This group research project consisted of an initial psychological experiment conducted on our own school, which aimed to assess the extent to which instrumentation affected musical preferences using various statistical and sampling methods, such as the two-tailed t-test.

After it was (surprisingly!) determined that there was insufficient evidence supporting that instrumentation meaningfully effected musical preferences, we used it to inform the production of a simple computer program which used genre as a proxy to predict musical preferences, using dynamics and harmony. The program took a large, assorted set of songs of various genres, and another smaller list of songs of a genre we wanted to pick out. From there, the program would output the list of assorted songs in order of how closely they match the songs we want to pick out.

Out of a list of 16 random songs (5 of which we wanted to pick out), the program successfully ranked 4 of the songs we wanted to pick out within the top 5 rankings.

This involved the production of a 5-minute video essay on the above question. During it, I covered key important concepts within the field of AI safety, such as the orthogonality thesis, instrumental convergence, the "race to the bottom" with capabilities development, and recent examples of sycophancy and deception within AI systems that could pose a serious risk when scaled up.

I went on to win a Special Commendation for the video essay, being recognised for an "excellent knowledge of state-of-the-art", as well as "clear on-screen summaries" and a "nice definition of intelligence".

My submission for this essay competition was on "back-of-the-envelope", or in this case, "napkin", calculations; i.e. Fermi problems. This (mainly fun) read begins with an estimation as to how many tennis balls fit within the Eiffel Tower, using techniques such as exponential interpolation to estimate the mathematical equation to which the tower's curve matches and knowledge of sphere packing, before then estimating how long it would take to find an atom hidden randomly in the universe, and then using the binomial formula to model the infinite monkey theorem given that, just like us, they are also allowed to make mistakes.

After a brief dive into some smaller estimation problems, the essay ends with a poignant calculation on how much the average relationship (and breakup) weighs, and then ways in which Fermi estimations are really used in real life.