Just won 1st Place and $40k at the Mercor x Etched x Cognition inference-time compute hackathon!
4 Stanford freshman (@radi_cho, @zeynebnkaya, @nicolesplaining, and I) built "LLaDA-R1: Scaling Reasoning at Inference Time with Diffusion-LLMs" in 24 hours
Joe Li
34 posts
Joined November 2024
- Built *brain typing* at Treehacks. Modded a BCI to capture 16-channels of EEG data. Used facial muscle kinesis (binary mapping) with Huffman encoding to map the entire English alphabet— 100% deterministic. Then used autocomplete to type 2-3x faster than Stephen Hawking.Replying to @AlexReibman5/ NeuroScribe Brain computer interface keyboard Translate thoughts into binary, which can be translated into any language
- Replying to @leloykunThank you for the feedback! Some comments: Weights are explicitly normalized, so weight decay is unnecessary (see pg. 1 of paper). LR is tuned. See Section A.6. Interestingly, adding qk norm to ngpt has minimal impact. See Section A.8 Table 6.
- Replying to @ycombinator @WillowVoiceAI and 2 othersthis is insane, congrats!






