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

Packed inside we have

  • The State of Multilingual AI
  • Why Meta’s latest large language model survived only three days online
  • and AEYE Health gets FDA clearance to use AI to screen diabetics in hopes of preventing blindness

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

Marcel Hedman

Key Recent Developments

Why Meta’s latest large language model survived only three days online

Meta Trained an AI on 48M Science Papers. It Was Shut Down After 2 Days
Galactica was supposed to help “organize science.” Instead, it spewed misinformation.

What: "Meta AI (formerly known as Facebook Artificial Intelligence Research) recently released an artificial intelligence named Galactica  with the intention of using machine learning to "organize science." It's caused a bit of a stir since a demo version was released online last week, with critics suggesting it produced pseudoscience, was overhyped and not ready for public use." A a result it was shut down in 2 days.

It was found that there were instances where the model struggled to do basic math and some of the citations attached to research seemed to be fabricated as well.

Key Takeaway: There is clearly a massive gap between a model that produces grammatically correct strings of words and phrases vs one that can produce content grounded in truth.  

The State of Multilingual AI

The State of Multilingual AI
This post takes a closer look at the state of multilingual AI. How multilingual are current models in NLP, computer vision, and speech? What are the main recent contributions in this area? What challenges remain and how we can we address them?

What: The growth of large language models is undeniable with popular models including BERT, GPT-3, BLOOM. However, many of these models are focused on the English language. With only 17% of the world speaking English, there is clearly a lot of room to explore multilingual approaches.

This post covers the current status of efforts to address the non-English based AI language models.

Key Takeaway: Challenges to the expansion to a multilingual AI world include:

  • A lack of non-english training data
  • Multiple language on a single model often leads to lower performance in the same way a multitask modelling approach would
  • Quality issues in existing multilingual resources
  • Lack of evaluation benchmarks

AI Ethics

🚀 AI Helps Ukraine - Charity Conference

🚀 Ubisoft and Riot are going to use AI to stop you from being horrible online

🚀 Tax Filing Websites Have Been Sending Users’ Financial Information to Facebook

🚀 To Regulate General Purpose AI, Make the Model Move

🚀AEYE Health gets FDA clearance to use AI to screen diabetics in hopes of preventing blindness

Other interesting reads

🚀 Tesla reports two new fatal crashes involving driver assistance systems

🚀 Amazon Robotics: Teaching robots to stow items presents a challenge so large it was previously considered impossible — until now

🚀 MIT solved a century-old differential equation to break 'liquid' AI's computational bottleneck

🚀 Get ready to swipe right on some AI-generated profile pics

🚀Intel's new deepfake detector can spot a real or fake video based on blood flow in video pixels

Cool companies found this week


AEYE Health - Applying AI expertise to retinal imaging to deliver broad diagnostic screening solutions that address care gaps in primary care. Recent news: "FDA clears AEYE’s autonomous screening system for diabetic retinopathy featuring groundbreaking diagnostic accuracy, >99% imageability and using only one image per eye"

Speech Recognition

Vosk - Practical speech recognition library which comes with a set of accurate models, scripts, practices and provides ready to use speech recognition for different platforms like mobile applications or Raspberry Pi. If you want to build practical applications with plug and play library, consider Vosk.

Computer Vision

Spot AI - Provides a cloud-based analytics system that “reads” CCTV and other kinds of security camera footage to get insights about not just security, but also safety and operational activity. They have just raised $40 million in funding.

and Finally...

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.


Marcel Hedman
Nural Research Founder

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!