hsikka

I think the issue with open endedness has been getting reliable training signal in the first place. Not sure how easy it is to systematically capture that signal, but cool to see teams are working on this

Hey HN, I’m one of the co-authors of this work. Very excited to see it here

We’ll actually be publishing a full paper soon, but happy to answer any questions in the meantime!

Hey, very cool work! How does your solution compare to directly building the context in a framework like LangChain?

I’m an AI researcher, so am a little further from this area, but am very curious.

Enlightening. Another far less covered topic is the role of Non Tenure Track “Research Faculty”, often titled Research Scientist, Research Professor, etc.

I occupy one of those positions, and their forward mobility and relation to traditional TT positions is strange, and I’d like to see it covered in more depth.

What a wonderful surprise! I remember playing with this back in 3rd and 4th grade during our computer lab, it was an absolute pleasure. I haven't seen it in over 15 years, and am just floored by how timeless this is. It just so happened that I walked by my old elementary school on Monday of this week. What a strange coincidence.

Truly a joyful memory.

A bunch of fellow researchers and I started Manifold Computing (https://manifoldcomputing.com), were we’re hoping to do live, open source research and build open source tools. As you said, time will tell but I hope we can do good work this way.

I’m not sure I follow this line of thought. If all the grocery stores did this, then I agree society would certainly take action: build better, more open grocery stores. surely the same principle applies here, A business like Twitter can choose to block a group of ideologies. If there is enough want in the part of people to propagate those ideologies, they’ll build a better forum and if those ideas are successful, that forum will be. Competitive, capitalistic behavior handles this.

That term is something I sort of came up with, but certainly there are some cool papers.

https://arxiv.org/abs/1905.10985 Is one, talking about using ideas like neural architecture search to build better learning systems. A lot of the references in it are golden.

I also wrote a set of notes formalizing what I see as the first step to building this infrastructure of intelligence. There’s some great references in there: https://osf.io/bv4qp/

I’d be happy to talk about this and get your thoughts, you can hit me up on my email: harshsikka123 @ Gmail.com

I’m leading a research collaboration with some researchers in academia and industry, working on this actively!

Building learning systems that can operate on multiple modalities, and are totally interpretable. I think of these requirements as the basis for the next big jump in software usability, (I.e much better intelligent user interfaces)

AGI is not what I want to build, someone else can do that. For now I want to increase the capacity for people to build neural networks that researchers don’t dream of doing, as easy as stitching together web APIs. I internalize this under the title “the Infrastructure of Intelligence”

Also, eventually I would like to work on a programming language for biology, and contribute to building wetware computers.

Another thing that I think the first project would help is longitudinal health tracking and quantifying human biology.

Great points, I’ll follow up via email!

Hey HN!

This thesis was the culmination of more than a year's worth of fulltime research. It was a blast to do, and in an exciting area. I've currently transitioned to a research scientist role at a defense company after completing my two Master's. I'd love to hear your thoughts or discuss anything!

I'm currently working on another DL project in my freetime with some collaborators. The goal is to build large sparse networks in a scalable way: https://www.harshsikka.com/creating-managing-and-understandi...

This is amazing, thank you! I will most definitely take you up on your offer!

I don’t know if this qualifies as Citizen science in the traditional sense, but I’m planning on doing a lot more I’d ependent research in Machine Learning this year. My focus is on Neural Architecture Search, Modularity, and biologically inspired prior in deep learning. I’ve written a bit about it here: https://www.harshsikka.me/the-diy-phd/

I personally think that there is still enormous room for improvement in applying even modern cutting edge techniques to real world usecases and industry.

Also, I think interpretability and the introduction of better understood priors will lead to extremely large, sparse networks being used across multiple modalities in applications and software.

Hey HN!

I've been contemplating doing independent research in ML and DL, and am really interested in building some informal structure around the activity. I just wanted to share my thoughts on the topic and get your feedback!

Wow thank you for the feedback! I love the idea that its a more natural, unobtrusive extension rather than an explicit agent sitting there.

I did a bootcamp and then went on to get my master's in CS. I think it would be more useful to you to do something like Lambda School or Hack Reactor, rather than go back for a 4 year degree. A lot of real engineering skills can and are learned by thousands on the fly as they work as Software Engineers. Programs like Lambda get you the essential toolkit and teach you how to learn, while degrees offer breadth in a lot of different types of topics including some very theoretical ones and can be offbeat with what is happening in industry. If you really do enjoy it, you can always work for a year or two as a software engineer, and decide you want to study Computer Science more formally and save up for that. Just my 2 cents, and I'm someone who's applying to CS PhD programs right now.

I believe it is an acronym for "Fresh Off the Boat", i.e. an immigrant. It's a slang term and I personally don't use it too much but have heard it used fairly often.

Hey HN, really thrilled to see this launch. I was a Research Fellow at the Paperspace Advanced Technologies Group this summer and saw this project develop. Gradient Notebooks were an indispensable tool in the work I was doing, and had several advantages for my work over other notebook services. As a researcher, it’s great to see an emphasis being placed on starting projects with no fuss or issues around setting up infrastructure and sharing/forking models.

Giving anyone access to free GPUs and powerful tooling seems like an incredible opportunity. I'd love to hear what you all think!

Thanks my friend, hope all is going well with you! I've been working on some cool stuff, will definitely share with you soon.

Second this! The course is solid and drops you right in to practice, which is great.

You could actually replicate all the work in this post after just watching the first lesson, that's how fast you start learning.

Hey HN, author here.

This post was the result of a small set of experiments that came out of the Paperspace Advanced Technologies Group. We've been working on some pretty ambitious research projects at the intersection of systems, ML, and HCI, and we were evaluating tools and libraries (i.e. Keras and Fast.ai) that would allow us to prototype concepts quickly. (More on our research approach and project structure coming soon). We found this interesting classification task and used it as a testbed for some small scale testing and the results were pretty cool!

Yes, it shocked me too! In this project I didn't get a chance to dive deep into some of the decisions the Fast.ai Library did, but I'm hoping to see if there's some inherent gain based on training.

I'm also very curious what the performance of some of the newer architectures, i.e. capsulenets would look like.

For what its worth, I recall this was like a student project from 2016 or something. I could be wrong, but it seemed like an impressive output under the circumstances.

This is enormously helpful, thank you!

Thank you!

What are the design tools that synthetic biology researchers use to create genomes? Is it some sort of EDA environment like Cello or Asimov, or is it done manually?

Damn, I was actually planning on applying to Comma for an internship because George Hotz was there, just to see if I could learn anything from him.

Brilliant, thank you!

This is really cool, I've been thinking about your comment for the past 15 minutes. I can imagine a swath of these BI tools, and also something that could be used in a consumer setting to manage your finances, health/exercise, mood, and more. Sort of like a personalOS.