/ akbir khan / blog

🎴 first year retrospective

October 5, 2022

highlights:

  1. Submitting CHAOS 15 minutes before the deadline
  2. Pair coding on VMAP code with Newton and Timon and being told i have “magic hands”
  3. Having drinks with Laura and then going back and hacking together the human data collection form for implicature work
  4. Lab Socials: 1) post NEURIPS dinner with Rob and Tim. Good food and a very entertaining discussion about why safety is important. 2) summer drinks with UCL dark and friends. Highlight was Ed shouting to me “Short my ass” when meeting Jesse
  5. Ice-creams on the FAIR roof in summer with Mika and the RL group
  6. UCL speaker series with Max Jaderberg - very much felt the full UCL dark energy

low-points:

  1. First meeting with Jakob to show IPD results in which the response was that I clearly had a bug in my code
  2. Stoping work on Bronson project. On the flipside, subsequent coffee with Tim and Ed was very positive. They say this is was fine but should only happen once - really appreciated their honesty and clarifying expectations
  3. Withdrawing Maestro because it needed more work / being unable to get good MultiCarRacing results
  4. Awkward conversations about paper authorship.
  5. Pushback from Buck and Co on the importance of MultiAgent Safety

improvements i’ve noticed:

  1. How to run a research project and the cadence required to produce results on a weekly basis.
  2. Identifying problem worth solving and if its tractable with your given approach.
  3. Approaching collaborations, organising work and what work i enjoy
  4. Worrying less about code quality for research

things I still need to learn:

  1. Get better at finding papers which will have high impact.