What AI made possible last week

Last week, there were several AI announcements that I found to be highly relevant to bootstrapped founders looking to build news products:

AI video avatars have become frighteningly realistic

We are currently passing the Turing test for video. Last week, HeyGen released version 3 of their AI avatars which look… scary real:

Other companies are following suit: One of them is Beyond Presence, which raised $3.1 million to build lifelike avatars.

The implications for indie hackers: Video is everywhere. The problem is, most of us are shy (myself included).

These lifelike avatars will unlock marketing channels like YouTube and TikTok for us, enabling us to create videos for our (mostly software) products which have the potential to go viral.

YouTube will help you create background music with the help of AI

Let's say you've created a lifelike avatar and gotten it to say something. Now you want to add some nice background music to your video.

The news: Last week, YouTube announced a broader launch of its "Dream Track" AI audio generator. This enables you to create custom background tracks for your short videos.

The implication for us, again, is video is everywhere, and features like this will help a lot in making a video from a text prompt.

On-device AI is becoming more widespread

Last week, Mistral released two new AI models which can run on laptops and tablets.

The models: Ministral 3B and Ministral 8B, both of which have a context window of 128,000 tokens, meaning they can ingest roughly the length of a 50-page book.

The implications for us: There are lots of use cases where our users want the privacy of interacting with AI on their local devices rather than through the web browser. These models are also cheaper to fine-tune and run in general.

So before you decide to pay $$$ for the "top 3" models (ChatGPT, Claude, Google), try out smaller models like Mistral and see if they do the job.