tristanc

They mention specifically fixing this in 2.1.14:

https://github.com/anthropics/claude-code/blob/main/CHANGELO...

> Fixed a regression where the context window blocking limit was calculated too aggressively, blocking users at ~65% context usage instead of the intended ~98%

This is one of the greatest news I've ever heard for the digital preservation community. Just so many projects over the years could have used resources like this. Thank you for contributing to humankind!

I'll bite, the article mentions:

> To calculate the wind speed on WASP-127b, the team tracked how fast molecules in the planet's atmosphere moved using the 'Very Large Telescope' (VLT) located in the south American country of Chile.

Could you provide anymore insight into how this was achieved? Any interesting challenges associated with the measurement? What are the practical limitations of this methodology? How did you pick the planet as a target to measure?

None of these questions are necessary, I'm just genuinely curious if you could provide any insight

Location: Washington D.C. and metro area

Remote: Preferred

Willing to relocate: no

Technologies: C++, Linux, Python, Rust, ML (PyTorch), 3D graphics programming (shaders, data processing, rendering, ...), Multimedia (FFMPEG, GStreamer, ...), Git, etc...

Résumé/CV: http://tristancharpentier.com/resume.pdf

Email: tristancharpentier[at]protonmail[dot]com

Hi! I'm Tristan, a recent graduate in C.S. from Sorbonne University in Paris but now based in the Washington D.C. area. Lately, I have been working on interactive 3D interfaces for music discovery with personalized recommendations using neural nets, pixel shaders for rendering weather forecasts, and real-time lidar processing and shading for augmented reality using a Kinect.

My projects are driven by intellectual curiosity, sometimes originating from conversations with my friends. I enjoy doing the research to achieve unique results. Here are some of my projects: http://tristancharpentier.com/projects.html

GitHub: https://github.com/hashFactory/

Youtube: https://youtube.com/@tridostudio

Location: Washington D.C. and metro area

Remote: Preferred

Willing to relocate: yes

Technologies: C++, Linux, Python, 3D graphics programming (shaders, data processing, rendering, …), Multimedia (FFMPEG, GStreamer, ...)

Résumé/CV: tristancharpentier.com/resume.pdf

Email: tristancharpentier[at]protonmail[dot]com

Hi! I'm Tristan, a Recent graduate in C.S. from Sorbonne University in Paris but based in the Washington D.C. area. Lately, I have been working on interactive 3D interfaces for music discovery with personalized recommendations using neural nets, pixel shaders for rendering weather forecasts, and real-time lidar processing and shading for augmented reality with a Kinect.

My projects are driven by intellectual curiosity, sometimes originating from conversations with my friends. I enjoy doing the research to achieve unique results. Here are some of my projects: https://tristancharpentier.com/projects.html and my github: https://github.com/hashFactory/

Hi atdrummond!

I'm sorry I missed your earlier thread I had no idea you were doing and was a little stumped but I think you are an amazing person for doing this.

I'm writing because I think I could use your help!

I'm a recent graduate from University in France (Sorbonne University) with a degree in CS. I recently moved back to the states to my parent's house to look for a job but it's been difficult to get an offer.

I'm passionate about CS and have been working on personal projects while waiting to land an interview. Most recently I've been working on projects like creating novel interfaces for exploring music by analyzing audio with a neural net and plotting in 3D according to genre classification.

(demos from my project here: https://www.youtube.com/watch?v=1UyjFeLjGJs&list=PLvzGE7O7Di... )

I've been having a great time developing this and my friends are all encouraging but it's been difficult to convince a company to hire me or a way to monetize this.

Furthermore, I only have access to an old family computer from 2013 so have been limited to doing audio processing and hosting on cpu.

I would be eternally grateful and appreciative if you would do me the favor of helping finance a PC with reasonable specs such as a GPU and proper storage capacity (stuck on a 512GB SSD right now...) to help me develop my projects further!

I would only ask for ~$800 and I could finance the rest but this would be a huge huge huge help!

I still can't believe someone would be so generous as you but I know HN is full of thoughtful people so I'd love to discuss!

I'm not sure which transfer method is best yet but I have a bank account in the US and paypal so I'm sure we could make something work.

Thank you so much!!!! My name is Tristan btw (: I am located in virginia and my email is tristancharpentier@protonmail.com

Use any model trained on the AudioSet dataset. There is one called EfficientAT i think that I use regularly and is pretty reliable

AFAIK, there is only support for AV1 decoding in the latest Snapdragon 8 Gen 2 and Apple's A17 Pro. No mobile SoC supports AV1 encoding (so far).

  Location: Washington D.C. and metro area
  Remote: yes or on-site
  Technologies: C++, Linux, Python, 3D graphics programming (shaders, data processing, rendering, …), Multimedia (FFMPEG, GStreamer, ...)
  Résumé/CV: https://tristancharpentier.com/resume.pdf
  Email: tristan_charpentier[at]hotmail[dot]com

Hi, I'm Tristan. I am passionate about math and CS and graduated with a B.S. in CS from Sorbonne University in Paris (June 2023).

Recently I have been working on interactive 3D interfaces for music discovery with personalized recommendations using neural nets, pixel shaders for rendering weather forecasts, and real-time lidar processing and shading for augmented reality with a Kinect.

My projects are driven by intellectual curiosity, sometimes originating from conversations with my friends. I enjoy doing the research to achieve unique results. Here are some of my projects: https://tristancharpentier.com/projects.html

Woah, is this suggesting that the cumulative human-years experienced by current living US population will add up to the age of the entire universe?

Mind-boggled again!

Interesting, I attempted to do the same as you but stopped just shy of BPM matching.

However I did get sound similarity working using an audio tagging neural net [1]. I chopped off the first and last 15 seconds of every song in my collection and ran them all through this analysis which produces a ~520 dimensional vector. I then targeted specific endings I wanted to match and used Euclidian distance to find the closest matching song beginning.

YMMV but I thought it actually worked pretty well, I just never got to automating the BPM matching. I can try to look for my old script if you're interested :)

[1] https://github.com/fschmid56/EfficientAT

In a similar situation as you, I’ve been having lots of luck using phind.com . They advertise as “the AI search engine built for developers” and its interface is a blend of google and chatgpt. You can also paste in long snippets of code or provide specific URLs you want it to use as a part of its answer. It then generates a few possible solutions with references.

For me it’s consistently been much more helpful than google.

An entire thread about apple rankings and no mention of Adam’s apples?

https://adamapples.blogspot.com/

I’ve been having tons of fun messing around using the depth sensor data from a Kinect for Xbox to create procedural visuals.

It’s been growing and growing to the point of now having 4 or 5 different input methods and accidentally discovering new emergent effects based on different combinations of parameters.

Also threw in intel’s 3D rendering engine Embree because why not.

Anyways, this playlist documents the progress in chronological order...

https://youtube.com/playlist?list=PLvzGE7O7DizdvCEOrwaVNcY_o...

For what it’s worth I followed this [1] tutorial about a month ago to jailbreak my 5th gen kindle paperwhite and was able to get a root shell and install and run python3. There is also a reader app which is much, much nicer (imo) than the native interface.

[1] https://www.mobileread.com/forums/showthread.php?t=346037

I’m pretty sure this just means it’s rated as appropriate for ages 9 and up. Had never noticed the ambiguity though...

Very neat project. With the new PinePhone case enabling LoRa communication as well, it seems like a lot of interesting possibilities are opening up.

The PNG format already allows grayscale 16bits / channel images. I regularly use this when rendering out depth maps from blender and ffmpeg seems to handle reading from these PNGs just fine (detecting the gray16 pixel format).

However I don’t know of any DEM tile apis that provide these sorts of PNGs but it sounds like a fun project!

Edit: I found this StackExchange post which shows how to generate 16-bit PNGs with gdal https://gis.stackexchange.com/questions/246934/translating-g...

Just a note for those of you who may be in a similar situation. YMMV.

If you have a couple of very large folders (even 1 TB or more) you care about holding onto from your unlimited drive, share it with a few (even throwaway) personal Gmail accounts. Once my HS disabled access to my account after I graduated, the school admin account somehow inherited these folders (the organization name is the new owner).

I still have access to these many TB large folders many years on in the “Shared with me” section of gdrive.

Could somebody please explain to me how TicketMaster on the iOS App Store manages to supposedly have 1.8M(?!) reviews averaging 4.8 stars when all of the written reviews I’ve scrolled through have 1 or 2 stars?

Do they trick people into reviewing the app favorably or something else?

I had a terrible experience with their app and don’t understand how they’re managing to game the App Store rating system.

You might be looking for ProjectM [1]. Reimplements WinAmp’s Milkdrop visualizer plugin to run standalone on practically every modern platform.

[1] https://github.com/projectM-visualizer/projectM

Does adding a header-only (C++) http server to projectM’s visualizer [0] count? I added a very basic http server to switch presets from any basic web browser (including a kindle paperwhite) using a static html page. [1]

It’s been great fun hosting parties and projecting visuals across the room onto an opposite wall. Then passing around a very old Android phone to go through the presets.

I highly encourage others to do the same with their favorite applications it’s fairly straightforward and makes it a pleasure to use.

[0] https://github.com/hashFactory/projectm

[1] https://github.com/hashFactory/projectm/blob/compression/src...

Not directly related to this project but in a similar spirit.

Recently, while exploring france’s national geography department (Institut de Géographie Nationale [1]) website, I came across a page [2] which allows anybody to download a Minecraft world representing any 25km^2 area of France.

The page lets you set the precise center and many other parameters such as whether to include buildings or even stylistic options.

The maps are generated on the fly and make use of a combination of the department’s own datasets (digital elevation, road networks, land parcels, building dimensions and height, and even forests).

I wish they would release whatever code they use for the terrain generation as I’m sure the open-source community would enjoy extending it with OpenStreetMap data!

[1] https://www.ign.fr/

[2] https://www.ign.fr/minecraft/

This is a fantastic idea! I forked [1] projectM a while back to work on a few modifications such as screen and audio recording (by saving raw framebuffers) and being able to select the active preset remotely from a web interface so I became somewhat familiar with the way projectm handles A/V input/output.

I just took a look at ffmpeg’s video filter interface and it looks like linking the two could be doable!

This feels like a much more elegant solution for recording frame-perfect captures so thank you!

I’ll definitely be looking more into it.

[1] https://github.com/hashFactory/projectm

You have to make sure to either feed it a microphone input or research how to set up a loopback monitor with pulsaaudio/pipewire/alsa (whichever you use). This way you can feed it whatever the audio output from any application. Then you many also need to specify an environment variable in your shell (I think it’s something like SDL_AUDIODRIVER I’m not sure and you set it to the specific driver you’re using) in any case you should be able to find instruction with “pulseaudio loopback stereo monitor sdl”

(Apologies for formatting i am on mobile)

I recently had to set this up on a raspberry pi 4 with manjaro / KDE plasma and it was a bit of a pain to setup (the setup is slightly different depending on your exact audio configuration) but it’s very rewarding once you do.

Feel free to PM as I spent a lot of time debugging the setup and may be able to help!

It’s not too often this comes up ... but here is an iOS app for finding drinking water fountains and public restrooms using Overpass’s OSM API.

https://github.com/hashFactory/ImThirsty

I feel like this comes up again and again. For some people Instagram or Facebook are sometimes the only way to communicate with new peers without introducing too much friction. In France, for example, most of my friends will exclusively use Instagram when connecting with new people.

Seems a little biased since they don’t even talk about its advantages. AFAIK not many companies use OCaml in practice but it is used quite a bit in academia in Europe.

That said (I may be biased as well) I was taught OCaml in uni and enjoyed its succinctness and expressiveness.

Unfortunately seems unavailable in Europe...

Very interesting and made me chuckle. Thank you.