• Starter AI
  • Posts
  • DeepMind's self-discover framework, Stability AI’s upgrade, Introducing: localllm

DeepMind's self-discover framework, Stability AI’s upgrade, Introducing: localllm

Creating incredible videos with AI gets even easier.

Author

Sthefania
February 08, 2024

Together with

Hello, Starters!

Thursday has arrived, and we’ll see you again on Monday with more AI goodness. But that doesn’t mean this issue lacks them; on the contrary, there’s a LOT to cover!

Here’s what you’ll find today:

  • DeepMind presents a “self-discover” framework for LLMs

  • Stable Video Diffusion gets upgraded

  • localllm: No GPU? No problem 

  • Google’s new feature for Lookout

  • OpenAI is reportedly working on “agents”

  • And more.

Researchers from Google DeepMind and the University of Southern California have introduced a prompting framework to enhance the reasoning capabilities of LLMs. "Self-discover" has been shown to improve the performance of popular models such as GPT-4 and PaLM2.

This framework is based on LLMs "self-discovering" task-intrinsic reasoning structures to find solutions. Essentially, the model will consider certain reasoning modules, like critical thinking or step-by-step thinking, and then choose the right technique for the task at hand.

Stability AI is well-known in the AI landscape for its state-of-the-art open models, such as Stable Diffusion and Stable Video Diffusion. The generative AI company constantly releases upgrades that leave users in awe of its realistic results.

Recently, a major update was announced for Stable Video Diffusion. The video model, SVD 1.1, represents a significant advancement over its predecessor, enhancing motion and consistency in generating four-second videos with 25 frames and a resolution of 1024 x 576 pixels.

Google has found an innovative solution to one of the biggest problems developers currently face when creating AI applications: the lack of GPUs. Thanks to Google Cloud's Cloud Workstations, developers can unlock the full potential of LLMs locally on CPU and memory.

This is achieved by using quantized models available on Hugging Face, which are optimized to run on local devices with limited resources. In addition to eliminating the need for GPUs, this approach also opens up new possibilities for AI development.

🔍 Google's Lookout app leverages an AI model in its Image Q&A feature that allows visually impaired users to ask questions about a specific image. By just uploading the pic, the user will get a detailed description of it, including colours, characteristics of a subject, text, and more.

🧑‍💻 According to insiders, OpenAI is currently working on new assistants called "agents." This type of software aims to automate complex tasks by assuming control of a user's device. For example, it will be able to create itineraries or even book flight tickets without the need for human commands.

Microsoft brings new design-focused features to Copilot

How much AI do you know?

Login or Subscribe to participate in polls.

Sponsored content

Stay up-to-date with AI.

AI won’t replace you, but a person using AI might. That’s why 500,000+ professionals read The Rundown– the free newsletter that keeps you updated on the latest AI news, tools, and tutorials in 5 minutes a day.

Thank you for reading!