🧠 AI Weekly Digest vol.15
by Federico German
Published in
2 min readAug 8, 2023
Finally, the new version of stable diffusion is out there, and it seems pretty great.
- It seems like decades have passed since Sam Altman pleaded before the US Senate for regulations. Time has passed, and the AI giants, Google, Microsoft, OpenAI, and Anthropic, have decided not to wait. They decided to create a forum for the “safe and responsible development of frontier AI models.”
- AI and robots have been working together for quite a long time. Now, Google is presenting a new model that combines vision and language to generate robot actions.
- Multimodal models are the next big thing. In this work they talk about not only images but also 3d information injected into the LLM
- Remember the days when you got an error and promptly went to Stack Overflow? It seems that, for half their users, that’s only a memory. So, Stack Overflow joined the AI craze with a model trained on all their data. Maybe this will keep them relevant.
- We all have been waiting for this. SDXL 1.0 is here, it’s great, and it’s open source. You can run it locally or in a colab.
- Stability AI keeps releasing LLM models. Now, FreeWilly is free to battle ChatGPT and all the others. According to their tests, it can hold its own against the greats, but I would like to try that for myself. Fortunately, like everything else from Stability.ai, it is open source.
- Computer-brain interfaces are going to come sooner or later. This paper shows how to reconstruct music from brain activity.
- Do you feel that LLM-based applications are safe? Think again. A group of researchers claim to have found an adversarial attack that works on every LLM.
