AI lineup (RLHF) and football prediction
The Bradley-Terry model has become central to chess players' Elo ratings, the best football match prediction models, language model alignment, and the democratic governance of recommendation AI. This video details
#TheFabulousProject #Sunflower #Bayes #Science4All
Playful:ai: https://playfulai.net/fr
Script + source code: https://github.com/lenhoanglnh/manusc...
Contribute to Sunflower: https://tournesol.app/
0:00 Introduction
2:01 The Elo score
5:18 The Bradley-Terry model
11:47 Playful:ai
17:19 Aligning language models
22:13 Our Bradley-Terry generalization
26:32 Elicitation and preference learning
29:03 Conclusion
Twitter: / le_science4all
Facebook: / science4allorg
LinkedIn: / l%c3%aa-nguy%c3%aan-hoang
The Formula of Knowledge (my first book):
https://laboutique.edpsciences.fr/pro...
The Fabulous Construction Site (my second book, with El Mahdi El Mhamdi):
https://laboutique.edpsciences.fr/pro...
Turing at the Beach (my third book, with Rachid Guerraoui):
https://www.dunod.com/sciences-techni...
SmartPoop (my fourth book, with Tristan Le Magoarou, aka @RisqueAlpha):
https://bit.ly/smartpoop-fr
The Dictatorship of Algorithms (my 5th book, with Jean-Lou Fourquet, aka @ApresLaBiere):
https://www.tallandier.com/livre/la-d...