'Tis The Season of Change

By Jaime Sevilla @ 2021-12-12T14:02 (+37)

A lot has happened since I wrote about my PhD in August. Seems like a reflection is in order!

In short, I have decided to suspend my PhD. For the next year I will be working as a contractor for the Open Philanthropy Project, helping them develop better models of how transformative AI might result in economic growth loops. If this goes well, I plan to transition into doing more contractor work for EA-aligned organizations.

But before we get to the juicy details, more on these last few months.

• Between September and October I did a PhD secondment in Maastricht University. I had a fantastic time and met some great people. Having an office and being surrounded by colleagues while working made a huge difference to my happiness and productivity.
  
  During the secondment I mostly worked on my ideas to explain Bayesian Networks. I was bearish about them before, but I have come around to seeing the value of my contribution to the field. 
  
  I wrote a summary of my approach. I have also released an open source package. A technical paper is available on the ArXiV.
  
  I believe my work is significantly better than the current state of the art on explaining Bayesian Networks. Unfortunately I think this speaks more about the state of the field than about my own research prowess.
   
• In the last week of October I participated in a one-week training session in Twente University about innovation and commercialization. There I finally met other ESRs in the same grant program as me. It was incredibly fun and motivating!
  
  As a result of this week I started my own Machine Learning art company, Connectome Art. We sell paintings made using Artificial Intelligence - a fun memento to commemorate the wild times we live in.
   
• Juan García, Ángela María Aristizábal and I have continued our work to promote GCR management in Spain and Spanish-speaking countries. 
  
  I am especially proud of our article on the Civil Protection System in Spain; if you are interested you can read our takeaways.
  We also published a website with information about GCRs in Spanish.
  Additionally we have been thinking about how other people can replicate our approach to Improving Institutional Decision Making.
  
  I will be dedicating less time in the future to this project, though I continue to think it is a very exciting one. If you have a background in policy or a GCR-relevant topic and want to collaborate, please reach out!
   
• Jonathan Lindblum and I published an article explaining how to model record progressions using Bayesian Statistics. Later we tried the model on some athletics data and got good results compared to previous frequentists approaches. I have been procrastinating on writing this up, sorry Jonathan.
   
• I have kept being involved in the Spanish-Speaking EA group, though less actively so. Our coordinators Laura González and Sandra Malagón have been absolutely amazing and have made the community a much more exciting place to belong to.
  
  I was planning on attending EAGx Prague to meet many of them for the first time IRL. The conference was canceled due to Omicron, but a few of us still were able to meet in Prague. It was fun!
   
• I got the green light to make public the article I wrote on intergenerational persistence! It exploded in popularity after being featured on Marginal Revolution, and received >5700 views. You can read my informal takeaways here. I am grateful to the Forethought Foundation for commissioning this work.
   
• I have been figuring out the best ways to aggregate forecasts. I wrote a well-received article defending logodd pooling, and a more nuanced recommendation later.
  
  This caused quite the uproar:
  Simon M tested my claims on Metaculus data, and found empirical support.
  Toby Ord responded in the comments bringing up some considerations.
  Alex Mennen wrote an article arguing for the opposite conclusion.
  I dug deeper into Mennen's argument and found a fascinating issue with Bayesian updating.
  
  The back and forth of arguments is the most fun I have ever had in research. I also feel quite accomplished about my research results, and I have had multiple people approach me for advice on the topic.
  
  I will definitely keep digging more into this topic.
   
• The team I was coordinating to research trends in Artificial Intelligence has grown from 3 people to 8 people. Everyone is tremendously smart and motivated, and we have made quite a lot of progress; expect news from us soon.
  
  Meanwhile, let me point out that our dataset is public and that we are excited to have other people build on top of it. We also built a cool visualization of the dataset.
  
  Our work has not gone unnoticed: researchers from Our World in Data reached out to us to inquire about our dataset, we have been exploring a collaboration with Neil Thomson's lab and Jack Clark used one of our graphics in a talk to DC policy-makers. It feels great to have already had such an impact!
   
• And finally, I have been talking with the Open Philanthropy Project about doing some contractor work for them. They will finance my research on AI trends and I will be working with Tom Davidson on developing a model of the interplay between automation and economic growth.
  
  This is a very exciting step for my career, and I am grateful to OpenPhil for the opportunity and to my PhD supervisors and colleagues for supporting this career move. To be able to do this I had to renounce my research grant, which is frustrating but I believe is the right move overall.
  
   

More fun things I have been up to:

• Dedekind's Army, the math student group I set up in my university, has been reborn! Some motivated students reached out to me a couple of weeks ago and I helped them take over the group resources.
• For complicated reasons I was hired to be a tour guide in an art exposition in Berlin for a weekend. I had not seen the exposition nor any of the exposed pieces before. Hilarity ensued.
• I wrote 30 more pages of my GCR-inspired fantasy novella. It is still really bad.
• I also wrote a short story about near term applications of AI. More for fun than a serious attempt at predicting the future, but I am very pleased with how much my writing has improved.

On the personal level, a mix of good and bittersweet news:

• After a year and a half of the pandemic I was able to visit my family again. My mom cried. I had missed them.
   
• I finally was able to renew my National Identity Card and open an investment account. I still need to make the decision of how much money to put in every month, but I am very glad I did this.
   
• Pablo Villalobos helped me with research and accountability to look into cryonics more seriously. I signed up as a member of Alcor, filed a declaration of intention to be cryopreserved and I got some quotes for life insurance through Unusual Risks, which Alejandro Ruiz helped me analyse. I am now procrastinating on deciding whether I want to buy the life insurance or invest the premium and pay for cryo off my own savings later.
   
• I decided to move elsewhere; I need a community around me working on similar topics to push my work to the next level. I have been invited to the FTX EA Fellowship in the Bahamas, and will be based there in the foreseeable future. Very excited to meet the community there!
  
  A sad side effect is that since neither I nor my partner want a long distance relationship we have broken up. I will miss her.

In short, my life has taken quite the 180º turn. I am very excited to be working with the Open Philanthropy Project on what I consider a very important problem. I’ll keep pushing my research on AI Trends and Technological Forecasting even further.