update

README.md

@@ -10,16 +10,16 @@ We mostly talk on discord.gg/gpumode in the popcorn channel. It's pretty casual
 ## What We're Working On
 
 ### Data
-We're collecting every CUDA and Triton kernel we can find. We need more examples of good GPU code than what's currently out there. We're also building proper benchmarks so we can actually tell if what we're doing is working.
+We're collecting every CUDA and Triton kernel we can find. We need more examples of good GPU code than what's currently out there.
 
 ### Infrastructure
 We need compute - we're building tools to track progress and share results in real-time through Discord.
 
 ### Research
-We're trying to figure out the fundamentals - what do we need to tell an LLM about GPUs for it to write good code? How many times should we sample to get better results? What feedback actually helps?
+We're exploring what LLMs need to know about GPUs to generate effective code, and how to best sample and train them.
 
 ### Language Design
-We think some abstractions like Thunderkittens might make this whole thing easier. If that proves true, we'll push to get more people using and improving these tools.
+We're exploring new abstractions to simplify GPU programming and will promote successful approaches.
 
 ## Collaborators and Sponsors