AI Newsletter #95 - Design platform Canva launches text-to-image AI powered by Stable Diffusion

Welcome to Nural's newsletter focusing on how AI is being used to tackle global grand challenges.

Packed inside we have
All about image generation models this week!

  • Design platform Canva launches text-to-image AI feature powered by Stable Diffusion
  • Easily Accessible Text-to-Image Generation Amplifies Demographic Stereotypes at Large Scale
  • and Meet Prompt-to-Prompt: An Artificial Intelligence AI Model That Brings Image Editing Capabilities to Text-to-Image Models

If you would like to support our continued work from £1 then click here!

Marcel Hedman


Key Recent Developments


Design platform Canva launches text-to-image AI feature

Canva’s text-to-image tool

https://www.theverge.com/2022/11/10/23450965/canva-text-to-image-ai-tool-free-users

What: Generative AI has gotten a lot of attention in 2022 and we have seen the rapid transition of this technology from research towards mass adoption by non-technical end users.

The latest to join this trend is design tool Canva, who have recently launched their own text-to-image AI feature. Their tool is powered by open source model Stable Diffusion with a few extra safety filters and a custom UI to help steer Canva’s users to get the results they want.

Key Takeaway: The embedding of generative models into everyday tools is rapidly on the rise. The news of Canva's mass use tool follows Microsoft's announcement of text-image models DALL-E into their Office suite and the launch of GitHub Copilot to allow for code generation. Where else will we see these generative applications pop up...


Meet Prompt-to-Prompt: An Artificial Intelligence AI Model That Brings Image Editing Capabilities to Text-to-Image Models

Meet Prompt-to-Prompt: An Artificial Intelligence AI Model That Brings Image Editing Capabilities to Text-to-Image Models
It is okay to assume everybody has heard about the Stable Diffusion or DALL-E at this point. The huge craze about text-to-image models has taken over the entire AI domain in the last couple of months, and we have seen really cool executions. Large-scale language-image (LLI) models have shown extrem…

What: With text-to-image models, it can be difficult to edit a generated image. When changing the intial prompt, typically the whole image will change given the stochastic nature of the underlying models. As such, researchers have created a model to enable such image editing capabilities through an amendment of the intial prompt. Thus preserving the original image but enabling for small tweaks to be made.

Prompt-to-prompt editing results. Source: https://prompt-to-prompt.github.io/

Easily Accessible Text-to-Image Generation Amplifies Demographic Stereotypes at Large Scale

Link to Paper

Biased outputs from a text-to-image model
Biased outputs from a text-to-image model

What: Researchers have conducted a review of existing generative image models and have found the underlying bias to be a cause for serious concern. They saw "that the amplified stereotypes are difficult to predict and not easily mitigated by users or model owners." These stereotypes spanned across race, gender and class lines.

Key Takeaway: As noted earlier in this newsletter edition, we are seeing the increased transition of generative image models into the hands of mass consumers. This means millions of images will be created daily which in turn will shape our belief about what is true. What will be the effect from the long term amplification of demographic stereotypes at scale?

Link to Paper


AI Precision Medicine Partnership Aims to Improve Chronic Disease Outcomes

https://healthitanalytics.com/news/ai-precision-medicine-partnership-aims-to-improve-chronic-disease-outcomes

What: "Maryland-based MedStar Health and precision medicine technology company Zephyr AI, Inc., have announced a partnership focused on developing artificial intelligence (AI)-enabled precision medicine tools to improve outcomes for chronic diseases, starting with type 2 diabetes (T2D)."

Why this matters: The shift away from a population level to an individual view of healthcare leads to tailored therapies and medicine for patients, unlocking huge health benefits. This individual view is something that can be opened using the power of ML and this partnership represents a significant step towards this personalised approach.


AI Ethics & 4 good

🚀 Alphabet X’s new effort to combat climate change with seagrass - cameras, computer vision, and machine learning to track the carbon stored in the biomass of the oceans

🚀 Amazon’s leaked ‘Clinic’ would connect patients to telemedicine

🚀 Dirty shrimp farms are punching a huge hole in the environment. A.I. could cut it in half

🚀 AI System to Leverage Social Media to Predict, Prevent Future Pandemics

Other interesting reads

🚀 5 takeaways from the FDA’s list of AI-enabled medical devices - 91 authorized in 2022 alone

Number of approvals and clearances by the Food and Drug Administration per year for AI-enabled medical devices

🚀 Zoom to Launch Virtual Co-Working Space and AI Chatbot

🚀 Nvidia establishes autonomous driving units in Israel

🚀 “Too easy”—Midjourney tests dramatic new version of its AI image generator

Eight images generated with the alpha version of Midjourney v4.

Cool companies found this week

Health

Zephyr AI - Zephyr AI is building novel transparent AI to address unmet needs across the full spectrum of Precision Medicine from prevention to treatment to improve patients’ lives.

Climate

Tidal - Tidal’s underwater camera system and machine perception tools bring visibility to ocean ecosystems so we can better understand and protect them.


...and finally

How-To Guide: Build a Website to Talk to GPT-3 using Bubble and OpenAI in 21 Easy Steps (Beginner)

AI/ML must knows

Foundation Models - any model trained on broad data at scale that can be fine-tuned to a wide range of downstream tasks. Examples include BERT and GPT-3. (See also Transfer Learning)
Few shot learning - Supervised learning using only a small dataset to master the task.
Transfer Learning - Reusing parts or all of a model designed for one task on a new task with the aim of reducing training time and improving performance.
Generative adversarial network - Generative models that create new data instances that resemble your training data. They can be used to generate fake images.
Deep Learning - Deep learning is a form of machine learning based on artificial neural networks.

Best,

Marcel Hedman
Nural Research Founder
www.nural.cc

If this has been interesting, share it with a friend who will find it equally valuable. If you are not already a subscriber, then subscribe here.

If you are enjoying this content and would like to support the work financially then you can amend your plan here from £1/month!