Google DeepMind's framework for evaluating AGI's progress
PLUS: Amazon Olympus = ChatGPT's replacement?
Good morning, human brains. Welcome back to your daily munch of AI news.
Here’s what’s on the menu today:
Is AI better than you? Not yet, but here’s how you’ll know 👩💻 🤖
Google DeepMind unveiled its new framework to label AGI’s progress.
Is coding easy? No, but in December, it might be a little easier 🦾 🤖
GitHub announced improvements to GitHub Copilot Chat.
Amazon Olympus = ChatGPT’s replacement? 🦹♀️ 👹
Amazon is working on a competitor to OpenAI / Microsoft’s AI offerings.
Is ChatGPT considered AGI, yet? 👩💻 🤖
In October, we reported on Google DeepMind’s proposed AI Safety framework. It particularly addresses the social and ethical risks of AI systems.
Last Saturday, Google DeepMind introduced its “Levels of AGI” framework. It’s a way to categorize and measure AGI progress.
What’s AGI? Asking for a friend…
We haven’t achieved it yet, it stands for Artificial General Intelligence.
It’s AI’s ability to understand, learn, and apply knowledge at a level comparable to a human.
This framework categorizes and measures AGI similarly to how we evaluate stages of autonomous driving.
Why did they spend time and money on this?
Google DeepMind claims that AI’s recent advancements have shifted AGI from a philosophical concept to a practical, urgent field of study.
It criticizes the existing definitions of AGI and suggests the need for a more robust, universally applicable definition.
How does it work?
The framework assesses current AI systems based on depth and breadth.
Depth refers to how well it performs.
Breadth refers to how many different tasks it can do.
The framework outlines six principles for defining AGI.
Here they are:
Where are we now?
The paper states that current LLMs, like ChatGPT, demonstrate some AGI capabilities, but require further development to reach full AGI.
DeepMind classifies ChatGPT at level 1 out of 5.
What are the main takeaways?
DeepMind believes the path to AGI should be seen as a journey with multiple stages, each with specific benchmarks and considerations, rather than a single endpoint.
It also claims AGI should be judged by potential ability rather than current application, not requiring a system to be deployed in the real world for it to be considered AGI.
SPONSORED BY WONDERCHAT
Build a Custom ChatGPT for your website in 5 minutes ⏱️
Create ChatGPT-powered chatbots to automate up to 70% of customer support for your website.
Spend 5 minutes to create a bot that saves you 5000 minutes of manual work.
Wonderchat empowers you to create:
✅ GDPR-compliant bots in 5 minutes trained on site-links and PDF files
✅ Multilingual chatbots that provide 24/7 customer support in over 80+ languages.
✅ AI chatbots that hand over to human support & integrate with 5000+ apps via Zapier.
Claim Your FREE Trial instantly (No credit card required).
BUZZWORD OF THE DAY
Leaner, meaner GitHub Copilot Chat 🦾 🤖
In September, we covered GitHub’s expanded beta access for Copilot Chat. It extended beta access from business clients to all Copilot users.
On Wednesday, GitHub announced improvements to GitHub Copilot Chat. The upgrades include slash commands, context awareness, and more.
It also announced it will be out of beta and generally available in December.
Remind me again, what is Copilot Chat?
It’s an AI assistant that can explain complex coding concepts, provide suggestions based on the user's current work, and help detect and fix security issues.
Developers can use Copilot Chat to discuss specific lines of code within their code editor, streamlining the development process.
What goodies is it getting?
Here is an easy-to-scan list to get you excited:
Increased code awareness: it will use your code as context, suggest code based on your open files/windows, and more.
Inline chat: you’ll be able to chat about specific lines of code directly in your editor.
Slash commands: you’ll be able to use commands like /fix and /tests to improve code, generate tests, and more.
Smart actions: you’ll be able to click where you need help and AI will review the code, offer suggestions, and more.
How much will it cost?
GitHub Copilot is included in the existing GitHub Copilot subscription without additional costs.
It is free for verified teachers, students, and open-source project maintainers.
When can I use it?
GitHub will release Copilot Chat with the new features in December.
A LITTLE SOMETHING EXTRA
Amazon’s new ChatGPT competitor 🦹♀️ 👹
Last week, we covered Amazon’s new service, EC2. It lets you rent out NVIDIA’s GPUs for short-term AI projects.
Now, Amazon is developing an AI model called Olympus. It’s a conversational AI that aims to directly compete with OpenAI and Microsoft.
How is it different from every other AI model?
Olympus integrates with Amazon’s online store, AWS, and Alexa.
It’s reportedly improved Alexa AI-powered generative capabilities for more intuitive user interactions.
How is Amazon making this thing?
It is reallocating funds, investing millions in AI technology, and reducing spending on its retail operations.
Its “Nile” project is underway to enhance Amazon's retail search with AI, making it more conversational and personalized.
How does it measure up against ChatGPT?
Olympus is reported to have 2 trillion parameters. GPT-4 has 1 trillion parameters.
That’s all we know as of now.
When can I use it?
No clue. The launch date has not been announced yet.
MEMES FOR DESSERT
YOUR DAILY MUNCH
Papers.day — an AI-powered search for scientific papers.
Vidiofy — quickly creates reels directly from articles.
Ozone — an AI-powered video editor to create professional short-form videos
AllGPTs — a up-to-date collection of GPTs.
Thinking of starting an AI startup? Read this first, how to create AI startups that can stay relevant.
Use case of OpenAI’s vision API. A GitHub repo on how to run inference of webcam feeds, videos, images, and more.
An interview with Microsoft’s CEO. How Microsoft has evolved since the ’90s, AI, AI safety, and more.
Scale AI unveiled SEAL, its Safety, Evaluations, and Analysis Lab. The purpose is to establish safety benchmarks and evaluation products for deploying LLMs.
Ghost Autonomy raises $5 million. The OpenAI Startup Fund invested in the startup to bring multi-modal LLMS to autonomous driving.
Microsoft updated its startup program. It includes a free Azure AI infrastructure option for high-end GPU clusters.
MFTCoder — a code-generating AI that trains on several tasks simultaneously to improve rapidly and efficiently.
VR-NeRF — it generates realistic VR environments that are walkable using a camera with HDR (high-dynamic range) capture and neural radiance field.
EmerNeRF — a self-learning AI model that enhances its understanding of raw inputs by segmenting motion and stasis.
If you like Bot Eat Brain then you’ll probably like Yaro on AI too:
TWEET OF THE DAY
xAI’s new AI model, Grok, summarizes the latest news articles.
Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.
Until next time 🤖😋🧠
What'd you think of today's newsletter?