jmpman
jmpman on Hacker News
Asked Google for "movie times". Musk wasn't listed, even though my local theater was showing it. Only when looking at the specific theater, did it show me the movie times.
Magnesium glycinate is what worked for me, but at the peak of my insomnia, I would have benefited from something other than Ambien.
Are the Russians attacking US planes or ships? If they want to cross that line, I expect the US has plenty of ways to ensure a proportional response.
I'd read that there are a few snack items resistant to GLP-1 induced food aversions, when I was on a GLP-1, Biscoff cookies felt like one of them. I may have an Amazon scheduled delivery of Biscoff, which every month, I intend to cancel, and somehow forget to follow through.
I never learned Solidworks or any other "proper" CAD packages and have always hacked my way to success with OpenSCAD. Fundamentally there are tons of shortcomings, but where it's recently become a joy is with AI. I can have Claude design something in OpenSCAD, and then I'll copy and paste it over to OpenAI, asking for it to review. A few back and forth cycles and the resulting output is surprisingly solid. If there's something I don't like, I'll just make a few edits, and send it back to the AI to parameterize (and critique). Once I have a good first draft on OpenSCAD, I'd like the AIs to then convert it to a true CAD package to productionalize... haven't achieved that yet.
Cool system. I was thinking a standard interceptor drone, but with a single shotgun barrel, and single round.
In an economic battle, I continue wondering why Ukraine has interceptors which detonate instead of being armed with projectiles. High speed cameras and low latency bridgewire ignition, coupled with AI targeting should enable single shot targeting with high probability of success. Multiple reusable interceptors seems like a better solution than expendable. At the same build rate, the number of interceptors per incoming drone will grow over time, instead of maintaining equivalence.
Was this "tent city"? I recall "tent city" was a Joe Arpaio creation that was ultimately closed down in 2017 after he lost reelection as sheriff. My closest experience with "tent city" was when a relative got busted for DUI and ended up spending a weekend there. Was quite the deterrent, as it should be, and his drunk driving was dramatically curtailed. Ultimately that's good for society.
Why didn't they select a site closer to the natural gas instead of 17 miles away? Ellison is wealthy enough, he could have bought his own gas field, and plumbed directly from there to the blooms.
I've had a friend pass from an OD, when consuming what he thought was cocaine, which was actually laced with fentanyl. I wonder if the Italian mafia has a more profitable cocaine market, and don't want it contaminated, as is likely to happen with fentanyl, as the cocaine market, which I assume is recreational, would contract.
I'm interested in seeing these results on cloud instances. I'd expect rather extreme differences in latency.
Thrilled to see your post. It was a fun find and had so many implications.
Thanks for the pointers. My goal - I really didn't like the Erik Satie Gymnopedie #1 version I'd just bought, and wanted to generate a new version. With the recent Astra release, I wanted to see how it and Fable did on creating their own versions based upon a standard MIDI. The encoding was actually just to make the PianoDisc player work better. Their iPad app will actually play MIDI directly without encoding, but they feel like second class citizens in the app, and there's some weird Bluetooth audio and a second Bluetooth MIDI interface on the app. I was having problems with the MIDI one, but the mp3 encoded connection seemed to work reliably.
When Fable came back with the decoder, I knew something was off. The decoy notes... even more evidence of wonkiness. Mostly posted it to HN out of disbelief that Fable just hacked my piano and had I blindly pushed it to a public repo, I would likely have broken some law.
Btw, the version I like is performed by Pascal Roge back in the 1970s. The AIs don't quite match his performance, and I keep trying to figure out how to guide them. I don't think his version has a delay between the melody and the chords, while the AIs were both convinced that was desirable. They seem to be lacking some taste in that area, but overall I prefer their version to the one I bought.
I haven't tried that, but when doing some quick discussion on the topic, Fable pointed me to some other models which do it better.
My task to fable was to take the MIDI and encode it into some proprietary mp3 format. It's still "binary" MIDI encoded in the mp3, it's not simulating the piano and generating waveforms... yet.
The encoder wouldn't be a concern if it didn't have details on the decoy notes - and I can easily ask fable to remove them. But, I'm not looking for lawsuits for what is really my wife's hobby.
That's a great data set. The main problem with it is the file labeling. All the details are in the external metadata file. I had Fable convert those MIDI into the PianoDisc mp3 format, and embed the metadata into the id3 tags. Should import into their application natively, with albums etc.
I had performed a quick GitHub search and not found any code. It certainly didn't have these decoy notes documented on its own. It only found them because I provided it a known Erik Satie piece, which has a specific number of notes, and it dug in to figure out what the excess notes were.
I'd found this code as well, and in a previous chat, Fable pointed me to it but first, I have a $100/month Claude subscription, and typically just barely run out every week, so I throw tokens at just about anything. Second, I'd come across that code, but it's windows software, not open source. Third, the encoder isn't anything exciting, and not really a concern except that the Fable version offers the ability to include the decoy notes. Finally, I literally just wrote "Yes, please write that encoder" when fable asked if I want it to "Bigger picture: since the Prodigy plays this format over Bluetooth, an encoder that writes your own MIDI into the right channel this way is probably the missing link for the MP3-to-piano project. I can build a test file if you want to try it."
Yes. Made a type and after I noticed it was too late to edit my post.
The funny thing is - I'm virtually tone deaf and can't play the piano. It's my wife's piano, but when we bought it, I had the player system installed, so I could get some enjoyment out of it.
Thanks. It was too late to edit after I noticed my typo.
I might just email the company. If they object, I won't make my GitHub repo public.
Posted by
jmpman on Saturday September 05, 2026 @10:54AM
from the self-post dept.
I have a self playing piano, using a system called PianoDisc Protigy. They have an online store which sells music for their system, from various modern artists along with classics such as Bach and Beethoven. Last night I saw they had released some music from Eric Satre, a 19th century French composer, which I bought. Curious if I could have just used AI to create these files, I began...
No, I was asking about basic physiology research.
Hi Muhammad,
Great paper. Enjoyed seeing the RAMCloud callout, as it was the first thing that popped into my head when reading it.
I've been thinking about a few other potential variants. Primarily the 3AZ scenario where you'd use Btrlog in 1AZ, and then send a replica to an S3-Express bucket in another AZ synchronously (a PUT to the local AZ endpoint does not incur a cross AZ data transfer cost, even if the bucket is in a different AZ), and then an async PUT to s3-std.
Love it, but when comparing the recorded version to the generated version, there are significant differences. Is this something which can be improved?
How was this study not done back in the 1950s?
But would you really put it past Russia?