Fund me please - I Work so Hard that my Feet start Bleeding and I Need to Infiltrate University

post by Johannes C. Mayer (johannes-c-mayer) · 2024-05-18T19:53:10.838Z · LW · GW · 37 comments

Contents

  Bleeding Feet and Dedication
  Collaboration as Intelligence Enhancer
  Hardcore Gamedev University Infiltration
  Over 9000% Mean Increase
  Don't pay me, but my collaborators
  Join me
  The Costly Signal
    Research Artifacts
    On Portfolios
None
37 comments

Thanks to Taylor Smith for doing some copy-editing this.

In this article, I tell some anecdotes and present some evidence in the form of research artifacts about how easy it is for me to work hard when I have collaborators. If you are in a hurry I recommend skipping to the research artifact section [LW · GW].

Bleeding Feet and Dedication

During AI Safety Camp (AISC) 2024, I was working with somebody on how to use binary search to approximate a hull that would contain a set of points, only to knock a glass off of my table. It splintered into a thousand pieces all over my floor.

A normal person might stop and remove all the glass splinters. I just spent 10 seconds picking up some of the largest pieces and then decided that it would be better to push on the train of thought without interruption.

Sometime later, I forgot about the glass splinters and ended up stepping on one long enough to penetrate the callus. I prioritized working too much. A pretty nice problem to have, in my book.

[Edit 2024-05-19] The point is that this is irrational, and I have the problem of working too much. But this is a problem that's much easier to solve than "I have trouble making myself do anything". More details here [LW(p) · GW(p)].

Collaboration as Intelligence Enhancer

It was really easy for me to put in over 50 hours per week during AISC^[1] (where I was a research lead). For me, AISC mainly consisted of meeting somebody 1-on-1 and solving some technical problem together. Methylphenidate helps me with not getting distracted when I am on my own, though Methylphenidate is only the number 2 productivity enhancer. For me, the actual ADHD cure seems to be to take methylphenidate while working 1-on-1 with somebody.

But this productivity enhancement is not just about the number of hours I can put in. There is a qualitative difference. I get better at everything. Seriously. Usually, I am bad at prioritization, but when I work with somebody, it usually feels, in retrospect, like over 75% of the time was spent working on the optimal thing (given our state of knowledge at the time). I've noticed similar benefits for my abilities in writing, formalizing things, and general reasoning.

Hardcore Gamedev University Infiltration

I don't quite understand why this effect is so strong. But empirically, there is no doubt it's real. In the past, I spent 3 years making video games. This was always done in teams of 2-4 people. We would spend 8-10 hours per day, 5-6 days a week in the same room. During that time, I worked on this VR "game" where you fly through a 4D fractal (check out the video by scrolling down or on YouTube).

For that project, the university provided a powerful tower computer. In the last week of the project, my brain had the brilliant idea to just sleep in the university to save the commute. This also allowed me to access my workstation on Sunday when the entire university was closed down. On Monday the cleaning personnel of the University almost called the cops on me. But in the end, we simply agreed that I would put on a sign on the door so that I wouldn't scare them to death. Also, I later learned that the University security personnel did patrols with K-9s, but somehow I got lucky and they never found me.

I did have a bag with food and a toothbrush, which earned me laughs from friends. As there were no showers, on the last day of the project you could literally smell all the hard work I had put in. Worth it.

Over 9000% Mean Increase

I was always impressed by how good John Wentworth is at working. During SERI MATS, he would eat with us at Lightcone. As soon as all the high-utility conversation topics were finished, he got up – back to work.

And yet, John said that working with David Lorell 1-on-1 makes him 3-5x more productive (iirc). I think for me working with somebody is more like a 15-50x increase. 

Without collaborators, I am struggling hard with my addiction to learning random technical stuff. In contrast to playing video games and the like, there are usually a bunch of decent reasons to learn about some particular technical topic. Only when I later look at the big picture do I realize — was that actually important?

Don't pay me, but my collaborators

There are multiple people from AISC who would be interested in working with me full-time if payed. Enough money to just pay one person would give me over 75% of the utility. If I need to choose only one collaborator, I'd choose Bob (fake name, but I have a real person in mind). I have almost no money, but I can cheaply live at my parent's place. So the bottleneck is to get funding for Bob.

Bob would like ideally $90k per year, though any smaller amounts would still be very helpful. I would use it to work with Bob full-time until the funds run out.

I might also consider choosing somebody other than Bob who would be willing to work for less.

Join me

Of course, another way to resolve this issue is to find other collaborators that I don't need to pay right now. Check out this Google Doc if you might be interested in collaborating with me.

The Costly Signal

Research Artifacts

I don't have a portfolio of polished research results. But what I do have is a bunch of research artifacts, produced during AISC - i.e. various <documents/whiteboards> that were created during the process of doing research. I expect faking something like this is very hard. 

Over 90% of the content in these artifacts is written down by me, though heavily influenced by whoever I was working with at the time. For a list of collaborators and a short description of what I am working on, see here.

I hope, at minimum, this demonstrates that I am able to put in the time (when I have collaborators). Optimistically, these artifacts not only show that I can put in the time (with collaborators) but also demonstrate basic technical competence.

None of these documents try to communicate the "why is this good". Rather, I hope that looking at them will make somebody think "This seems like the kind of research artifact that somebody who moves in the right direction might produce." I expect that if I were to look through similar lists made by other people, it would allow me to better evaluate them. But I am not sure to what extend other people would be able to do this, and I am highly uncertain about how well I would be able to do it myself.

In any case here is the list of artifacts:

• Here are all the pictures I took of my physical whiteboards during AISC.
• In the Science Algorithm Research Log, I write down rough notes about things I understood after figuring them out on a whiteboard. I created this doc just as AISC ended.

During AISC, we used Eraser extensively; here are a few of the boards, ordered after some intuition of what I think would be best for you to look at (note that some are very large and might take some time to load in):

• What is a good concept
• Bottleneck Abstraction and Planning
• Various topics: Abstraction / Program Synthesis / Hull Tiling / Defining World Model / Didactic Goals
• Running A World Model Backwards to Plan
• Minecraft Crafting and World Model Inference
• Forward VS Backward Planning / Network routing and Amorphous computing
• World Model Structure / Planning Proving Isomorphism / Running the world model Backwards
• A bunch of considerations about inferring world models (Note that there are multiple spaced apart columns of content)
•  A tiny board about Program Search

I also have 61 hours of video recordings of me working during AISC (for logistical reasons, not linked here).

If anybody does vaguely consider funding me, I expect that you would significantly update towards funding me after talking for me in a video chat. Empirically, people seem to think a lot higher of me when I meet them face to face, 1-on-1. Also, I can make you understand what I am doing and why.

Alas, I think it's quite unlikely that this article will make somebody fund me. It's just that I noticed how extremely slow I am (without collaborators) to create a proper grant application.

On Portfolios

Let me briefly spend some time outlining my current models of what a good research portfolio would look like, in the hope that somebody can tell me how to improve them.

I have this game design portfolio website that showcases all the games I worked on. Making all these games took 3 years. This is a signal that's very hard to fake, and also highly legible. When you watch the video for a game, you can easily understand what's going on within a few moments. Even a person who never tried to make a game can see that, yes, that is a game, and it seems to be working.

I don't have such a convincing portfolio for doing research yet. And doing this seems to be much harder. Usually, the evaluation of such a portfolio requires technical expertise - e.g. how would you know if a particular math formalism makes sense if you don't understand the mathematical concepts out of which the formalism is constructed?

Of course, if you have a flashy demo, it's a very different situation. Imagine I had a video of an algorithm that learns Minecraft from scratch within a couple of real-time days, and then gets a diamond in less than 1 hour, without using neural networks (or any other black box optimization). It does not require much technical knowledge to see the significance of that.

But I don't have that algorithm, and if I had it, I would not want to make that publicly known. And I am unsure what is the cutoff value. When would something be bad to publish? All of this complicates things.

Right now, I have a highly uncertain model which tells me that without such a concrete demo (though it probably could be significantly less impressive) people would not fund me. I expect that at least in part this is because my models about what constitutes good research are significantly different from those of most grantmakers. Multiple people have told me that projects about "Let's figure out <intelligence / how to build an AGI we understand>" are not well liked by grantmakers. I expect this is because they expect the problem to be too hard. People would first need to prove that they are actually able to make any progress at all on this extremely hard problem. And probably, they think that even if somebody could solve the problem, it would take too much time for it to make a difference.

If anybody has better insights than me into why this is the case, that would be helpful. What would make somebody happy to fund a project like "Let's figure out <intelligence / how to build an AGI we understand>"?

---

Lastly, I leave you with another artifact: a digital whiteboard created over maybe 10-20 hours during AISC (again, you are not supposed to try to understand all the details):

1. ^{^}

   Note that for unknown reasons I need to sleep 10-12 hours. Otherwise I get a significant intelligence debuff.

37 comments

Comments sorted by top scores.