Sunday Rundown #92: Chinese Models & Epic Moves

Sunday Bonus #52: Get live insights from foreign news sources.

Happy Sunday, friends!

Welcome back to the weekly look at generative AI that covers the following:

• Sunday Rundown (free): this week’s AI news + a fun AI fail.

• Sunday Bonus (paid): an exclusive segment for my paid subscribers.

Let’s get to it.

🗞️ AI news

Here are this week’s AI developments.

👩‍💻 AI releases

New stuff you can try right now:

1. Anthropic says Claude can finally search the web. (Looks like they listened to me.) Available to paid accounts in the US now; rolling out to everyone soon.

2. Adobe news:

   1. The Adobe Experience Platform Agent Orchestrator (oooh, catchy!) helps businesses manage AI agents across Adobe and other systems.

   2. Speaking of agents: Adobe Marketing Agent lets users create assets directly inside Microsoft 365 Copilot, PowerPoint, and Word.

3. Baidu launched two new models:

   1. ERNIE X1 is competitive with DeepSeek-R1 but costs 50% less.

   2. ERNIE 4.5 performs on par (or better) with GPT-4.5 but is way cheaper.

4. Google simply refuses to stop shipping stuff for the third week in a row:

   1. The new Canvas works much like the similarly named feature in ChatGPT (or “Artifacts” in Claude), outputting code or text results in a separate window.

   2. The famous NotebookLM Audio Overviews are now coming directly to the gemini.google.com chatbot.

   3. The Chirp 3 audio model is coming to the Vertex AI platform.

   4. New Interactive Mindmaps in NotebookLM turn your notes and sources into mind maps you can explore dynamically.

5. Krea AI now lets you train the Wan 2.1 video model on your own styles, objects, and motions.

6. Meta is finally rolling out Meta AI in Europe, limited to text-based chat inside WhatsApp, Instagram, Messenger, etc.

7. Mistral AI launched Mistral Small 3.1 which outperforms Gemma 3 and GPT-4o mini on text and multimodal tasks while also being faster.

8. NVIDIA released the Llama Nemotron family of open reasoning models post-trained for better accuracy and faster inference to power AI agents.

9. OpenAI news:

   1. o1 and o3-mini now support Python-powered data analysis in ChatGPT, so you can ask them to run complex data crunching and visualization tasks.

   2. There are three new audio models: Two speech-to-text models that are better than Whisper, and a text-to-speech model that can change “how” it speaks based on instructions. (Try the fun free demo here: openai.fm)

10. Pika Labs added a whole bunch of goofy video effects for you to play with.

11. Roblox has a foundational model called Cube 3D that lets users quickly create in-game 3D objects to interact with.

12. Tencent…

    1. …launched two new open-source 3D models: Hunyuan 3D 2.0 MV and 3D 2.0 Mini.

    2. …and Hunyuan-T1, a reasoning model that’s hyper fast, accurate, and competitive with GPT-4.5, DeepSeek-R1, and o1 on many benchmarks.

13. Topaz Labs released Gigapixel AI v8.3.0, its image upscaling and restoration tool—now capable of sharper, cleaner fixes and with a simpler interface.

    

14. Zoom added new agentic skills to its AI Companion, letting it execute tasks and manage meetings more effectively.

---

🔬 AI research

Cool stuff you might get to try one day:

1. Anthropic is reportedly working on a voice mode so you might soon be able to finally talk to Claude naturally.

2. China researchers introduced ReCamMaster, which can reimagine videos from new camera angles based on a single reference video.

3. OpenAI is working on ChatGPT connectors for business customers to link the chatbot to apps like Google Drive and Slack.

4. Pika Labs teased an object manipulation feature that lets you control specific characters or items in a video without affecting anything else in the frame. (Apply to the Pika Creative Partner program for early access.)

5. Stability AI previewed its Stable Virtual Camera that can turn even a single input image into a controllable 3D video view.

---

📖 AI resources

Helpful AI tools and stuff that teaches you about AI:

1. “6 tips to get the most out of Gemini Deep Research” [ARTICLE] - quick primer from Google on its now-free “Deep Research” feature.

2. “File prompting strategies” [ARTICLE] - another guide by Google on prompts to use along with your uploads. (Unless no-prompt prompting works in your case.)

---

🔀 AI random

Other notable AI stories of the week:

1. Google researchers demonstrated that you can make language models significantly smarter via sampling-based search, where they generate multiple responses and self-verify them.

🤦‍♂️ AI fail of the week

Me: “Hey, Sora, do you know parkour?”

Sora: “You bet!”

---

💰 Sunday Bonus #52: Tap into real-time local news in other countries and languages

Ever wanted to find out what’s currently happening somewhere in the world?

Not the watered-down international news headlines, but the stuff locals are reading in their native language?

Now that practically every language model is plugged into the Internet, you’d think getting up-to-date information from news sources would be a cakewalk.

Yet when it comes to developing news stories, web search in most models is sorely lacking in my experience.

Here’s a seemingly simple query that most chatbots somehow struggle with:

What are the top five tech headlines of the past 24 hours?

ChatGPT feeds me week-old news items and offers a weirdly prioritized story mix:

Google’s Gemini returns a vague “stuff is happening somewhere, probably” take. (If only there were a search engine that Google could use to dig up the latest news.)

Grok—despite being hooked up to the firehose of real-time content that is X—digs up headlines that are at least 3-5 days old.

Even Perplexity’s “Pro Search” includes DeepSeek-R1 news from back in January in its March 22 overview:

Things only get worse when you try to search for local news in foreign languages.

Every item above pulls from just a single German site, most of the headlines are at least several weeks old, and more than half of them aren’t even German local news.

This is a pattern.

My guess? Much of this is due to restrictions on which sites AI chatbots are allowed to crawl and the way fast-moving news is parsed by AI search features.

I’ve always felt this was a huge shame, seeing how most LLMs are multimodal and therefore perfectly positioned to bridge the gap between truly local news and readers in other countries.

How awesome would it be to get an English summary of foreign-language headlines from anywhere in the world?

Well, what if I told you there's a free yet surprisingly effective AI tool that, in my experience, does a better job of finding and parsing local news than most chatbots?

Let me show you the tool and the exact prompt I use to get local takes on any topic, with a worked example of them in action. I’ll even share a link to my fill-in-the-blanks template that lets you pull together varied local news insights in any country.