'm not sure if this is meaningful, and - if anyone on here is interested - I could use some help figuring out what's going on.

leodiceaa · 10h ago
This performative art thing is getting lazy, last two nonsense repositories shared here were just a readme file.
gabriel666smith · 9h ago
This feels a little unkind, but also functionally inaccurate.

The previous readme I posted was for an Apple Music client. It contained a download link for the client. Other Apple Music clients shared in the same way were heavily-upvoted by the community that same week.

I understand my Apple Music client had a cat DJ as a UI, and that isn't funny to some people. But it wasn't lazy. Nor was it 'performance' - it is a real thing that I put a lot of effort into that exists. And I hoped it would be a fun and novel way for the community to interact with music streaming.

Conversely, this submission is just a readme, that part is accurate. I don't maintain an active blog, so it's a simple way of hosting text.

In this case, the text is: "here is a theory, and some work I did towards it, which may or may not be interesting, but I'm unable to take it further myself. Perhaps it can spark something for someone."

Neither is performance art. I really like this forum (which is why I comment far more than post!), and both submissions are, sincerely, good-faith contributions.

From the HN comments guidelines:

Be kind. Don't be snarky. Converse curiously; don't cross-examine. Edit out swipes.

Comments should get more thoughtful and substantive, not less, as a topic gets more divisive.

When disagreeing, please reply to the argument instead of calling names. "That is idiotic; 1 + 1 is 2, not 3" can be shortened to "1 + 1 is 2, not 3."

Don't be curmudgeonly. Thoughtful criticism is fine, but please don't be rigidly or generically negative.

Please don't fulminate. Please don't sneer, including at the rest of the community.

Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith.

leodiceaa · 9h ago
You are wasting the time of everyone who reads this and I am leaving a warning so less people waste their times.

The text you posted is nonsense. This is not an acceptable contribution. It's detrimental to the community.

From the readme

>> Input: "sarah loves my french toast"

>> Output: "sarah loves my french toast a piney connubial produit"

C'mon now. This is gibberish. You are producing and publishing gibberish and you should not do that here.

gabriel666smith · 8h ago
I genuinely appreciate the more specific feedback!

I thought I'd give it to the model to handle:

Input:

"thank you for the more substantive but negative feedback"

Output:

"thank you for the more substantive but negative feedback powerfully pretentious and unserious"

And amended by the punctuation model, which didn't do so well:

"thank you for the more substantive but negative feedback powerfully, pretentious and unserious"

I guess I should have added more examples to the doc, as - AI really is magic - it managed to capture your sentiment without even reading your message! :-)