@moinnadeem
Pinned post
I don’t think people realize that if they use a coding harness with some inference provider, they’ve effectively given that inference provider shell access to that machine.View quoted post on X
Insane untapped (for now) attack vector.
phonic.ai Sign up for Phonic The speech-to-speech model company New on DIAL: Phonic-1 from @phonic_co and Realtime v1 mini from @ElevenLabs.
Overall preference:
🆕 Phonic-1: #2, 4.04/5
🆕 Realtime v1 mini: #12, 3.60/5
We evaluate these models through live, multi-turn conversations with human raters. Updated overall preference rankings below.
Performance can be separately compared across Consumer and Business conversations, with breakdowns across naturalness, audio quality, content and the issues users flag during their calls.View quoted post on X
Full-Duplex Models and Systems. Key Ideas and Papers. Moshi (@neilzegh @lmazare), Full-Duplex-Bench (@HungyiLee2 @shinjiw_at_cmu), τ-Voice (@karthik_r_n), STITCH (@LINJIEFUN), FLAIR (@tianyu_zh), Stream RAG (@Sid_Arora_18), Qwen-Audio (@TONGYI_SpeechAI) x.com/i/article/2118…View quoted post on X
