• The Bright Journey with AI
  • Posts
  • OpenAI Project "Strawberry" At the Cutting Edge | DeepMind Propose Tiny Expert for Scale | Five-Tiers of AI | System 2 Thinking from Meta

OpenAI Project "Strawberry" At the Cutting Edge | DeepMind Propose Tiny Expert for Scale | Five-Tiers of AI | System 2 Thinking from Meta

The Bright Journey with AI

📆 July 16th 2024📆 

The needle keeps on moving - reasoning is making way for advanced reasoning and, who-knows-what next 🤓 

  • OpenAI hint at the next big thing with project Strawberry

  • Google DeepMind introduce a new concept to allow scaling of AI experts efficiently

  • OpenAI propose there are five tiers to AI which also serves as a roadmap of sorts

  • Meta announce better reasoning capabilities, bring system 2 thinking to LLMs. Direct reference to Thinking Fast & Slow, if you haven’t read it.

  • As always, too much to put in headlines. Check out the Other News Rollup section for a quickfire review of the top stories in AI.

Dive in for all the details!

🆘 Quick Tips 🆘 

Did you know that recent AI advances have made image recognition really capable? This tip has many use cases and can help save you time manually entering information and avoid mistakes.

  • Step 1 - Take a picture of a receipt, bill or hand written note you want to extract information from

  • Step 2 - Go to https://chatgpt.com/

  • Step 3 - Upload your image and enter the following prompt

    • For this example I uploaded two images; one with handwritten notes and another with a shopping receipt & asked for a TODO list

Carefully read the following images and extract the content. You will present the output in a list of todos, using the receipt to create a shopping list

And there you have it - easily extracted content that you can now use across any digital product to do more.

Does this spark any ideas? Contact Me for a free discovery call. Let’s unlock AI’s potential for your business today!

💫Help Support my Work💫

I really enjoy researching and writing about AI & appreciate your support. If you enjoy this content please use any of the options below to help

📰 News 📰 

OpenAI Nears Breakthrough in Reasoning AI with New Framework

OpenAI is reportedly close to achieving a significant breakthrough in artificial intelligence, focusing on advanced reasoning capabilities. This development is part of a project known internally as "Strawberry," which aims to enhance AI's ability to perform complex reasoning tasks more efficiently and accurately. This new AI model could dramatically improve the way machines understand and solve problems, bringing them closer to human-like thinking​.

Despite the promising advancements, the project has raised concerns among some OpenAI staff about the potential risks associated with powerful AI technologies. These concerns were highlighted in a letter to the board, which reportedly contributed to the temporary dismissal of OpenAI's CEO, Sam Altman. The incident underscores the ongoing debate about AI safety and the need for stringent oversight as AI systems become more capable​.

Read More At: Tom’s Guide

DeepMind's PEER Leverages Tiny Experts for Scalable Language Models

DeepMind has introduced PEER (Polymerized Expert Encoding Representation), a novel approach to scaling language models using millions of specialized micro-models, or "tiny experts." This innovative method enhances the efficiency and performance of language models by distributing the computational load across a multitude of smaller, specialized models, each expert in different types of data or tasks. This approach allows for better handling of diverse linguistic challenges, from understanding nuanced context to generating more precise and relevant responses.

The distributed nature of the PEER system also means it can scale more efficiently than traditional monolithic models, potentially reducing computational costs and improving speed.

Read More At: VentureBeat

OpenAI Unveils Five-Tier AI Progress Framework

OpenAI recently introduced a five-tier system to gauge its advancement towards Artificial General Intelligence (AGI), shared during an all-hands meeting with employees. This framework is designed to provide a clear understanding of AI progression, although it describes hypothetical technologies that do not yet exist. This system appears to be aimed at attracting investment by highlighting aspirational goals.

The Five Stages of Artificial Intelligence:

  1. Level 1: Chatbots - Current AI, including models like GPT-4, capable of conversational interactions.

  2. Level 2: Reasoners - AI that can solve problems at a human level, comparable to someone with a doctorate degree but without external tools.

  3. Level 3: Agents - Autonomous systems capable of working on tasks for extended periods.

  4. Level 4: Innovators - AI that can assist in creating new inventions.

  5. Level 5: Organizations - AI that can manage entire organizations.

This framework, still in development, aims to communicate AI milestones and gather feedback from stakeholders.

Read More At: Ars Technica

Meta Integrates System 2 Thinking into LLMs for Enhanced Reasoning

Meta's AI researchers have successfully integrated System 2 thinking into large language models (LLMs), significantly improving their performance on complex reasoning tasks. This new approach, developed by Meta's Fundamental AI Research (FAIR) team, enables LLMs to simulate more sophisticated, human-like thought processes.

System 2 thinking, known for its slow and deliberate reasoning, contrasts with the more intuitive and fast System 1 thinking. By incorporating elements of System 2, such as intermediate thought processes and strategic problem-solving, the LLMs can achieve higher accuracy and efficiency with fewer computational resources. This innovation is expected to advance the capabilities of AI in various applications, from natural language processing to decision-making systems​​.

Read More At: VentureBeat

🔨 AI Powered Tools 🔨 

  • Framer - Uses AI to enhance design workflows by automatically generating text, images, and layout suggestions. It leverages AI for content creation, improving design efficiency and creativity.

  • Descript - Offers AI-powered video and podcast editing, enabling users to edit media by editing text transcripts. Features include AI-generated transcripts, voice cloning, and text-based video editing, simplifying complex editing tasks.

  • Numerous - Integrates AI into Google Sheets and Excel, offering functionalities like content summarization, classification, and generation of marketing materials. It simplifies data management and content creation within spreadsheets using the =AI function.

🗞️ Other News Rollup 🗞️ 

Nvidia's Siggraph Innovations - Nvidia showcases cutting-edge AI advancements at Siggraph 2024, including rendering, simulation, and generative AI innovations.

AI Model Cost Decline - OpenAI expects AI model costs to decrease with the introduction of the affordable GPT-4o model, enabling new use cases.

AI Success Story - Instabase wins top award for AI innovation, automating document workflows and unlocking unstructured data for enterprises.

AI Datacenter Trends - Remote datacenters with cheap energy are ideal for AI training. Challenges include weather and operational risks.

Graphcore Acquisition by SoftBank - SoftBank buys Graphcore for $600 million, enhancing AI technology with innovative accelerators.

AI Revolutionizes Robotics - Google's Gemini AI model boosts robot navigation and interaction, opening doors for broader AI integration in robotics.

AI security funding boost - Kindo raises $20.6M for AI security, acquires WhiteRabbitNeo, enhancing enterprise AI safety and control.

AI and Employee Replacement - CEO advocates for augmenting, not replacing, employees with AI to improve satisfaction and performance, emphasizing leadership's irreplaceable role.

AI-Powered Ad Revolution - AdCreative.ai launches AdLLM Spark, predicting ad performance with over 90% accuracy, outshining competitors.

AI Act Compliance Costs - EU AI Act raises costs for Chinese firms, non-compliance leads to significant fines. Innovation impact questioned.

🎶 Prompts 🎶 

Prompts used to generate some of this issues images. Unless otherwise stated all images are using Dall-E and ChatGPT Plus.

Generate a wide banner style image of an AI thinking deeply. The scene should be set in a home study with evening light streaming in the window, creating a warm atmosphere. The perspective should be from the side looking towards the window
Generate a wide banner style image that shows thousands of mini AIs sitting in an amphitheatre. Each should be shouting their ideas which are being merged into one at the centre
Generate a wide banner style image of a strawberry. It should be bright red and glistening with moisture, signifying that it is ripe and delicious. The image should have one strawberry as the focal point

Thank You for Subscribing

Enjoying what you’re reading? Help me get better so I can continue to provide you with the most relevant content.

Reply

or to participate.