The week's developments in AI, quantum, and biotech, explained | 23.07.24

If you’re enjoying this briefing, sign up here — and share it with a friend. Reading time: ~4 minutes

This week, scientists warn of AI chatbots spiralling into gibberish when trained on too much AI-generated data, star-gazing tech finds a new purpose in spotting digital fakes, and researchers discover deep-sea metal nodules that can produce oxygen without light. Plus, Alzheimer's diagnostics get a Cambridge AI boost.

— Charlie and the Research & Intelligence Team

🚀You are now able to sign up for the super early bird offer (50% off) to the CogX Leadership Summit in London on October 7th here.

Innovation and Releases

🤖 Turns out AI’s worst enemy is its own data: When chatbots learn primarily from their own output, they can degrade and start producing nonsensical replies. This "model collapse" happens because AI models overfit on popular data, meaning they “forget” the less-common elements of their original training data.

🔭 A tool to combat deep fakes: Techniques used in astronomy are helping researchers identify real photos from AI-generated fakes by analysing how light reflects in eyeballs. While not foolproof, this breakthrough offers a powerful new tool in the fight against deep fakes and manipulated media.

🌑 'Dark oxygen' from the deep-sea floor? Scientists have discovered deep-sea "battery rocks" that produce oxygen without light or the presence of living organisms. This "dark oxygen" throws a wrench into our understanding of how life began — and could strengthen the case against deep-sea mining.

AI Tools of the Week

🔧Compare multiple LLMs side by side: Nailedit.ai streamlines AI content generation by enabling simultaneous comparison of outputs from multiple large language models.

🎨 Create 3D models from prompts: Artefacts.AI is an AI-powered platform that transforms text descriptions or images into 3D models.

💰Streamline your workflow: Zerve AI provides a unified workspace for streamlined data science & AI projects.

Latest Research

🧠 Alzheimer’s diagnostics gets an AI boost: Cambridge scientists have developed a new AI tool that predicts Alzheimer's development with 82% accuracy, far exceeding current methods. The tool uses only routine cognitive tests and MRI scans, eliminating the need for expensive and invasive procedures.

👨‍🚀 Why meals taste bad in space? A new study suggests the space station environment itself intensifies some food smells, particularly sweet ones like vanilla and almond. This breakthrough could lead to personalised food experiences for astronauts, improving their diets in space.

💊 Scientists have unravelled a key part of the placebo effect:  A brain circuit involving the brainstem and cerebellum has been linked to placebo pain relief in mice. This discovery could lead to entirely new ways to treat pain, without relying on medications.

🌖 Anybody want moon water? The UK Space Agency is awarding £30k in funding  to 10 innovative teams to develop solutions for making clean water on the moon. Ideas include using microwaves, sound waves, and even repurposed Tesco appliances.

In case you missed it

We've been wowed by videos of robots doing parkour and helping to sort laundry. So, why don't we have better robots yet? Despite impressive feats, it seems like we're still a ways off from having our very own robot butlers. 

✍️ Enjoying this newsletter?  Subscribe to our free weekly briefings on Preparing for AI, Cinema & AI, The Race to Net Zero, AI & DeepTech and The Future of Work.

🚀 You are now able to sign up for the super early bird offer (75% off) to the CogX Leadership Summit in London on October 7th here.