Vincent Ethier

Corporate escapee. Unreasonably curious. Builder of things.

[Vincent Ethier @vinidlidoo]

Rewired my voice agent for GPT-Live today and kept thinking about Kahneman's "Thinking, Fast and Slow." When the fast model hits something hard, it hands off to a slower one while keeping the small talk going, the way we do. A little spooky how these architectures are starting to look like how our heads work.

https://t.co/wQjHqU1qEq

[OpenAI Developers @OpenAIDevs]

GPT-Live-1 is now available in the API.

Bring ChatGPT’s natural back-and-forth to your app, with voice agents that listen while they speak and work with the models and harness you choose.

[Vincent Ethier @vinidlidoo]

@mavenroger @kaanbuildsai @shaneparrish That one cracked me up.

[Vincent Ethier @vinidlidoo]

A little late to the party, but I've switched to Soniox v5 real-time for dictation on my Mac, replacing ElevenLabs Scribe v2.

Transcription accuracy and speed have improved noticeably. Really remarkable work by the Soniox team, especially with mixed-language dictation.

https://t.co/2fx70XQoTv

[Soniox @soniox_ai]

Soniox v5 Real-Time is now available.

Live speech AI is not batch transcription with lower latency. It has to turn raw, noisy, continuous audio into structured intelligence while people speak.

What’s new:
• Higher accuracy across 60+ languages
• Completely reengineered speaker separation
• Better spoken language identification
• Higher-quality real-time translation across 3,600+ language pairs
• Faster semantic endpointing for voice agents
• Better alphanumeric recognition
• More robust native context handling

Built for voice agents, meetings, captions, translation, dictation, customer support, contact centers, and multilingual products.

Read more:
https://t.co/z01GyIOvPh

[Vincent Ethier @vinidlidoo]

@rlacombe Thanks Romain. We should catch up sometime!

[Vincent Ethier @vinidlidoo]

Spent the past two months building Waza, an AI coach for your hands. Demo videos in the post. If you own Meta glasses and an iPhone, I'd love to have you as a tester.

https://t.co/4UTXOnXSeD

[Vincent Ethier @vinidlidoo]

@foysavas hah! yeah, that could be a fun version of the chopsticks one. are you good with them?

[Vincent Ethier @vinidlidoo]

@natolambert in the neighborhood and super eager to connect with like-minded folks. if already full, pls count me in for the next one!

[Vincent Ethier @vinidlidoo]

@yutaAsano1991 @YuriMunchkin @yutaAsano1991 合コンだけではありません。

[Vincent Ethier @vinidlidoo]

disclosure: this article is pure love, unsponsored, and I receive no commission from VoiceInk.

[Vincent Ethier @vinidlidoo]

https://t.co/esgZzeXqWv

[Vincent Ethier @vinidlidoo]

and again, big shoutout to @JoshiPax from @tryvoiceink for the awesome work.

[Vincent Ethier @vinidlidoo]

@ClaudeDevs @claudeai gh issue: https://t.co/Sn4Sf4KQFo

[Vincent Ethier @vinidlidoo]

beware of Claude Code misusing the ScheduleWakeup tool call. it ran one of my skills in a loop every 25min and just cost me $50 in extra usage. @ClaudeDevs @claudeai

[Vincent Ethier @vinidlidoo]

@mitchellh Minuet.nvim does the trick for me. Supports local models, custom prompts and manual triggers. https://t.co/PIsuIjOyTz

[Vincent Ethier @vinidlidoo]

Part 2 turns that equation into a real proof: trusted setup, eight commitments, five pairing checks, zero knowledge. https://t.co/DL4apF30nx

[Vincent Ethier @vinidlidoo]

I wrote these two posts to teach myself the foundational mechanics behind zero-knowledge proofs. The series starts from zero and gets to a working SNARK in under 30 minutes. It's a bit math-heavy, yes (Part 2 especially), but I tried to make each line follow naturally from the last, and had fun using Nano Banana 2 and GPT Image 2 for the figures. ⬇︎

[Vincent Ethier @vinidlidoo]

Part 1 reduces a program to a single polynomial identity (R1CS → QAP → one equation). https://t.co/jh1D0BoSAA

[Vincent Ethier @vinidlidoo]

https://t.co/9ITruPDVFd

I like how we're discovering that agents need to learn what all of us had to learn working in hierarchical orgs: know your limits and ask for help, don't micromanage, remember your reports don't share your context, surface anything useful to manager and teammates, flat orgs past a certain size collapse into chaos... Labs should maybe open a role for an Organizational Behavior speciality within their post-training teams.

[Walden @walden_yan]

https://t.co/iY0G5ieQhJ