Mystery AI ✨ Patient Deaths Down 31% 💊 AI Combats Bears 🐻

Your weekly AI wrap

May 02, 2024

This Week in AI is Changing

This newsletter is about the future of work and how it is evolving with AI, however along with the weekly digest of events, soon you will also get a weekly UX teardown of the best AI products on the market.

Mystery AI Chatbot with Advanced Skills Appears and Disappears

A new AI chatbot called "gpt2-chatbot" briefly appeared online, demonstrating abilities similar to cutting-edge models like GPT-4.

Created by an unknown source, it wowed users with its ability to solve complex math problems and even code games.

Its origins are a mystery, with many speculating it's a secret test version of OpenAI's next-generation GPT-4.5 or GPT-5.

OpenAI’s CEO Sam Altman even posted this cryptic post:

Its sudden appearance and disappearance highlight a lack of transparency in AI development.

Why it matters: This incident underscores the secrecy and hype surrounding competitive AI research. It raises questions about the motivations behind anonymous testing and potential use of testing platforms as marketing tools.

This lack of transparency could hinder the evaluation of AI progress and potentially erode public trust in this rapidly evolving technology.

Lastly, the chatbot's brief appearance could suggest a potential new rival to major players like OpenAI and Google, fuelling excitement and questions about the future direction of AI.

📝 Gizmodo Article on the mystery model

𝕏 Great thread on the saga

AI Reduced Overall Deaths Among High-Risk Patients by 31%

Google's new AI model, Med-Gemini, surpasses previous AI systems in medical reasoning tasks. It combines advanced language understanding with the ability to process images, videos, and other medical data.

This gives it the potential to analyse complex patient information, search for relevant medical facts, and assist doctors in making more accurate decisions.

Med-Gemini has achieved record-breaking scores on medical benchmarks and, with further development, could lead to personalised healthcare.

Why it matters: This breakthrough highlights the potential for AI to enhance patient care by supporting doctors with complex diagnoses, providing comprehensive analysis of patient data, and potentially flagging high-risk cases early on.

📰 Article by New Scientist

📝 Read the research paper here

First Wave of AI Gadgets Disappoints, But Hope Remains

Recently launched AI-focused devices like the Rabbit R1 and Humane AI Pin have faced criticism for lacking basic features and overall poor performance.

While disappointing, this is common with first-generation products in a new category. Iterative design is key for improvement.

Why it matters: These early releases highlight the challenges of pioneering new AI hardware. Despite the initial setbacks, the potential for future iterations remains high.

This bumpy start underscores that even groundbreaking technologies need time for refinement to reach their full potential.

Japan Combats Bear Attacks With AI Warning System

Japan is facing a surge in bear attacks, prompting the launch of a pilot AI system in Toyama Prefecture.

The system will use cameras to identify bears, predict their movements, and issue immediate warnings to authorities. This effort aims to prevent further attacks and ensure public safety.

Why it matters: This project demonstrates how AI can be used to address real-world safety concerns.

The system tackles a problem likely caused by factors like climate change and reduced natural food sources for bears. Its success could lead to wider adoption in other areas facing human-wildlife conflicts.

📰 Article on News18

China Debuts Vidu, a Rival to OpenAI's Text-to-Video AI

Chinese start-up Shengshu Technology, backed by Tsinghua University, has launched Vidu, an AI tool that generates short videos from text descriptions.

While currently limited to 16-second clips, Vidu demonstrates China's progress in this emerging AI field.

Why it matters: Vidu highlights China's ambition to become a leader in generative AI, even facing challenges like restricted access to advanced computing hardware.

This development suggests that the text-to-video AI race is rapidly intensifying.

𝕏 Examples from the model

Adobe's VideoGigaGAN AI Improves Blurry Videos Resolution by 8x

Adobe researchers have developed a new AI model called VideoGigaGAN that can dramatically upscale the resolution of blurry videos.

It offers superior results to existing methods, producing sharper images without the distracting visual artifacts that often accompany AI upscaling.

Why it matters: This technology could transform how old footage is restored, adding clarity and detail while minimizing the "AI weirdness" of previous methods.

Additionally, VideoGigaGAN has the potential to improve the quality of low-resolution videos and become a powerful tool for video editors working to enhance the visual impact of their content.

📰 Article by the Verge

📝 Read the research paper here

Meta's AI: Is More Really Better?

Meta is aggressively pushing AI features across its platforms, with its "Meta AI" chatbot now prominently featured in Instagram search and Facebook groups.

However, users are finding the AI intrusive, misleading, and often indistinguishable from genuine content.

Why it matters: Meta's AI integration highlights its ambition but also raises serious concerns about user experience and online authenticity.

The flood of both company-generated and user-created AI content blurs the line between real and fake, potentially fostering confusion and distrust.

📰 OpenAI's blog post

Apple Assembles Secretive AI Research Lab in Zurich, Luring Talent from Google

Apple has quietly established a European AI research lab in Zurich, attracting dozens of top AI experts from Google.

This secretive "Vision Lab" likely focuses on developing advanced AI models that combine text and visual inputs, similar to the technology behind ChatGPT.

Why it matters: This move signals Apple's determination to catch up in the AI race and potentially integrate cutting-edge AI features into future iOS releases.

Apple's emphasis on reliability could lead to innovative yet carefully controlled AI experiences.

ChatGPT’s New Memory Feature

OpenAI has introduced a "Memory" feature for ChatGPT Plus users. This allows the AI to remember key points from previous chats, making conversations more contextual and less repetitive. Users can control what the AI remembers and forgets for enhanced personalisation.

Why it matters: This upgrade has the potential to make ChatGPT more useful for tasks that require continuity and understanding of a user's preferences.

The ability to learn over time could signify a step toward an AI-powered "personal assistant."