There has been promising work on olfactory training, which you can do very inexpensively at home. If you can, I would consider seeing any ENT first to rule out polyps, etc.
Maybe at the bottom your marbles could land on surfaces with different accustic properties. Track selection would determine the surface and release time would determine the timing.
I live in a big city. DoorDash always delivers to an address 10 minutes away walking from my building. It is quite inconvenient in the winter. With human drivers, you can at least try to convince them to use Google Maps, but with an AI?
On second thought, prompt injection via delivery instructions?
Building Blocks was a pack-in when I bought Syntorial in last years sale. It's technically different (web-based instead of a standalone app like Syntorial), and tries to do with music theory as a whole what Syntorial does with synthesis. Since it goes through the same 'theory, practice and create' cycle every time (eg. once for every chord type) I found it getting a bit tedious after a while.
What I don't quite get is why manufacturers of midi controllers (Arturia, Novation, NI, etc.), with the exception of, possibly only Korg, don't release any of their digital instruments as mobile apps. After sitting the whole day in front of my computer, the last thing I want to do is to swap VS Code for Ableton or Kontakt and spend a few more hours in the glow of my monitors.
(I do get that if you are very serious about making music you need a proper computer set up. I am just a mere amateur hobbyist.)
This is why I bought a teenage engineering op-1. Yes it’s overpriced, yes there are definitely better little desktop synths but I can whip this one out and just go wild without having to look at a (real) screen.
Same reason I keep my Roland Fantom around - has everything built in to the device.
NI are probably more focused on DJ stuff at the mobile app level.
Arturia have tried a few different hardware-hosts over the years, but seem to be focusing on their Astrolab platform rather than supporting iOS (Android is a non-runner due to latency).
Novation's main offerings are about analog signal paths. Back in the early 00s they had a few weird integrations like the X-Station, but its the analog nature of the bass-station and subsequent *Brute line that maintained their USP and cachet. Things like the Circuit/Launchpad are obvious AIO attempts at taking the share from similar form-factor iPad sequencing and clip launching utilities.
The m8 and the recently, heavily promoted Woovebox 2 are the Emacs/Vim of grooveboxes. They hide a vast amount of functionality behind what initially seems like an impenetrable jungle of button presses and shortcuts; a system that ultimately proves* to be highly ergonomic.
*Based on what I read. Sadly, I don't own these devices.
I own an M8 (and also, happen to be a vim user lol) and it's sorta true, but it feels a lot more natural if you're familiar with the LSDJ-style workflow. LSDJ is a DAW built for the GameBoy so was working with very constrained hardware. Both are derived from sequencer workflows, but I'm not quite sure which or where those originated.
That being said, the M8 workflow isn't all that bad, but it's definitely less visual than most other DAWs.
> Why is it so hard to install PyTorch, or CUDA, or libraries like FlashAttention or DeepSpeed that build against PyTorch and CUDA?
This is so true! On Windows (and WSL) it is also exacerbated by some packages requiring the use of compilers bundled with outdated Visual Studio versions, some of which are only available by manually crafting download paths. I can't wait for a better dev experience.
Stuff like that led me fully away from Ruby (due to Rails), which is a shame, I see videos of people chugging along with Ruby and loving it, and it looks like a fun language, but when the only way I can get a dev environment setup for Rails is using DigitalOcean droplets, I've lost all interest. It would always fail at compiling something for Rails. I would have loved to partake in the Rails hype back in 2012, but over the years the install / setup process was always a nightmare.
I went with Python because I never had this issue. Now with any AI / CUDA stuff its a bit of a nightmare to the point where you use someone's setup shell script instead of trying to use pip at all.
Lets be honest here - whilst some experiences are better/worse than others, there doesn't seem to be a dependency management system that isn't (at least half) broken.
I use Go a lot, the journey has been
- No dependency management
- Glide
- Depmod
- I forget the name of the precursor - I just remembered, VGo
- Modules
We still have proxying, vendoring, versioning problems
Python: VirtualEnv
Rust: Cargo
Java: Maven and Gradle
Ruby: Gems
Even OS dependency management is painful - yum, apt (which was a major positive when I switched to Debian based systems), pkg (BSD people), homebrew (semi-official?)
Dependency Management is the wild is a major headache, Go (I only mention because I am most familiar with) did away with some compilation dependency issues by shipping binaries with no dependencies (meaning that it didn't matter which version of linux you built your binary for, it will run on any of the same arch linux - none of that "wrong libc" 'fun'), but you still have issues with two different people building the same binary in need of extra dependency management (vendoring brings with it caching problems - is the version in the cache up to date, will updating one version of one dependency break everything - what fun)
NuGet for C# has always been fantastic, and I like Cargo, though sometimes waiting for freaking ever for things to build does kill me on the inside a little bit. I do wish Go had a better package manager trajectory, I can only hope they continue to work on it, there were a few years I refused to work on any Go projects because setup was a nightmare.
The main NuGet was problematic for a long time e.g. not providing any control over transitive dependencies (like pip at the time). You had to use https://fsprojects.github.io/Paket/ if you wanted safe and consistent resolution. NuGet since got their act together and it’s not as flawed now.
I agree. Here are some things that I (a science researcher and professor) like about R and CRAN:
1. There are a lot of build checks for problems involving mismatches between documentation and code, failed test suites, etc. These tests are run on the present R release, the last release, and the development version. And the tests are run on a routine basis. So, you can visit the CRAN site and tell at a glance whether the package has problems.
2. There is a convention in the community that code ought to be well-documented and well-tested. These tend not to be afterthoughts.)
3. if the author of package x makes changes, then all CRAN packages that use x will be tested (via the test suite) for new problems. This (again because of the convention of having good tests) prevents lots of ripple-effect problems.
4. Many CRAN packages come with so-called vignettes, which are essays that tend to supply a lot of useful information that does not quite fit into manpages for the functions in the package.
5. Many CRAN packages are paired with journal/textbook publications, which explain the methodologies, applications, limitations, etc in great detail.
6. CRAN has no problem rejecting packages, or removing packages that have problems that have gone unaddressed.
7. R resolves dependencies for the user and, since packages are pre-built for various machine/os types, installing packages is usually a quick operation.
PS. Julia is also very good on package management and testing. However, it lacks a central repository like CRAN and does not seem to have as strong a culture of pairing code with user-level documentation.
Do I get it right that this issue is within Windows? I've never heard of the issues you describe while working with Linux.. I've seen people struggle with MacOS a bit due to brew different versions of some library or the other, mostly self compiling Ruby.
The mmdetection library (https://github.com/open-mmlab/mmdetection/issues) also has hundreds of version-related issues. Admittedly, that library has not seen any updates for over a year now, but it is sad that things just break and become basically unusable on modern Linux operating systems because NVIDIA can't stop breaking backwards and forwards compatibility for what is essentially just fancy matrix multiplication.
I had issues on Mac, Windows and Linux... It was obnoxious. It led me to adopt a very simple rule: if I cannot get your framework / programming language up and running in under 10 minutes (barring compilation time / download speeds) I am not going to use your tools / language. I shouldn't be struggling with the most basic of hello worlds with your language / framework. I don't in like 100% of the other languages I already use, why should I struggle to use a new language?
On Linux good luck if you're not using anything besides the officially nvidia-supported Ubuntu version. Just 24.04 instead of 22.04 has regular random breakages and issues, and running on archlinux is just endless pain.
Have you tried conda? Since the integration of mamba its solver is fast and the breadth of packages is impressive. Also, if you have to support Windows and Python with native extensions, conda is a godsend.
It is not fast. Mamba and micromamba are still much faster than conda and yet lack basic features that conda has to provide. Everyone is dropping conda like a hot plate since the licensing changes in 2024.
I would recommend learning a little bit of C compilation and build systems. Ruby/Rails is about as polished as you could get for a very popular project. Maybe libyaml will be a problem once in a while if you're compiling Ruby from scratch, but otherwise this normally works without a hassle. And those skills will apply everywhere else. As long as we have C libraries, this is about as good as it gets, regardless of the language/runtime.
Have you tried JRuby? It might be a bit too large for your droplet, but it has the java versions of most gems and you can produce cross-platform jars using warbler.
And yet operating distributed systems built on it is a world of pain. Elasticsearch, I am looking at you. Modern hardware resources leave the limitations of running/scaling on top JVM to be an expensive, frustrating endeavor.
In addition to elasticsearch's metrics, there's like 4 JVM metrics I have to watch constantly on all my clusters to make sure the JVM and its GC is happy.
In-house app that uses jdbc, is easy to develop and needs to be cross-platform (windows, linux, aix, as400). The speed picks up as it runs, usually handling 3000-5000 eps over UDP on decade old hardware.
I'm surprised to hear that. Ruby was the first language in my life/career where I felt good about the dependency management and packaging solution. Even when I was a novice, I don't remember running into any problems that weren't obviously my fault (for example, installing the Ruby library for PostgreSQL before I had installed the Postgres libraries on the OS).
Meanwhile, I didn't feel like Python had reached the bare minimum for package management until Pipenv came on the scene. It wasn't until Poetry (in 2019? 2020?) that I felt like the ecosystem had reached what Ruby had back in 2010 or 2011 when bundler had become mostly stable.
Bundler has always been the best package manager of any language that I've used, but dealing with gem extensions can still be a pain. I've had lots of fun bugs where an extension worked in dev but not prod because of differences in library versions. I ended up creating a docker image for development that matched our production environment and that pretty much solved those problems.
That's one of the reason I prefer a dev environment (either physical install or VM) that matches prod. Barring that I would go with with a build system (container-based?) that can be local. Otherwise it's painful.
This is the right direction for Python packaging, especially for GPU-heavy workflows. Two concrete things I'm excited about: 1) curated, compatibility-tested indices per accelerator (CUDA/ROCm/CPU) so teams stop bikeshedding over torch/cu* matrixes, and 2) making metadata queryable so clients can resolve up front and install in parallel. If pyx can reduce the 'pip trial-and-error' loop for ML by shipping narrower, hardware-targeted artifacts (e.g., SM/arch-specific builds) and predictable hashes, that alone saves hours per environment. Also +1 to keeping tools OSS and monetizing the hosted service—clear separation builds trust. Curious: will pyx expose dependency graph and reverse-dependency endpoints (e.g., "what breaks if X→Y?") and SBOM/signing attestation for supply-chain checks?
Given that WSL is pretty much just Linux, I don't see what relevance Visual Studio compiler versions have to it. WSL binaries are always built using Linux toolchains.
At the same time, even on Windows, libc has been stable since Win10 - that's 10 years now. Which is to say, any binary compiled by VC++ 2015 or later is C-ABI-compatible with any other such binary. The only reasons why someone might need a specific compiler version is if they are relying on some language features not supported by older ones, or because they're trying to pass C++ types across the ABI boundary, which is a fairly rare case.
If you have to use, e.g., CUDA Toolkit 11.8, then you need a specific version of VS and its build tools for CUDA's VS integration to work. I don't know why exactly that is and I wish I didn't have to deal with it.
In my experience, Anaconda (including Miniconda, Micromamba, IntelPython, et al.) is still the default choice in scientific computing and machine learning.
It's useful because it also packages a lot of other deps like CUDA drivers, DB drivers, git, openssl, etc. When you don't have admin rights, it's really handy to be able to install them and there's no other equivalent in the Python world. That being said, the fact conda (and derivatives) do not follow any of the PEPs about package management is driving me insane. The ergonomics are bad as well with defaults like auto activation of the base env and bad dependency solver for the longest time (fixed now), weird linking of shared libs, etc.
I agree. Pixi solves all of that issues and is fully open source including the packages from conda-forge.
Too bad there is nowadays the confusion with anaconda (the distribution that requires a license) and the FOSS pieces of conda-forge. Explain that to your legacy IT or Procurement -.-
When was that ever a part of the definition? It was part of the early Unix culture, sure, but even many contemporary OSes didn't ship with compilers, which were a separate (and often very expensive!) piece of software.
OTOH today most Linux distros don't install any dev tools by default on a clean install. And, ironically, a clean install of Windows has .NET, which includes a C# compiler.
GitHub of the person who prepared the data. I am curious how much compute was needed for NY. I would love to do it for my metro but I suspect it is way beyond my budget.
(The commenters below are right. It is the Maps API, not compute, that I should worry about. Using the free tier, it would have taken the author years to download all tiles. I wish I had their budget!)
The linked article mentions that they ingested 8 million panos - even if they're scraping the dynamic viewer that's $30k just in street view API fees (the static image API would probably be at least double that due to the low per-call resolution).
OCR I'd expect to be comparatively cheap, if you weren't in a hurry - a consumer GPU running PaddlePaddle server can do about 4 MP per second. If you spent a few grand on hardware that might work out to 3-6 months of processing, depending on the resolution per pano and size of your model.
Their write up (linked at top of page below main link, and in a comment) says:
> "media artist Yufeng Zhao fed millions of publicly-available panoramas from Google Street View into a computer program that transcribes text within the images (anyone can access these Street View images; you don’t even need a Google account!)."
Maybe they used multiple IPs / devices and didn't want to mention doing something technically naughty to get around Google's free limits, or maybe they somehow didn't hit a limit doing it as a single user? Either way, it doesn't sound like they had to pay if they only mention not needing an account.
(Or maybe they just thought people didn't need to know that they had to pay, and that readers would just want the free access to look up a few images, rather than a whole city's worth?)
i just hashout out the details with claude. apparently it would cost me ~8k USD to retrieve all Taipei street images from gmap api with 3m density. Expensive, but not impossible.
> I've never seen Claude Code do anything that git couldn't easily reverse, so I'm not that worried about that any more.
I had one agent that installed something with pip and used it to rewrite my git history. It used the wrong flag and what was left was one file. I am glad I caught it before auto-commit and push triggered. It was one of my hobby projects, so not a huge deal. I can't imagine the fallout of this happening in a work environment.
