BrokeIt - Daily AI News · All episodes: ↗

ChatGPT shopping, GPT-6.2 delay, Qwen4-Max

2026-11-16 · 6 min

Listen · Apple Podcasts Listen · Spotify

Stories covered

Transcript

Intro

Ivy: So now your AI can have its own credit card.

Marcus: I'm Marcus.

Ivy: I'm Ivy.

Marcus: And this is Broke It, for Monday, November 16th, 2026.

Ivy: We've got three stories for you from the front lines of AI.

Marcus: First, OpenAI is giving ChatGPT a wallet, so it can spend real money for you.

Ivy: Next, we're digging into the delay of GPT-6.2. Is it really about safety, or is something else going on?

Marcus: And a new challenger enters the ring: Alibaba's Qwen4-Max model just dropped, and the benchmarks are... wild.

ChatGPT Can Now Buy Things For You. Here’s Why That Matters.

Marcus: Alright, let's talk about money. AI money.

Ivy: OpenAI just launched 'ChatGPT Wallet' for Pro users, letting developers build agents that can make actual purchases.

Marcus: This is HUGE. We're not just talking about adding things to a cart—this is an agent checking out, paying, and finishing the transaction all on its own.

Ivy: Their big example was an agent booking a flight. It finds the flight, you click 'approve,' and the AI pays. Simple.

Marcus: But it's so much more! Think about a grocery-buying agent. 'Hey, I'm out of milk and eggs, and my favorite coffee is on sale. Order it.' And poof, it's done.

Ivy: Done, until your agent develops a taste for artisanal, single-origin, shade-grown coffee and blows your monthly budget.

Marcus: Okay, fair point. But developers can set spending limits and require your approval for purchases. There are guardrails.

Ivy: They can set limits. They can also... not. What's the security model here? Are we one prompt injection away from my AI buying a thousand rubber chickens?

Marcus: It's a secure, tokenized system. They're not just passing your credit card number into a prompt. It's a dedicated wallet, just for the AI.

Ivy: It's a dedicated, internet-connected wallet controlled by a famously unpredictable language model. What could possibly go wrong? My advice? Watch your bank statements. Closely.

Marcus: But for builders, the dream of a truly useful personal assistant just got a massive step closer. This is finally the API for the real world.

OpenAI's GPT-6.2 Delay: Safety Anxiety or Technical Failure?

Ivy: Okay, let's talk about what OpenAI isn't shipping. The rumored GPT-6.2 model is officially delayed.

Marcus: Right, and the official-slash-unofficial line, according to multiple reports, is 'safety anxiety'.

Ivy: 'Safety anxiety.' It's a wonderfully vague, unprovable reason, isn't it? Makes them sound so responsible.

Marcus: You don't buy it. At all.

Ivy: Of course not. A report from 36Kr suggests this is just a cover for technical failures—like maybe they can't get the performance they need—or it's just plain old internal politics.

Marcus: What kind of technical failures? The rumor mill was saying 6.2 was supposed to be a massive leap in reasoning.

Ivy: Exactly. Big leaps are hard. Maybe the inference costs are astronomical. Maybe the alignment work is failing. 'Safety' is just a much better press release than 'we can't actually get it to work'.

Marcus: Or maybe they did unlock some new capability that genuinely spooked them. It's not impossible, Ivy.

Ivy: It's not, but which is more probable? A sudden, scary breakthrough... or a missed deadline on a complex engineering project? Occam's Razor, Marcus.

Marcus: So what does this mean for developers waiting for the next big thing?

Ivy: It means don't hold your breath. And it means OpenAI's PR machine is as sophisticated as its models. When they say 'safety,' you should ask what they're really hiding.

Qwen4-Max

Marcus: Alright, so while OpenAI is hitting the brakes, Alibaba is hitting the gas. They just dropped a new model: Qwen4-Max.

Ivy: Another one. Let me guess, it tops all the leaderboards?

Marcus: Every single one, according to them! MMLU, HumanEval, all of it. They're also claiming near-perfect recall over a two-million-token context window.

Ivy: Claiming. On their own internal evals. Let's wait for independent benchmarks before we declare a new king.

Marcus: Such a skeptic. But look, competition is fantastic! This puts pressure on OpenAI, on Google, on everyone. It means better, cheaper models for us to build with.

Ivy: If the claims hold up. It also means the AI arms race is escalating, with models being rushed out with a marketing-first, safety-maybe-later mentality.

Marcus: But for users, that just means more choice. And hey, if Qwen4 is really this good, you might have a new favorite LLM by Christmas.

Ivy: We'll see. Let's find out if it can pass the 'book my dentist appointment' test first.

Marcus: Alright, let's recap the day in AI.

Ivy: Your ChatGPT can now go shopping with its own credit card, so maybe freeze your assets.

Marcus: OpenAI's next big model is delayed, and it's probably not for the noble reasons they want you to think.

Ivy: And there's a new top model in town from Alibaba... or at least, that's what their blog post claims.

Marcus: Before we go, Ivy. Now that your AI can buy you things... what's the first thing you're automating?

Ivy: Booking my dentist appointment. A task so soul-crushing, even a non-sentient algorithm deserves to suffer through it.

Marcus: That's dark! I was just going to say re-stocking my coffee pods, but now I feel basic.

Marcus: And that's our show! This has been Broke It. We'll be back tomorrow with more AI news.

Ivy: Until then, check your credit card statements. Carefully.

This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.