This is a short summary of this week's stories from my newsletter, AI'm informed. Get the full versions straight to your inbox, 3 days per week -?sign up here.
- Mustafa Suleyman, co-founder of DeepMind and Inflection, said that Generative AI is just a phase. The next step will be Interactive AI, bots that can carry out tasks for us. Like Yann LeCun, he’s not worried about LLMs causing human extinction, and instead keeps his focus on practical problems like regulation, bias, and data privacy.
- There’s a cool new approach in self-driving systems and it’s called VLAM, vision-language-action models. Lingo-1 is leading the way, you can talk to it while driving and it explains its decisions on the road.?
- Deloitte presented 60 potential use cases across six industries in their Generative AI Dossier. It’s lacking a bit in real-word examples, but it’s good inspiration if you’re thinking about building something or using GenAI in your company.
- An awesome project I stumbled upon is Petals, a community-driven, BitTorrent-like system where people pool GPUs together so that you can run inference and fine-tune on your local machine or Google Colab.
- Even players in the big leagues stumble. Microsoft's AI team accidentally leaked a whopping 38 TB of private data, showcasing the risks involved in large-scale, open-source AI development.
More stories and links to sources in the full edition of my newsletter (scroll down the page to explore all past editions in the archive).
