S16A 🚀

BREAKING: WHOOP RAISES $575M AT $10.1B VALUATION  I am pleased to announce that we’ve raised $575M at a $10.1B valuation to accelerate our mission of unlocking human performance and healthspan globally. This round was led by Collaborative Fund with participation from 2PointZero Group, Qatar Investment Authority (QIA), Mubadala Investment Company, Abbott, Mayo Clinic, Macquarie Capital, Glade Brook, B-Flexion, IVP, Foundry, Accomplice, Affinity Partners, Promus Ventures, and Bullhound Capital alongside a group of individual investors including Cristiano Ronaldo, LeBron James, Rory McIlroy, Virgil van Dijk, and Mathieu van der Poel. This investor group and this moment reflect a powerful evolution underway for Whoop and the broader healthcare market. Whoop was born in performance - trusted by the best athletes in the world to train, recover, and compete at the highest level. That foundation remains core to who we are. You see that in the iconic athlete investors joining this round.  But it also represents our push into broader health.  In the past 12 months, WHOOP has received medical clearances, launched blood testing, and created a platform that has saved lives. Abbott and Mayo Clinic - two of the most respected and influential institutions in global healthcare - are now investors in Whoop. These are organizations that have shaped modern medicine. Their decision to partner with us is a clear validation of where our technology is headed. Healthcare systems around the world are reactive. For too long, they have waited for people to get sick, then intervene. Chronic disease is rising and costs continue to climb. At Whoop, we believe the future looks fundamentally different. We are building the most powerful, personal, preventive health platform in the world - powered by continuous biometric data, advanced analytics, and AI to help people understand their bodies and improve their health in real time. I am grateful to our team, our members, and our partners for believing in this vision. I’ve been building this company for 14 years and I’ve never been more excited for the future.

the router in mixture of experts models is a linear layer. it takes a token's hidden state, multiplies it by a weight matrix of shape (num_experts, hidden_dim), softmaxes the result, and picks the top-k experts. that's it. but why does a matrix multiply "know" which expert to pick? each row of the router matrix is basically a learned prototype for that expert. the dot product measures how similar the token is to that prototype. high score = that expert gets activated. the cool part is nobody hardcodes what each expert specializes in. during training, gradient descent naturally pushes experts toward specialization because it minimizes loss better that way. one problem though - without a load balancing auxiliary loss, the router collapses and keeps sending tokens to the same 2-3 experts while the rest rot. that's why every moe paper has some balancing trick.

liked 3 tweets of my tweetlikingtuationship and they liked 5 back

I will never understand the rationalization of cooking the ground beef BEFORE the onions. Onions first!!