Sunday Rundown #111: Everybody Builds & Javelin Jack
Sunday Bonus #71: Prompt Crafter that turns vague ideas into full prompts.
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.
In case you missed it, here’s this week’s Thursday deep dive:
I was gone last weekend, so we have an extra week of AI news to catch up on.
Let’s get to it.
🗞️ AI news
The last two weeks of AI developments.
👩💻 AI releases
New stuff you can try right now:
Alibaba news:
Qwen3-ASR-Flash is a multilingual speech recognition model that can handle noisy/choppy audio and transcribe songs with background music.
Qwen3-Next is a hybrid attention language model with a mixture-of-experts design, making it comparable to larger LLMs while being 10x faster.
Anthropic news:
Claude can now create and edit files like Excel, PowerPoint, and more. Available on Max and Teams plans, with Pro plans soon to follow.
The new Memory feature for teams lets Claude remember details and context across your entire team working on the same project.
Baidu news:
ERNIE-4.5-21B-A3B-Thinking is a long-context (128K) reasoning model with efficient tool use and high scores on math, coding, and science benchmarks.
ERNIE X1.1 is a reasoning model that hallucinates less, accurately follows instructions, and matches GPT-5 and Gemini 2.5 Pro on many benchmarks.
ByteDance released Seedream 4.0, a top-tier image model that excels at both generating and editing images. (I covered it in detail here.)
ElevenLabs news:
Agents Testing lets you safely evaluate your voice assistants before publishing them.
Productions is a paid end-to-end service that offers AI-generated captions, dubbing, transcripts, and audiobooks refined by human experts.
Google news:
AI Max for Search campaigns is a one-click solution for advertisers to experiment with and improve their Google Ads campaigns.
AI Mode is now available in five additional languages to reach more users worldwide.
AI Quests is a collection of educational games that teach students AI literacy through hands-on challenges and tasks. (Try it here.)
Canvas in Gemini added a Select and Ask feature that lets you visually highlight specific areas to edit.
Chrome got a major feature drop with built-in Gemini, multi-tab summaries, AI-powered search, safer browsing, and much more.
Gemini app finally lets you share your custom Gems with others. (I look forward to building some Sunday Bonus goodies with this.)
NotebookLM added Flashcards and Quizzes as output options to help everyone learn the source material.
Veo 3 got cheaper and added support for 1080p resolution and vertical video.
Kling AI launched Avatar, letting you animate a character from an input image, audio, and prompted expressions and emotions.
Luma AI launched Ray3, the first reasoning, physics-aware video model that can generate 4K HDR footage and comes with a “Draft” mode for fast iterations.
MiniMax released Music 1.5, which can generate complete four-minute songs with natural vocals and rich instrumentals (Try for free.)
Notion launched Notion 3.0, complete with a team of AI agents that can create and update docs, databases, workflows, and more on your behalf.
OpenAI news:
GPT-5 Codex is faster, more reliable, and can debug code or work independently on projects.
ChatGPT now has full MCP tool support, so you can build connectors to trigger workflows, automations, etc. in developer mode.
Reve relaunched Reve Image with drag-and-drop image editing, a creative assistant, and API access.
Stability AI released Stable Audio 2.5, an enterprise-grade audio model that can generate commercially safe sound effects and music clips of up to 3 minutes.
Tencent open-sourced HunyuanImage 2.1, a prompt-adherent image model that produces images at 2K resolution. (Try it for free on Hugging Face.)
Udio introduced Voices, which lets users create or select consistent vocals for their music tracks.
WorldLabs upgraded its 3D world model with better geometry, larger environments, and more. (Try for free during public beta.)
YouTube news:
YouTube Shorts has new AI tools like Veo 3 video generation with sound, photo animation and editing, and soundtrack creation.
YouTube Studio also has new AI-powered features like the Ask Studio chat assistant, auto-dubbing with lip sync, and other tools to grow your channel.
🔬 AI research
Cool stuff you might get to try one day:
Adobe is expected to soon incorporate Nano Banana (Gemini 2.5 Flash Image) directly inside Photoshop.
OpenAI might be adding an Orders feature to ChatGPT, with native checkout and order tracking, so it can function as a shopping hub.
Perplexity’s Comet browser may soon support local MCPs, letting the Comet Assistant access local files and apps.
📖 AI resources
Helpful AI tools and stuff that teaches you about AI:
“Anthropic Economic Index: Understanding AI’s effects on the economy” [REPORT] - excellent interactive page with lots of useful insights (US based).
"Detecting and reducing scheming in AI models” [RESEARCH] - a research paper on alignment by OpenAI.
“MCP Registry” [REFERENCE] - new resource from GitHub that helps you discover relevant MCP servers.
🤦♂️ AI fail of the week
His technique is unconventional, but it gets the job done!
Send me your AI fail for a chance to be featured in an upcoming Sunday Rundown.
💰 Sunday Bonus #71: Turn vague requests into polished prompts with Prompt Crafter
I’ve long argued against uncritically reusing other people’s off-the-shelf prompts:
I also once shared several ways to get chatbots to do their own prompting:
This week, I thought: “Wouldn’t it be cool to have a tool that incorporates these lessons and helps someone turn their initial ask into a fleshed-out prompt?”
So then I went and made one, because I’m a doer, dammit!
Say hello to Prompt Crafter.
This Claude Artifact takes your half-baked idea, interviews you, and spits out a complete prompt to use with your favorite chatbot.
Here’s how it works:
Step 1: You type in your rough request.
Step 2: Claude asks you follow-up questions tailored to your situation.
Step 3: Claude drafts a structured prompt using the CRAFT framework (Context, Role, Action, Format, Tone). You can review and tweak it.
Step 4: Claude provides a formatted version you can copy or save as a text file.
Here it is in action:
Note: Using the Prompt Crafter requires a free Claude account. If you don’t feel like making one, I’ll also share my custom GPT that does largely the same thing via chat.
Here you go: