Over the years, AI/ML research has become more and more centralized (at least in my opinion) with companies like OpenAI, DeepMind, etc. This has its pros and cons, but I think it would be great for the field to have more independent researchers with the flexibility to explore new ideas without being tied to any one institution. Independent research used to be much more common (with pretty great results: https://en.wikipedia.org/wiki/Independent_scientist), and I think with the internet and remote work, there’s no reason it can’t make a comeback. To that end, I’ve been working on a platform to connect independent ML researchers to independent backers who can provide ongoing funding on a monthly basis (similar to a Substack or Patreon model). This is definitely a platform I’ll be interested in using myself, but I’m curious to see if others would be as well and to hear any feedback.
What are some ways this will differ from Patreon and similar platforms? Put another way, what are some changes being made to that funding model to better support machine learning research specifically?
Supposing this platform is successful, is there an interest in expanding beyond machine learning? Why machine learning specifically? Is it what you know best, or is there some other reason to fovus specifically on ML?
Good questions. I think the key difference will be in the topic/market focus. The creator/artist market (e.g., Patreon) is different enough from the researcher market that I don't think people browsing Patreon (on average) would be interested in backing graph neural network research, for example. From there, it will be about tailoring the platform to researchers' (and their backers') needs which could include a bunch of things like facilitating connections between other researchers and "advisors", (hopefully) being able to provide cloud/GPU discounts, having different standard rewards (e.g., a backer's name listed on a paper), etc.
If there is enough interest, I'd definitely want to expand to other areas, but AI/ML seemed like a good place to start since it's where I'm most familiar and there's not much overhead in terms of needing lab space, etc.
Thanks for the answers. More questions if you're feeling up to it:
Will there be any sort of requirements to "demonstrate progress" to the funders (even if that means the results of a failed experiment or some sort of open lab notebook)?
Does all research need to terminate in a paper (as opposed to, say, tooling)?
In fact, how does accountability work in this system? Do researchers need to provide a provisional timeline? Research can fail or take unexpected turns away from the initial goal, and this may or may not make funders nervous. I guess this might be a way of asking: are participants funding people, or are they funding projects?
Will there be requirements to make all papers produced freely available? All source code? Is this at the researcher's discretion?
Suppose something profitable, or merely patentable, comes out of a Strato funded stint. Do the funders have any stake in such an outcome?
This sort of thing excites me, so I do hope you find some sort of success.
Over the years, top AI/ML research has become more and more centralized thanks to companies like OpenAI, DeepMind, etc. This has its pros and cons, but I think it would be great for the field to have more independent researchers with the flexibility to explore new ideas without being tied to any one institution. Independent research used to be much more common (with pretty great results), and I think especially with the internet and remote work, there’s no reason it can’t make a comeback. To that end, I’ve been working on a platform to connect independent AI researchers to independent backers who can provide ongoing funding on a monthly basis (similar to a Substack or Patreon model). This is definitely a platform I’ll be interested in using myself, but I’m curious to see if others would be as well or to hear any feedback.
This is the first feature of a tool I'm developing to quickly analyze large bodies of text. Try copying and pasting the text from the Moderna vaccine protocol as an example: https://www.modernatx.com/sites/default/files/mRNA-1273-P301.... Curious to hear any thoughts or feedback.
I was having some fun playing around with Google's Music Transformer so I built this site to make it more accessible and to generate sheet music so people could play along or use it to spark creativity. The songs can be hit or miss, but some turn out surprisingly well with noticeable long-term structure. I'm curious to see how others use it and am open to any feedback on what features to add next.
Really nice. I have a particular curiosity on this topic because I am studying data science with machine learning and until the covid19 pandemics started, I had a small business where I made music to free big chains/retail shops from paying music copyrights. I wonder where all this AI/Machine Learning thing will take music (and the business) to.
http://takeafive.com - Productivity tool that allows you to open a self-destructing tab during breaks. Solid and steady user base. Profitable, but haven't monetized much yet. contact: cole@takeafive.com
This was a really powerful essay. Don't know if using "8" is a good way to measure if there are not a lot of something. 8 light-years is pretty far and 8 tons is pretty heavy. But overall, great insights.
My favorite: "One heuristic for distinguishing stuff that matters is to ask yourself whether you'll care about it in the future. Fake stuff that matters usually has a sharp peak of seeming to matter. That's how it tricks you. The area under the curve is small, but its shape jabs into your consciousness like a pin."
