Sunday Rundown #125: Mini Models & Cocky Mascot
Sunday Bonus #85: Tool that automatically improves any prompt.
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 paid subscribers.
In case you missed it, here’s this week’s Thursday deep dive:
If you’re consistently missing out on my emails, remember to check your “Promotions” tab and mark whytryai@substack.com as a “Safe Sender.”
Let’s get to it.
🗞️ AI news
Here are this week’s AI developments.
👩💻 AI releases
New stuff you can try right now:
Alibaba open-sourced Qwen3-VL-Embedding and Qwen3-VL-Reranker, which help find and rank information across text, images, screenshots, and video.
Google is rolling out Gemini inside Gmail with tools that summarize emails, suggest and improve replies, and highlight what matters most in your inbox.
Lightricks open-sourced its LTX-2 video model with native audio, so you can download the weights and code and run it offline. (Get the model here.)
Microsoft is rolling out Copilot Checkout that lets shoppers buy items without leaving the chat, starting in the US.
NVIDIA news:
Nemotron RAG models make it easier to extract accurate information from text, images, audio, and video.
Nemotron Speech ASR lets you build low-latency transcription and voice agents with open models.
Nous Research released NousCoder-14B that builds on Qwen3-14B to deliver better competitive programming performance. (Get the model.)
TII open-sourced Falcon-H1R-7B, a compact reasoning model that delivers strong step-by-step thinking at lower cost. (Try on Hugging Face.)
🔬 AI research
Cool stuff you might get to try one day:
DeepSeek is preparing to release DeepSeek V4, rumored to outperform existing top-tier models on coding tasks.
OpenAI introduced ChatGPT Health that grounds your chats in data from medical records and wellness apps. (Join the waitlist.)
📖 AI resources
Helpful AI tools or stuff that teaches you about AI:
“AI Models” [DATABASE]—comprehensive database breaking down factors behind machine learning progress across 3,200 tracked models by Epoch AI.
“How Consumer Industries Are Adopting AI: Key Survey Insights” [SURVEY]—research summary from Ropes & Gray.
“State of AI in Retail and CPG” [REPORT]—NVIDIA’s annual survey on AI adoption.
🔀 AI random
Other notable AI stories of the week:
X restricted Grok image generation to paid subscribers after backlash over non-consensual sexualized images.
🤦♂️ AI fail of the week
That time I naively asked ChatGPT for a rooster-themed “Butt Nugget” mascot…
Send me your AI fail for a chance to be featured in an upcoming Sunday Rundown.
💰 Sunday Bonus #85: Fix any prompt with “Prompt Medic”
The “Image Prompt Cleaner” GPT is one of my most popular Sunday Bonuses. It automatically fixes confusing and overstuffed image prompts based on my “splatterprompting” insights and industry best practices.
It’s…pretty great.
But recently I thought: “Why just image prompts? And why just a chat-based GPT?”
So I went ahead and created "Prompt Medic,” a neat Claude Artifact that takes your prompt, categorizes it, automatically applies relevant best practices, and spits out an optimized version.
Like any fancy diagnostics tool, Prompt Medic gives you a neat count of solved issues in addition to the optimized prompt itself:
From here, you can simply one-click copy the new prompt and use it as is.
But if you’re feeling nerdy, Prompt Medic gives you a comprehensive overview of changes and feedback to learn from.
While Image Prompt Cleaner is still excellent for fixing image prompts, think of Prompt Medic as your new Swiss Army Knife for all kinds of requests.
I’m really happy with how it turned out, and I’d love to hear your feedback once you get to test it. All you need is a free Claude account.