Tech brief
Wednesday signal: the ceasefire called off, NATO's €70B, and ChatGPT learns to interrupt
Strikes return to Hormuz, Ankara closes with a Ukraine pledge, Exxon's war windfall, a split Fed, and a housing bill running out the clock — press a signal for coverage.
Signalpoint TeamBrief
Tech
Voice is becoming the storefront and the big models the back office — whoever owns the conversation decides which AI gets the hard questions behind it.
BackgroundVoice assistants have historically worked like walkie-talkies: they wait for you to finish, then respond. Making AI conversation feel human requires models that can listen and talk simultaneously — and interrupt, or be interrupted.
Points
- Two models ship globally today: GPT-Live-1 as the default for paid Go, Plus and Pro tiers, a mini version for free users — iOS, Android and web, but no developer API at launch.
- Handing hard questions to the bigger model closes voice ChatGPT’s intelligence gap: OpenAI’s tests show gains over the old voice mode on science, web-research and multi-step support tasks.
- More than 150M people use ChatGPT Voice weekly; the nine voices were remastered with safeguards against imitating real people, and OpenAI insists this is not an AI-companion play.
- Gaps to watch: no video or screen sharing yet, uneven non-English fluency — and Google, ElevenLabs and Deepgram already sell voice toolkits to developers, a market GPT-Live can’t reach without an API.
Unlock the full brief
Sign in to read every signal, takeaway, and source. Free account — Apple, Google, or email.