AI News: We're One Step Closer To AGI This Week!

Introduction

Introduction

Here’s some AI news you might have missed this week! Starting off, OpenAI has mapped out their five levels towards the progress of AGI. Let’s break them down:

OpenAI's AGI Levels

1. Level 1: ChatBots and AI
   • Current state (ChatGPT, Claude, LLaMA 3).
2. Level 2: Reasoners
   • AI with human-level problem solving. OpenAI claims they are very close to achieving this.
3. Level 3: Agents
   • Systems that can autonomously perform actions (booking flights, responding to emails).
4. Level 4: Innovators
   • AI that aids in creating novel ideas and inventions.
5. Level 5: Organizations
   • AI systems capable of running entire organizations.

OpenAI's New Reasoning Technology: Project Strawberry

OpenAI has been working on a new reasoning technology codenamed “Strawberry.” This emerged from a leaked document seen by Reuters. Strawberry aims to autonomously navigate the internet for deep research and perform long-horizon tasks.

Whistleblower Allegations against OpenAI

OpenAI faces allegations from whistleblowers claiming they prevent employees from talking to government regulators and may illegally remove rewards for whistle-blowing. OpenAI refutes these claims, stating they protect employees’ rights to make protected disclosures.

OpenAI’s DALL-E Model Update

Speculation exists that the DALL-E image model may have received an update, as seen from improved text clarity in generated images. If you don't have a ChatGPT Plus account, you can use DALL-E 3 for free on Bing’s website.

New Demos from Soro

New black-and-white video demos from Soro look impressive, though Runway Gen-3 and Luma’s Dream Machine currently hold enthusiasts over.

HubSpot’s Free Resources for ChatGPT

HubSpot offers a free bundle called "Five Essential Resources for Using ChatGPT at Work." This bundle includes flowcharts, templates, and checklists to help enhance your work with AI.

Andrej Karpathy’s New Venture: Eureka Labs

Former OpenAI member Andrej Karpathy announced a new AI+education company called Eureka Labs. The company aims to build AI-native schools with AI teaching assistants to help scale learning.

Anthropics Claude Goes Android

Anthropic’s Claude AI is now available on Android, having been on iOS for a couple of months. This offers competition to the ChatGPT app, which is noted for its conversational voice abilities.

Google’s Gemini and Google Vids

Google’s Gemini can answer general questions even when your Android phone is locked. Google has also announced an AI-powered video creation app called Google Vids, integrated with the Workspace suite.

Apple, NVIDIA, and Anthropic Data Usage

Reports claim Apple, NVIDIA, and Anthropic have used thousands of YouTube videos to train their AI. Apple clarified that although they used this data for research, it is not part of their primary models.

Microsoft Designer Updates

Microsoft Designer, similar to Canva, is integrated into various Microsoft apps. It now offers a mobile app and restyling features.

AI Models and Forensics

A new AI system has achieved 96% accuracy in determining sex from dental X-rays, potentially useful for forensic purposes.

New Models and Collaborations

• OpenAI’s GPT-4 Turbo Mini: Released to replace GPT-3.5.
• NVIDIA and Mistral's Model: New efficient model for local deployment, useful in privacy-sensitive and limited-internet environments.

Meta Multimodal Models and EU Limitations

Meta’s multimodal models will not be available in the EU due to regulatory uncertainties but will be available in the UK.

Tencent's 3D Character from Selfies

Tencent’s new app to turn selfies into 3D printable characters is generating buzz, though details remain scarce.

Google's Olympic AI Sponsorship

Google is the official AI sponsor for Team USA and will feature AI products prominently during the Summer Olympics.

Keywords

• OpenAI
• AGI
• Strawberry
• Whistleblower
• DALL-E
• Soro
• HubSpot
• Eureka Labs
• Claude AI
• Google Gemini
• Google Vids
• AI Training Data
• Microsoft Designer
• Forensic AI
• GPT-4 Turbo Mini
• Nvidia Mistral
• Meta EU Ban
• Tencent 3D
• Olympic AI Sponsorship

FAQ

What are the five levels of AGI according to OpenAI?

The levels are ChatBots/AI (Level 1), Reasoners (Level 2), Agents (Level 3), Innovators (Level 4), and Organizations (Level 5).

What is OpenAI’s Project Strawberry?

Project Strawberry is a new reasoning technology aimed at autonomously navigating the internet to perform deep research and long-horizon tasks.

What allegations are OpenAI facing regarding employee whistleblowers?

Whistleblowers claim OpenAI prevents employees from talking to regulators and may remove rewards for whistle-blowing. OpenAI denies these allegations.

How can I use DALL-E 3 without a ChatGPT Plus account?

You can use DALL-E 3 for free on Bing’s website at bing.com/create.

What is Eureka Labs by Andrej Karpathy?

Eureka Labs aims to create AI-native schools with AI teaching assistants to help scale learning and assist with educational content.

What new AI-powered updates have Google announced?

Google introduced Gemini for answering questions with a locked phone and Google Vids, an AI-powered video creation app integrated with Workspace.

What is Microsoft Designer and its new features?

Microsoft Designer is similar to Canva and now includes a restyling feature, and a mobile app for creating and editing images on the go.

What is NVIDIA and Mistral’s new joint project?

They have released an efficient model called Mistral Nemo, suitable for on-device use, especially in privacy-sensitive or limited-internet environments.

Why are Meta’s multimodal models not available in the EU?

Meta cited regulatory uncertainties related to GDPR compliance as the reason for not offering their multimodal models in the EU.