Deep dive into Deepseek’s chain of thought and MoE models this week. Family’s in Delhi, binged Love is Blind (surprisingly good watch for couples). Keeping up with resistance training, prepping for GPL tournament.
Explorations
I was deeply impressed by the ‘chain of thought’ which shows up in the case of Deepseek R1. I spent some time reading about ‘chain of thought’ and Mixture of Experts (MoE) model. Most literature hints at how ‘chain of thought’ is probably a step towards model explainability, but I disagree. ‘Chain of thought’ is probably a clever optimization to get these models to ‘pause’ and ’think longer’ on some problems. Nevertheless - I love reading the ‘chain of thought’ summaries for Deepseek R1 prompts.
Family & Friends
Mona and Ibhan left for Delhi over the weekend. Caught up on some Netflix / I stumbled upon “Love is Blind - US / UK / Habibi Edition” and binge-watched the latest seasons. I really found the format of the shop very involving and in all honestly I would recommend it even for married couples to see the show together.
Workout
Continued with resistance training + momentum of 11k steps / daily. Kept rest of the workouts to a minimum in prep for the Greenage Premier League 2025.