. . AI: Google is releasing open AI? (2.21.24)

DeepMind, ByteDance, Nous Hermes

Friends, in a funny twist of fate, a few weeks ago we saw stories about OpenAI potentially launching search related products. Today we bring you Google, which is now bringing 'open AI' models. As these two power houses try and out disrupt each other, we have to zoom out way back to May 2023 and remember the internal memo that was leaked that said "we have no moat, and neither does OpenAI". Each week we talk to a number of AI founders and we are hearing that, amidst this competition, they are broadening their API infrastructures to integrate seamlessly with various providers like Google and OpenAI, ensuring plug-and-play compatibility with the top models of the day.

Thanks for reading and for sharing this newsletter with friends and colleagues.

Now here’s the news,

Marshall Kirkpatrick, Editor

First impacted: Software Developers, AI Model Developers
Time to impact: Short

Google and the team from Deepmind launched Gemma, a family of lightweight, open models built from the same research and technology used to create the Gemini models. The models come in two sizes: Gemma 2B and Gemma 7B, along with a new Responsible Generative AI Toolkit. These models can run on multiple hardware platforms including NVIDIA GPUs and Google Cloud TPUs. The models appear to be achieving strong performance for their sizes compared to other open models, and can apparently can run directly on a laptop or desktop. Check out more in the link. [Gemma: Introducing new state-of-the-art open models] Share this story by email

First impacted: AI researchers, Software Developers
Time to impact: Short

ByteDance has launched a text-to-image generative model named SDXL-Lightning, which the company says can produce high-quality 1024px images. The model is only currently released for research purposes. [ByteDance/SDXL-Lightning · Hugging Face] Share this story by email

First impacted: AI developers, AI researchers
Time to impact: Short

NousResearch has launched a new model, Nous Hermes 2 on Mistral 7B DPO, which they say shows improved performance in a variety of benchmarks. The model uses the ChatML prompt format, enabling a structured system for multi-turn chat dialogue and compatibility with the OpenAI endpoint. [NousResearch/Nous-Hermes-2-Mistral-7B-DPO · Hugging Face] Share this story by email

That’s it! More AI