Machine Learning Street Talk
This week Dr. Tim Scarfe, Dr. Keith Duggar and Yannic “Lightspeed” Kilcher respond to the “Algoshambles” exam fiasco in the UK where the government were forced to step in to standardise the grades which were grossly inflated by the schools. The schools and teachers are all paid on metrics related to the grades received by students, what could possibly go wrong?! The result is that we end up with grades which have lost all their value and students are coached for the exams and don’t actually learn the subject.
We also cover the second Francois Chollet interview on the Lex Fridman podcast. We cover GPT-3, Neuralink, and discussion of intelligence. See the Lex interview here; https://www.youtube.com/watch?v=PUAdj3w3wO4
00:00:00 Algoshambles
00:45:40 Lex Fridman/Chollet: Intro
00:55:21 Lex Fridman/Chollet: Neuralink
01:06:28 Lex Fridman/Chollet: GPT-3
01:23:43 Lex Fridman/Chollet: Intelligence discussion
PODCAST VERSION OF THIS HERE: https://anchor.fm/machine-learning-street-talk/episodes/UK-Algoshambles–Neuralink–GPT-3-and-Intelligence-ej8c11
Thanks to Robert Lange for our specialised ML Street talk logo! https://roberttlange.github.io
#timscarfe #yannickilcher #keithduggar #machinelearning #deeplearning #aiethics
First
This channel is quite underrated. It sincerely deserves some attention (because, that's all you need!).
So the distribution in my data is same in all RGB channels. Can I say my data is unbiased??
Thoughtful conversation. I enjoy these discussions, fairness is a difficult subject. We can learn a thing or two from ethics, philosophy, sociology, anthropology.
Enjoyed this. Thanks fellas
Really interesting stuff, thank you! I do think there will be a large change in the approach to higher education in the near future with a shift to focus on opportunities such as apprenticeships where you can build your knowledge through experience. In my opinion, the whole system should focus on teaching students how to learn, not what to learn, and provide an environment where they can develop that skill with others.
Quite bizarre that Jannic's ARC videos are underviewed relative to his others. IMO it's some of the most important and exciting material there is in ML research.
Asking these "dumb" questions helps you make explicit what you're talking about. It's not about whether lex read the paper it's whether the listeners read the paper.
I'd give Lex the benefit of the doubt on the Priors etc… I have heard him often ask 'what is X' in various contexts where it's clear he must know the answer already but likely does it for the benefit of the viewer.
Honestly 2 numbers instead of 1 would be much better. The first one would be like grade where we say that this is the students/schools/other historic performance. The second one would be like variance and capability. Like though the child has scored grade X, he/she can work hard/less and improve/worsen by Y(should depend only on student). This will allow many students to be included because they will have high capacity with low grades.
Another major advantage would be that emphasis on student's other social/co-curricular/beneficial activities will increase when grades are close.