I was surprised to not find the funding freeze discussed on HN yesterday and I am speechless today at some of the comments I am reading in here. Are readers not aware of the scope and impact of this? I will tell you my story: I am funded by NASA to develop and maintain open-source machine learning software that they use directly on satellites. I am not the only one: many open source scientific software projects that are in widespread use are funded via federal grants of one form or another.
Such grants are funded on a reimbursement basis (or at least mine is): each month, I submit an invoice (via the scientific software 501(c)3 of which I am a part). Then NASA pays it. When the EO was announced, my guidance was simply that NASA was not going to pay out for a while and my February paycheck was just going to not show up for a while. To my knowledge, this was basically the guidance for every NASA, NSF, NIH, etc. grant. (I believe DARPA grants are not frozen.) These do not just fund what the more conservative among us might call 'mathematical wankery', they fund all kinds of things across science.
The issue is not whether some of these grants should have been issued in the first place. The issue is that suddenly, a large group of researchers is either not going to get paid, or their organizations are going to have to float their salary themselves on behalf of the federal government, because the federal government has just said that they plan to renege on their agreements. How many will miss rent or mortgage payments?
Although the order has been blocked for now, it is still unclear what it will mean for me and others. (I would submit an invoice on Feb. 1... would it be paid?) It's not like I just got laid off and can go look for other work now: the funding is likely to come back, I just get the joy of having no idea when.
For those of you who seem to have little problem with the EO itself: please take at least a few moments to consider whether your principles outweigh the real human costs here, and whether there might have been a less brutalistic way of achieving the same principles.
I completely agree, the whole type of decision making wrecks people's life. Fine, if the government wants have a discussion of what programs to fund, lets have that, but this id.otic 'stop funding' orders is just horrific for people involved. There could be people on grants, who are just planning to move to the US from some other country, who finished their previous contracts whose lives will be suspended.
I remember the same thing happened last time with the 'travel bans'.
Obviously this new government does not give a sh.t about people and rational decision making, but it's bizarre how some people in this thread are trying to justify that...
This you-know-better-than-me mindset is why all technology is converging to garbage. I can't speak for OP, but for me, you are half correct: I am indeed ultimately searching for answers, but I want the piles of web pages for an hour, and that is how I will find my answer. I don't like it when systems remove my agency by believing they can correctly predict my intent. (They can't.)
But there is no agency being removed here. You are being given more options by adding a button to chat if you so desire. It’s very small and unobtrusive
Ads, sponsoring and SEO spam were also very unobtrusive for a while. A few decades passed and we have to sift trough all the garbage to find anything relevant in our searches, it's very likely that search engines will push harder and harder for these AI assistants while making finding organic content and authoritative sources even harder, resulting in it being nearly impossible to do research by yourself rather then trough an AI that has god knows what bias and perhaps subtle ad driven bias into it.
In a lot of cases today’s AI systems can easily predict your intent once they accumulate enough context about you. Not in all, but when they predict incorrectly it’s easy to correct them using natural language. And nobody is removing the option of digging in the pile, it’s just that I don’t see why anyone would want to, in most cases. I certainly don’t - I got better things to do with my time.
It's slower but maybe the target audience is different? Armadillo prioritizes MATLAB like syntax. I use armadillo as a stepping stone between MATLAB prototypes and a hand rolled C++ solution, and in many scenarios it can get you a long ways down the road.
On this exact sequence, is there a LLM of choice that is really performant in this translation task? To armadillo, Eigen, Blaze or even numpy?
I have had very little success with most of the open self-hosted ones, even with my 4xA40 setup, as they either don't know the c++ libraries or generate very good-looking numpy stuff, full of horrors, simple and very very subtle bugs...
Looking for the same thing from any linear algebra library or language to cuda BTW (yes, calls to cu-blas/solver/sparse/tlass/dnn are OK), I haven't found one model able to write cuda code properly - not even kernels themselves but at least chaining library calls.
Linear algebra routines seem like one of the worst possible use cases for current LLMs.
Large amounts of repetitive yet meaningfully detailed code. Algorithms that can (and often are) implemented using different conventions or orders of operations. Edge cases out the wazoo.
A solid start seems like it would be using LLMs to write extensive test suites which you can use to verify these new implementations.
Yet for me all this C++/CUDA code is a lot of boilerplate to express dense and supposedly very tired concepts. I thought LLMs were supposed to help with the boilerplate. But yeah I guess it won't work.
And yes, it's nice to build unit test and benchmark harnesses. But those were never really such time-wasters for me.
Tough to say something as blanket as "it's slower"... there are lots of operations in any linear algebra library. It's not a direct comparison with other C++ linear algebra libraries, but hard to say Armadillo is slow based on benchmarks like this:
This toolkit was originally part of the mlpack machine learning library (https://github.com/mlpack/mlpack) before it was split out into a separate, standalone effort.
Yes, you use it to select the parameters of you optimiser, like step size and momentum schedules and other tuneable things. Look up MAML (meta abstract machine learning?)
11 years ago, I did something like that. I was trying to use particle swarm optimization to efficiently update neural network weights, it didn’t work well.
Don’t ask me why I even tried, I was a dumb but code happy undergrad!
Personally I don't like depending on musicbrainz or other online sources; any album that enters my collection I manually (and tediously...) tag via information found on discogs and other sources. I get that this is not... a typical desire, but, for anyone else who came across this post and thought "cool! I like music tagging!", other things worth mentioning are the really useful `eyeD3` utility [1] and `exfalso` [2] (which is from the same the `quodlibet` media player project, which I have found handles large libraries very well!).
Maybe someday I will relax my meticulousness enough to use online sources for automatic tagging... :)
If possible you can help other lazy people like me by adding album art and correcting the information in the MusicBrainz or Discogs catalogs.
It's very difficult to find print media nowadays and so all the liner, covers and booklets never get seen by people in the age of streaming music. Music is so depersonalised in streaming. I used to open the booklet and liner notes while putting on an album and it was a nice way to enjoy the album with some context.
I do the same, except I edit the information on musicbrainz (including adding album art) and then use that as the source of data. That way I'm contributing to the database and making sure my tags are perfect.
I also do the same for every album that enters my collection and I share a similar distaste for getting info directly off of online sources. Rather than an auto-tagger I've actually thought about writing a tag linter: something that checks that every track in my library is tagged according to my guidelines, maybe even reference online services for reference to check that there's no albums with missing tracks or that there's no typos, or that every track has cover metadata, and so on.
Very cool to see more C++-based machine learning efforts. The language still needs a good dataframe abstraction (maybe XFrame?), but with matrix algebra provided by Armadillo/Eigen and other long-time machine learning libraries like mlpack, Shogun, Shark (and if you want to include C, Darknet), personally I think the future is bright for machine learning in C++---especially for production and deployment applications.
100% agree, and there are a number of efforts in the space. mlpack (https://www.github.com/mlpack/mlpack/), Shogun (https://www.shogun-toolbox.org/), and Shark (https://www.shark-ml.org/) are three that have been around for over a decade now. They're a little niche because C++ is not that popular for data science, but they are generally pretty fast (especially mlpack, which focuses on speed).
Such grants are funded on a reimbursement basis (or at least mine is): each month, I submit an invoice (via the scientific software 501(c)3 of which I am a part). Then NASA pays it. When the EO was announced, my guidance was simply that NASA was not going to pay out for a while and my February paycheck was just going to not show up for a while. To my knowledge, this was basically the guidance for every NASA, NSF, NIH, etc. grant. (I believe DARPA grants are not frozen.) These do not just fund what the more conservative among us might call 'mathematical wankery', they fund all kinds of things across science.
The issue is not whether some of these grants should have been issued in the first place. The issue is that suddenly, a large group of researchers is either not going to get paid, or their organizations are going to have to float their salary themselves on behalf of the federal government, because the federal government has just said that they plan to renege on their agreements. How many will miss rent or mortgage payments?
Although the order has been blocked for now, it is still unclear what it will mean for me and others. (I would submit an invoice on Feb. 1... would it be paid?) It's not like I just got laid off and can go look for other work now: the funding is likely to come back, I just get the joy of having no idea when.
For those of you who seem to have little problem with the EO itself: please take at least a few moments to consider whether your principles outweigh the real human costs here, and whether there might have been a less brutalistic way of achieving the same principles.