In partnership with
Hey friends 👋
Today we've got OpenAI's new voice model that's getting uncomfortably close to sounding human, a US-China standoff that's turning AI models into geopolitical bargaining chips, and a study on how many of us are already telling chatbots about our relationship problems. Let's dig in.
OpenAI just made ChatGPT sound a little too human
OpenAI launched GPT-Live this week, and it's the biggest jump yet in making AI voice assistants feel like an actual conversation instead of a call center hold loop.
The trick is a new full-duplex setup that lets the model listen and talk at the same time. That means it can drop in little "mhmm" sounds while you're talking, jump into quick back-and-forth banter, or just give you space to think without cutting you off. It also splits the voice model from the heavy lifting. When a question needs real reasoning, GPT-Live quietly hands it off to GPT 5.5 in the background and keeps chatting with you while that model works. In one press demo, it translated a live conversation into Hindi in real time and just chimed in with the finished translation instead of freezing up with an awkward "thinking" pause like the old Advanced Voice Mode used to.
OpenAI says the result beats Advanced Voice Mode on both human preference and technical benchmarks like scientific reasoning and agentic search. It's rolling out as GPT-Live-1 for paid ChatGPT users and a smaller GPT-Live-1 Mini for free users, with API access coming soon.
Here's why this one actually matters beyond the demo reel. Text has always been the friction point that kept AI out of daily life. A voice assistant that finally feels natural is the thing that gets AI out of the browser tab and into the background of your actual day. But there's a catch nobody's really grappling with yet. The more human something sounds, the more we tend to believe it, and GPT-Live still hallucinates just like every other model before it. We're making AI more persuasive faster than we're making it more accurate.
How owning AI deployment expands your career
Across product, ops, and CX teams, a new kind of role is taking shape: the person responsible for making AI actually work, day to day.
On July 16, three people living this shift join a live roundtable: Simone Santiago Broad (Yoco), Yelva Espinoza (Zumba Fitness), and Fin's Dave Lynch. You'll hear what the job really looks like across industries, how they carved out these roles, the skills they'd hire for, and the challenges they're tackling now. Bring your questions, since the best moments happen live.
Register for the roundtable to save your spot.
AI models are now bargaining chips in US-China relations
If you needed proof that AI has officially become a national security issue and not just a product category, look at what both governments did this week.
The US already restricted international access to Anthropic's Mythos and OpenAI's GPT-5.6 after Washington got worried Mythos could help attackers find and exploit software vulnerabilities faster than anything seen before. That lockdown gave the government and major US companies about three weeks to shore up their defenses before wider access returned. Now China appears ready to return the favor. Reuters reports China's Ministry of Commerce met with labs including Z.ai, Alibaba, and ByteDance to discuss restricting international access to their most advanced models, and even floated limiting foreign investment into Chinese AI startups.
This isn't just symbolic. Open-weight Chinese models like DeepSeek, Qwen, and GLM are widely used by US companies and developers trying to cut AI costs while keeping things private and running locally. OpenRouter data shows more than 30% of tokens run on Chinese models come from the US, spiking as high as 46% in some weeks this year. If China locks that down, it could be a real boost for US-made open models like Nvidia Nemotron, Google Gemma, and OpenAI's GPT-OSS. Meanwhile the rhetoric keeps climbing. China's Ministry of Industry and Information Technology claimed this week that Claude Code has a security backdoor, even though Claude Code isn't officially available in China, and Alibaba banned employees from using Anthropic tools entirely.
Here's the part that actually matters once the headlines cool off. None of this is likely to stop American and Chinese developers from quietly using each other's models. Mythos is now broadly available again. Older Anthropic and OpenAI models never left the market even during the US restriction. China is still just discussing limits, not enacting them. The loud diplomatic posturing is real, but underneath it, the actual usage probably keeps flowing across borders like it always has.
A few more things worth knowing
A new Oxford Internet Institute survey of 2,000 UK adults found 31% of regular LLM users lean on chatbots for emotional support, and 38% specifically ask for relationship advice.
Two-thirds of respondents said they turn to AI for health information, and a quarter want it for meaningful conversation.
Women were more likely to use AI for emotional support, men leaned more on it for practical tasks, and younger users used it across the board more than anyone else.
Despite all that, 75% of respondents said they were enthusiastic about AI chatbots overall. Worth noting: Stanford research has already found AI therapy chatbots can reinforce harmful stigma and give dangerous responses, so "enthusiastic" doesn't mean "risk-free."
That's what stood out to me today. Reply and tell me what caught your eye, I read everything.
Talk tomorrow,
Hatman 🎩