August's 'August' AI Announcements
Today and yesterday’s AI announcements and releases:
Despite the name, OpenAI’s models have not been open-source since GPT-2, from 2019. However, today they released open-source reasoning models that do well on benchmarks; we’ll see how they hold up to real usage. GPT-5 is rumored to be around the corner, so OpenAI may feel like they have to less to lose by releasing a powerful open-source model.
AI news from Google (note: I am employed by Google, but this Substack represents my own opinions):
Genie 3 - Genie goes beyond video generation to generate actual 3D environments. While previous releases were able to generate shorter demos, Genie 3 generates consistent, ongoing 3D worlds on-the-fly. I didn’t expect something like this to be possible yet, it will be cool to try this once it’s released.
Gemini released a slightly random feature - the ability to generate storybooks inside Gemini from a single prompt. Here is a storybook I created based on the recent podcast episode on meditation.
Kaggle will host a game arena for AI models to duke it out against each other, starting with chess. This is a little random since LLMs aren’t made for chess, although they’ve gotten better over time. It will also be interesting to see how they play on a variety of different games, as they’re added to the platform. I wonder if Google will try to inject a little AlphaZero-style thinking into their LLMs.
Not to be left out of announcements, Anthropic released a smaller update, Claude Opus 4.1. They also mention more releases coming soon.
In other areas, the AI voice company Eleven Labs announced a music generation service; they’re collaborating with the music labels to avoid lawsuits. For now I will still try Suno, here’s a song inspired by the recent meditation podcast (text generated in AI Studio).
Bonus link, may comment on it in the future: ChatGPT and the Meaning of Life: Guest Post by Harvey Lederman
Stay tuned for bigger announcements!