Generative AI News 2023-03-27
Bing Image Creator
Microsoft has launched an AI image generator, and it’s OK. It’s likely based on DALL-E 2. It seems to understand verbs and subjects, which other image generators struggle with. It seems to struggle with hands though. The following is the result of “Robot reading a newspaper”.
OpenAI Launches Plugins for ChatGPT
Giving the chatbot “eyes and ears” (and hands). Allowing it to do things like make a reservation (OpenTable) or buy ingredients for a recipe (Instacart).
GPT-4 Is (a Little Bit) AGI
In a paper by Microsoft Research, they state that GPT-4 exhibits remarkable capabilities across a variety of domains and tasks. The authors claim that GPT-4 is part of a new cohort of models that exhibit more general intelligence than previous AI models, and could be viewed as an early version of an artificial general intelligence (AGI) system.
Microsoft Loop - A NotionAI Knockoff
Organized by components, workspaces, and pages, Microsoft Loop App. In the demo video, it looks like a web based OneNote / Sharepoint-ish solution with generative AI.
Adobe Firefly
Way more than image generation. Developed ethically, on Adobe Stock, open-licensed images, and public domain images.
GitHub CoPilot X
“GitHub Copilot is evolving to bring chat and voice interfaces, support pull requests, answer questions on docs, and adopt OpenAI’s GPT-4 for a more personalized developer experience.”
ChatBot Tech Coming to Siri?
Compared to the capabilities of ChatGPT - Siri, Alexa, and Google Home seem… dumb. I’m sure they’re all working on backing these services with large language models, but people have found concrete clues that Apple is testing new AI language generation capabilities for Siri, code-named "Bobcat," as discovered in the tvOS 16.4 beta.
Dolly - The Databricks ChatGPT Clone
Databricks has released the code for an open-source large language model (LLM) called Dolly that can be used to create instruction-following chatbots. The model can be trained on very little data and in very little time. Databricks believes that most machine learning (ML) users are best served by directly owning their models rather than sending data to a centralized LLM provider that serves a proprietary model behind an API.
Thanks for reading!