ICYMI: Our researcher @B_S_N_Y at NeurIPS. Thoroughly enjoyed the robotics showcase. Key insight from the teams presenting: at enough scale, vision-language-action models stop separating human videos and robot data. They learn a shared latent space. Once a VLA can control

key learning from recent convos with researchers: voice ai is increasingly circling hardware and wearable devices, and voice to action interfaces datasets and RL environments etc. will need to follow this new product paradigm

NeurIPS 2025 was a reminder that AI progress isn’t about louder demos, it’s about better data and solving the hard problems Great conversations with researchers pushing the edges of multimodal intelligence.

The datasets are fully OSS: https://t.co/jgthgHJmC6 If you’re training multimodal models, especially voice-heavy ones, feel free to DM. Happy to trade notes on multimodal AI!

ICYMI: We're excited to have contributed data to Pipecat’s Smart Turn v3.1. We're moving voice AI closer to truly fluid, interruption-aware dialogue. Our datasets power improved turn-taking and build on our work with frontier labs and leading voice AI teams. Link to HF below

Was great working on this with the @pipecat team. Completely OSS means open weights, data, and code. Pipecat is the leading Voice AI framework and Voice AI is the natural interface of AI. Happy to contribute data and see it public

Midcentury’s contribution involved 500+ diverse speakers, 1,000+ prosody-rich conversational “turns,” in multiple English accents that pushed the accuracy from 86% → 99%+. Check out the blog on how we built these datasets👇 https://t.co/gjIRP1Dhse

Thrilled to partner with @pipecat_ai on Smart Turn v3.1 Midcentury contributed high-quality human audio that helped unlock major accuracy gains for the OSS voice community. Our purpose-built datasets are rapidly closing the gap between synthetic & real human dialogue.