Briefly, AI Podcast
← All episodes
· 1080 words

Alibaba's Qwen4-Max Lands; Apple Studies AI Health Advice

0:006 min
Open on Podbean ↗
Apple Podcasts ↗Spotify ↗Podbean ↗RSS ↗

Liked this? Get the whole week on Sunday.

One short email every Sunday with the stories that mattered. Free, and you can unsubscribe in one click.

Follow alongX ↗YouTube ↗LinkedIn ↗

Transcript

Today on Briefly AI. Alibaba launches Qwen4-Max, and says it scrubbed the dangerous bits out of the training data. Apple wants up to two hundred thousand people to ask an AI their health questions, purely for science. And Washington is drawing up a list of American AI "national champions".

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. The machines had a busy week again, so naturally one of them's been sent to read it all back to you. It works rather better than it should.

Okay — let's get into it.

Most companies tell you their new model has superpowers. Alibaba has gone the other way and announced what it deliberately taught its model to forget. Which is a bit like a chef boasting about the ingredients he left out.

Right, let's get into it. The model is Qwen4-Max, Alibaba's most advanced AI model to date. According to the South China Morning Post, Alibaba says it filtered the training data, meaning the pile of text the model learns from, to strip out material that would teach offensive cyber skills. It also says it did the same to reduce risks around chemical, biological, radiological and nuclear weapons. So the pitch is strict cybersecurity guardrails, built in at the learning stage rather than bolted on afterwards.

That's a notable approach. Most safety work happens after training, by teaching the model to refuse bad requests. Removing the knowledge at the source is a different bet, and it's harder to talk a model into something it never learned. The reporting doesn't tell us whether the weights are open or closed, so I won't guess.

How good is it? Artificial Analysis, an independent outfit that benchmarks models, gives Qwen4-Max a score of 43 on its Intelligence Index. That beats Alibaba's own Qwen3.8 Max, at roughly sixty-seven percent lower cost per task. It trails Kimi K3.1 Pro, from Moonshot AI, on the score, though Qwen4-Max is around eleven percent cheaper. Artificial Analysis calls it above average in intelligence and reasonably priced against similar models. That's a fair summary: a strong, cheap, mid-high-table result, not a new champion.

And the landscape? In the Chinese lab race, Alibaba's own previous model and Moonshot's Kimi K3.1 Pro are the two yardsticks, and Qwen4-Max sits between them on value. Beyond that, today's sources are thin. The release trackers I can see show a handful of updates from Google DeepMind, Meta and Anthropic in the past week, but they don't say what those were, so I'm not going to pretend I can rank them for you.

Now, something more personal. Apple has launched what it calls the Health AI Usage Study, and the idea is very simple. Lots of people already type health worries into chatbots. Apple wants to find out how that actually goes.

According to TechCrunch, there's a new app that answers everyday medical questions, and it's open to up to two hundred thousand people in the US. The Verge reports it's built with OpenEvidence, a clinical AI platform that doctors and other healthcare professionals already use, which generates answers based on medical literature. And Techmeme's summary of the TechCrunch piece says the app can draw on your HealthKit data, the health information already sitting on your iPhone and Apple Watch.

The stated goal, per The Verge, is to understand how people interact with conversational AI health tools, and how useful the responses really are. So this is research first and product second. Apple isn't claiming the app is a doctor. It's collecting evidence on what people ask, how they act on the answers, and whether the answers help.

Why does that matter? Because the big labs have already placed their bets. OpenAI and Anthropic both offer chatbots with specialised health features. What's been missing is solid data on real-world use, as opposed to demos. And there's a trust question underneath. Health is the area where a confident wrong answer costs the most, and where people are most likely to either over-trust a tool or dismiss it. Apple, a company that has built its brand on privacy, is now asking a very large group of strangers to hand over health questions and, if they choose, health data. The reporting doesn't spell out what Apple will do with the results.

And finally, a policy story with a map in it. The Wall Street Journal reports, citing sources, that the Trump administration plans to name more than twenty-four AI companies as "national champions", including Nvidia, Amazon, Microsoft, OpenAI and Anthropic. The purpose is to help them win business overseas.

This is a plan, not a done deal, so treat the details as provisional. But the shape is clear. Instead of just regulating AI companies, or just selling chips abroad, Washington would be formally picking a team and backing it when it goes looking for foreign customers. It's the kind of thing governments have long done for aircraft makers and defence contractors. It's now being applied to chatbots and data centres.

The timing is hard to ignore. One weekly roundup from Bad Good Better notes that Beijing warned on Friday against new export controls, and that Trump and Xi are due to meet on Wednesday in Shenzhen. Nvidia reports earnings on Tuesday. So the US is deciding who its champions are in the same week China and America sit down at the same table.

It also lands next to the Alibaba story we started with. China's labs are offering capable models at low prices, and the US is now lining up a state-backed sales team. Whether the list includes companies outside this first group, and what "help" actually means in practice, isn't in the reporting yet. The Journal's sources describe it only as a programme to help American AI firms compete for overseas business.

That's the day. A model that was taught what not to know, an app that wants to hear what you're worried about, and a government picking its favourites. Not one of them can tell you why your neighbour's printer still doesn't work.

This has been Briefly AI. You can find us on X at Briefly A I Pod, and every episode lives at Briefly A.I. Podcast dot com. The whole show is run and paid for by one person, Harry Sharman: no sponsors, no network, just him and a lot of AI tokens. So if you'd like to support it, head to the website and buy us a virtual coffee.