The narrative on $META is about to change. 1. $META really surprised us with Spark 1.3, as it is a highly capable model on par with many even frontier models, while it is still not $META's big Watermelon model (coming soon according to Zuck and Alex's comments), which I expect to be significantly better than Spark 1.3. 2. The cadence of shipping models on which $META is working shows us that they set the foundation well and are now climbing on that ladder with compute. Muse Spark 1.2. was released just 1 month ago, and Spark 1.3 is much better than 1.2. More specifically, I think the ladder and the latest improvements of Spark 1.3 show that $META has figured out RL training well, which makes sense given $META's vast library of data, employees, and Wang, who knows the importance of data
View on X ↗