Sunday Rundown #98: Everybody Draws & Inflatable Ninja

Sunday Bonus #58: Custom GPT that helps you pick the right OpenAI model.

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. Amazon released an Enhance My Listing tool that helps sellers auto-generate product titles, descriptions, missing details, etc.

2. Anthropic added web search to its API, letting developers ground Claude with real-time data.

3. Figma news from its Config 2025 conference:

   1. Figma Buzz lets teams co-create branded assets like social posts, ads, and event materials.

   2. Figma Draw is a powerful vector illustration tool with brushes, dynamic strokes, and texture effects.

   3. Figma Make is a prompt-to-code tool that turns designs into interactive prototypes with working components and internal logic.

   4. Figma Sites lets people design, build, and publish responsive websites directly in the Figma workflow.

4. Genspark launched AI Sheets, an agentic spreadsheet tool that analyzes, auto-fills, and visualizes data from web searches or uploaded files.

5. Google news:

   1. The new Gemini 2.5 Pro (I/O Edition) sits at #1 on the WebDev Arena Leaderboard for coding and can turn video input into interactive apps.

   2. A new study on text simplification resulted in a Simplify feature for the Google iOS app that can shorten complex text without losing its meaning.

   3. Gemini 2.0 Flash Image Generation now has better visuals, more accurate text rendering, and lower censorship. (Try it on Google AI Studio - select “Gemini 2.0 Flash Preview Image Generation” in the model dropdown.)

6. HeyGen launched Avatar IV, which can turn a single photo and a spoken script into a lifelike talking avatar video. (But does that make deepfakes too easy?)

7. Higgsfield launched Effects Mix, letting creators blend multiple video effects into creative, unique combos.

8. Hugging Face released Open Computer Agent powered by Qwen-VL models, which can navigate the web and click on-screen elements. (Try the demo here.)

9. Lightricks open-sourced LTXV 13B, a video model that’s 30x faster than competitors, capable of multi-keyframe conditioning, and can run on consumer hardware.

10. Microsoft added new experiences for Copilot+ PCs, including Recall for timeline search, Copilot Vision for on-screen analysis, and new AI-powered workflows.

11. Mistral AI released Mistral Medium 3, a frontier model that delivers GPT-4o-level performance at 8X lower cost and is optimized for coding and STEM.

12. Netflix launched a redesigned TV experience featuring AI-powered search and real-time, mood-based recommendations.

13. NVIDIA news:

    1. Apriel Nemotron 15B is a compact LLM built for enterprise AI agents that offers fast and affordable reasoning for real-time workflows.

    2. Parakeet-TDT-0.6B-v2 is an open-source speech recognition model that can accurately transcribe and time-stamp English-language audio. (Try the demo.)

14. OpenAI news:

    1. Reinforcement fine-tuning with o4-mini and supervised fine-tuning for GPT-4.1 nano let developers tailor models for complex use cases and domains.

    2. Deep Research can now connect to GitHub repos and return detailed reports with citations based on source code and PRs.

15. Recraft launched Advanced Style Creation and Control, which lets you pick from an infinite library of styles and remix them into your own consistent look.

16. Windsurf rolled out Wave 8, with lots of new features for teams and enterprises.

---

🔬 AI research

Cool stuff you might get to try one day:

1. Alibaba proposed ZeroSearch, a method that trains LLMs to search the web without relying on search engines, making information retrieval much cheaper.

---

📖 AI resources

Helpful AI tools and stuff that teaches you about AI:

1. “ChatGPT Enterprise - Models & Limits” [ARTICLE] - handy guide by OpenAI on when to use each of its primary models.

---

🔀 AI random

Other notable AI stories of the week:

1. OpenAI news:

   1. The company is becoming a Public Benefit Corporation (PBC) with the non-profit arm retaining overall control.

   2. It is now officially set to acquire Windsurf for about $3 billion.

🤦‍♂️ AI fail of the week

Technically, it’s my fault for requesting an “absurdist ninja fight.”

---

💰 Sunday Bonus #58: Pick the right OpenAI model with my Custom GPT

If there’s one thing OpenAI sucks worse at than naming its models, it’s helping you decide when to use each of them:

But what if I want fast everyday reasoning for slow visual coding tasks?

Sam Altman once hinted they’d like to kill the model picker altogether.

But until that glorious day, we’re stuck navigating the ever-shifting lineup in ChatGPT.

That’s where my “Chat Model Picker” GPT comes in.

Just describe your task, and it’ll recommend the best model for the job (sometimes asking a few follow-up questions to get it right).

It’s pretty straightforward, but it does the trick!