Faynt: learning competitive Melee
One policy for all 26 characters. A study of imitation, winning demonstrations, distillation, and reinforcement learning.
Research
Decoder-Only Transformer Optimization 101
A training-performance report following kernel fusion, memory reuse, CUDA-Oxide, and architecture changes for a Melee policy on one RTX 6000 Pro.
Why Aren't We Using Her ?
An essay on consumer voice AI adoption, examining the interface, missing capabilities, and reasons impressive demos have limited everyday use.
Moshi for Mere Mortals
A visual study of Moshi and Mimi with diagrams, equations, and pseudocode for low-latency speech systems.
What is full-duplex? How does half-duplex imitate it?
A guide to overlapping speech, interruptions, backchannels, and the design of full-duplex and half-duplex systems.
It's 2026...where are all the AI NPCs?
An examination of the economics, evaluation difficulties, and social behavior that complicate AI characters in games.
What if we made SIMA2 from Temu
A Roblox field report using screenshots, a fast action loop, a slower planner, and recorded failures to improve movement and recovery.
Grafting a Speech Head onto Gemma 4 E4B
An architecture walkthrough and early experiment mapping Gemma hidden states to Mimi audio tokens with a trainable speech head.
AI Gaming Voice Models, Ranked
A dated comparison of voice systems through the needs of gaming companions: interruption handling, response time, context, tools, and character consistency.
Inside the Mimi Codec
A browser demo that lets readers encode audio with Kyutai Mimi and switch codebook levels to hear their effect on reconstruction.
Why Frisson Labs?
The founders explain their interest in AI companions built around experiences shared with people.
No articles found.
Try a different topic or search term.