Pinned Tweet
Haystack is a music discovery app that learns your taste from how your music sounds, not from its tags or likes. It has a server-side (docker) discovery service and a client-side ratings app that work together to do a better job of finding new music that you love.
github.com/kurtlieber/haysta…
Kurt Lieber retweeted
Gen X, how often do you get halfway through typing a response to a post, just to realize you don’t actually care enough to respond, and just keep scrolling?
I was puzzled why I saw such limited discussion of Claude Duet on X. Today, I was able to install and try it out for the first time, and now I understand. It's definitely a research preview and the actual utility of Duet is...limited.
Here's hoping for rapid progress!
2
3
100
Imagine announcing a new model and then making it available to your customers on THE SAME DAY.
There's a lesson in here for you @Google. You'll ignore it, of course, but it's a lesson nonetheless.
1
6
135
This is how @SpaceXSI chose to communicate cuts to Grok Bot account limits. No warning, no formal communication - just an "oh, by the way" reply from Elon.
It's like they looked at all the heat Anthropic and OpenAI took for their own account cuts and said, "wait, hold my beer"
Mind-bogglingly dumb.
103
Kurt Lieber retweeted
Anthropic is going to burn through a lot of developer goodwill if it keeps imposing its own ideas of what people should and shouldn't build.
The timing is especially funny given Roblox just reported that over 80% of its top 100 games used AI during development and Tim Sweeney is pushing AI across gaming, an industry where copying and iterating on existing ideas is practically the foundation of how new games get made.
Developers are paying for tools to help them build software, not for Anthropic to make moral and ethical judgments on their behalf, and the more dependent everyone becomes on these models, the more concerning that gets.
ANTHROPIC IS CRACKING DOWN ON VIDEO GAME REWRITES 🚨
Claude Code picked up a new rule around Oct 27, sent from Anthropic’s servers under a flag called "tengu_heron_brook". There’s no setting to turn it off.
The full rule: “Do build original games, games in a genre, and games from the user’s own designs and assets. Don’t build a recreation of a specific existing game, even if it is renamed or called original.”
Is this the end of game rewrites, modding and everything in between?
Kurt Lieber retweeted
I think I speak for everyone when I say "@AnthropicAI , fuck you and the horse you rode in on."
I don't personally care about doing video game rewrites myself. But the whole concept that some busybody can tell me what I may not do with the tool I am paying for is deeply offensive.
You will be replaced by open-weight and open-source models. Every asshole move like this makes you more enemies and brings that day closer.
ANTHROPIC IS CRACKING DOWN ON VIDEO GAME REWRITES 🚨
Claude Code picked up a new rule around Oct 27, sent from Anthropic’s servers under a flag called "tengu_heron_brook". There’s no setting to turn it off.
The full rule: “Do build original games, games in a genre, and games from the user’s own designs and assets. Don’t build a recreation of a specific existing game, even if it is renamed or called original.”
Is this the end of game rewrites, modding and everything in between?
A discovery engine is only as good as its feedback loop. haystack does not guess whether you liked a recommendation. It watches what you actually do.
Every hour, the sync job checks your Spotify plays, likes, and verdicts from its companion frontend, needle.
Like a track on Spotify or approve it in needle? That acoustic profile reinforces future positive matches.
If a liked track is missing from your local filesystem, haystack automatically routes it to a "To Rip" playlist so you know to acquire a permanent copy.
Leave a track unliked by the time next Monday's batch runs? Silence counts as a rejection. haystack logs the negative outcome, clears the slate, and sharpens the next batch.
github.com/kurtlieber/haysta…
83
Kurt Lieber retweeted
Heads up, @OpenAI and @AnthropicAI. Rumor reaches my ears that you're doing a stupid thing, and should stop before you damage yourselves.
I hear you've now nerfed your models so that when people try to use them to clean the spamware and spyware out of their televisions, you block that.
If you really want Chinese open-weight models and open-source projects like OpenWaldo to eat your lunch, crippling your models so that people paying expensive subscriptions can't use them for completely legitimate purposes is a great way to accomplish this.
If, on the other hand, you'd rather have a volume business, maybe stupid arbitrary restrictions that look like they have the sole purpose of protecting corporate sleazeballs aren't such a good idea.
A word to the wise should be sufficient.
How do you find a needle in a haystack without getting trapped in an algorithmic echo chamber?
Every morning at 05:00, haystack pulls candidate tracks from ListenBrainz and Deezer. It streams 30-second previews directly into memory, pipes them through ffmpeg, and generates audio embeddings on the fly. Nothing touches the disk, and rate limits stay strictly polite at one request every 1.1 seconds.
Scoring is mathematical and transparent:
Candidates are ranked by their closest acoustic match to music you already love.
They take a direct score penalty if they sound too close to tracks you have previously rejected.
Every Monday at 08:00, the top 15 tracks drop into your Spotify Discoveries playlist (capped at one track per artist, with 20% of the slots reserved for mid-score exploration).
github.com/kurtlieber/haysta…
1
1
83
Most music recommendation engines rely on collaborative filtering and messy genre tags. They recommend songs because strangers with overlapping playlists clicked like, not because the track actually fits what you want to hear.
I built haystack to fix this. It is a self-hosted music discovery engine for a single Spotify account running on an always-on Linux box.
Instead of trusting metadata, haystack learns your taste from the raw audio itself. It runs every track in your local library and your liked Spotify songs through Essentia's Discogs-EffNet model, analyzing up to nine 30-second windows per track. It maps how your music actually sounds, scouts new candidates, and builds a weekly Discoveries playlist.
github.com/kurtlieber/haysta…
2
4
158
Kurt Lieber retweeted
Amazing how quickly we went from "oh no LLMs are going to flood the Internet with insecure code" to "oh no LLMs have discovered the internet was already flooded with insecure code"
Dusting off this account to share some of my apps:
Up first, haystack: It learns your music taste from the audio itself, scouts new tracks by how they sound, and puts the best in a weekly Discoveries playlist. Requires a Spotify Premium account.
lieber.org/haystack/
1
4
146
Haystack pairs best with needle, which is the client-side component. needle plays the queues haystack fills, shows how each track reads against your taste, and hands your verdict back to haystack.
lieber.org/needle/
1
50















