Sunday Rundown #83: NVIDIA Blitz & Sad Bear

Jan 12, 2025

∙ Paid

Happy Sunday, friends!

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

Let’s get to it.

🗞️ AI news

Here are this week’s AI developments.

New stuff you can try right now:

Google is trialing a new experiment called “Daily Listen” that turns your Discover feed into a podcast a la NotebookLM’s “Audio Overviews.” (Currently only available via the Google app, only in the US, and only in English.)
Hailuo AI (“God tier” in my image-to-video test) has a new S2V-01 model that generates videos with consistent characters from a single reference image. (Try it here…but you might need to be very patient as everyone else is also doing it.)
NVIDIA made many announcements at CES, including these relevant AI ones:
1. New Nemotron model family for agentic AI:
  1. Cosmos Nemotron, which enables video search and summarization.
  2. Llama Nemotron, which helps developers create and deploy AI agents.
2. Custom AI Blueprint for video search and summarization built with the above models.
3. New ACE framework for creating truly autonomous video game characters powered by its LLMs.
Microsoft’s Phi-4 model (first announced a month ago) is now fully open-sourced on Hugging Face.
Stability AI is upping the stakes in 3D generation with Stable Point Aware 3D, capable of turning a single image into a fully realized 3D object. (Try the demo.)
xAI has made its Grok chatbot available as an iOS app in the US.

Cool stuff you might get to try one day:

Adobe Research showcased TransPixar, a video model that can turn text prompts and starting images into videos with transparent background.
VLC video player (I’m already a fan) previewed a feature that automatically subtitles videos in real time and can translate those subtitles into 100+ languages.

Helpful AI tools and stuff that teaches you about AI:

“The Definitive Guide to Building AI Agents in 2025” [VIDEO] - a great introduction to agents by AI Software Developers.
“Timelines Shift Forward” [VIDEO] - another solid dive from one of my favorite AI YouTube channels, AI Explained.

Serves me right for writing “Bearder man” instead of “Bearded man” by accident.
(Proper version used in this post.)

Did you know that you can ask ChatGPT to manipulate images?

Technically, this isn’t an innate abillity of ChatGPT itself, but it can use Python libraries to honor such requests.

While there are standalone tools for image editing, it’s often useful to have the option of doing this directly in ChatGPT.

Here are three different ways you can ask ChatGPT to manipulate uploaded images: