ChatGPT iOS, Calls for Regulation, HuggingFace Agents, and Drag Your GAN
From ChatGPT iOS to HuggingFace Agents, it's been a busy week in the world of machine learning. Here's a roundup of some of the latest developments in the field.
Created on May 21|Last edited on May 21
Comment
ChatGPT iOS
OpenAI's newest ChatGPT installment is their brand-new iOS app! This does pose an interesting question: how will ChatGPT compete against Alexa, Cortana, or Siri? The same text-to-speech models could be used as an extension to ChatGPT.
Integrating newer LLMs may change the way we interact with our virtual assistants. Asking questions and setting up calendars and schedules may be done more dynamically and naturally via language, with a lot more room for flexibility and plug-ins!
Call for Regulation
In other OpenAI news, Sam Altman recently testified in front of a congressional committee calling for regulating AI. There's lots of speculation surrounding the true intentions of regulation. Could it be for centralizing AI into the hands of a few? For countering the AI development in other companies and organizations?
HuggingFace Agents
HuggingFace recently released a new experimental library called Agents! It's a natural language API wrapper on HuggingFace Transformers. It seems like a natural HuggingFace-supported extension of the LangChain library.
From their Quickstart page, Agents has three modes of execution: single run, chat-based, and remote execution. Single runs are for single-command prompts, while chat-based ones are more conversational. Remote execution is built with inference endpoints and is a flag in both the .run() and .chat() methods.
The overall architecture is detailed above in their diagram. A user instruction is supplemented by a prompt engineered behind the scenes. The HuggingFace Agent decides and applies which tools to use. The results are returned to you in a user-friendly manner.
A few of their core supported tasks include:
Building LLM and smart AI apps has never been easier!
Drag Your GAN
Drag Your GAN is a point-based manipulation of GAN-generated images. GANs have always had limited control over the actual orientation and detail of the resulting image. The past few years of GAN research have dramatically increased the user's control over the GAN's generated image (though diffusion models have picked up greater traction in recent years). The project page provides a nice interface for dragging points on demo images.
References
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.