Everyone's betting big on AI Agents

Hi there, folks. This week we talk about AI Agents, Self-discover framework (allowing machines to think) and Gemini Ultra release.

We cover

• AI Stuff you should know:

  • Microsoft, OpenAI, Meta and a bunch of companies betting big on AI Agents

  • Google’s SELF-Discover Framework

  • Gemini Ultra

• Top AI Products launched this week

• Video Guide for Free Professional Avatar Generation

• This week in AI (Quick skim through top AI News)

Last week we saw how AGI is going to be big this year, well, AGI will be enabled via something called Agents, if you don’t know what these are, about time you pay attention.

Let’s dive in.

Stuff you should know

Microsoft, OpenAI, Meta and Open source AI Agents

AI agents are becoming increasingly sophisticated. OpenAI is focusing more on AI agents, Microsoft has published two innovative AI papers, underlining its commitment to advancing AI technology.

These AI agents are smarter than existing AI tools, capable of offering proactive suggestions, completing tasks across different applications, and learning over time by remembering user activities and identifying patterns in behavior. Think of these agents like a piece of code and a manager to manage all these pieces of code.

Numerous open-source autonomous agents are also available, including AutoGPT, BabyAGI, SuperAGI, ShortGPT, ChatDev, AutoGen, and MetaGPT.

AI agents are set to make services, currently deemed too costly for many, widely accessible, significantly impacting healthcare, education, productivity, and the domains of entertainment and shopping.

Business Use Cases

• AI agents can dispense healthcare advice, tutor students, manage shopping, and enhance worker productivity.

• Businesses can develop AI agents that mirror their brand values and enhance customer service experiences.

• AI Agents can handle various routine tasks such as reorganizing and classifying data, composing emails, and managing calendars.

• Open-source autonomous agents like AutoGen can automate task coordination, optimize workflows, and possess advanced multi-agent conversational abilities to adapt to changes in workflows or objectives.

Why it Matters

AI agents represent not merely a technological evolution but a shift in how we engage with technology. They promise to democratize services, making them accessible to a broader audience, and to significantly elevate productivity across numerous sectors. However, like any potent technology, they carry risks, including potential misuse and the dissemination of misinformation. It is therefore critical to persist in discussions on ethical guidelines and protections for AI.

DeepMind’s SELF-DISCOVER Framework

DeepMind, together with the University of Southern California, has developed a groundbreaking framework known as SELF-DISCOVER for large language models (LLMs) such as GPT-4 and PaLM 2. This framework enables LLMs to autonomously identify task-specific reasoning structures, significantly enhancing their problem-solving capabilities for complex reasoning tasks. By moving beyond predefined structures, SELF-DISCOVER allows LLMs to independently develop task-specific reasoning architectures, marking a notable advancement in the development of LLMs.

Stats and Bullets

• SELF-DISCOVER significantly boosts the performance of GPT-4 and PaLM 2 on difficult reasoning benchmarks, including BigBench-Hard, grounded agent reasoning, and MATH, showing improvements of up to 32% over the Chain of Thought (CoT) approach.

• The framework surpasses more computation-heavy methods like CoT-Self-Consistency by over 20%, while requiring substantially less computational power for inference, between 10 to 40 times less.

• The reasoning structures identified by the SELF-DISCOVER framework are universally applicable across various model families, from PaLM 2-L to GPT-4, and from GPT-4 to Llama2, showing similarities to human reasoning patterns.

Why it Matters

The SELF-DISCOVER framework is a significant leap forward in AI-driven problem-solving. It equips LLMs with the ability to self-discover intrinsic reasoning structures, thereby improving their capability to address complex reasoning problems. This advancement holds the promise of driving unprecedented progress in AI problem-solving, with the potential to revolutionize a variety of sectors by boosting the performance and efficiency of AI applications. Nonetheless, the emergence of such powerful technologies underscores the importance of ongoing discussions on ethical guidelines and safeguards for AI.

The Release of Gemini Advanced (Ultra)

Gemini Advanced is the new face of Google's AI, previously known as Bard, and it's making waves with its Ultra 1.0 model. With capabilities like coding, logical reasoning, and creative collaboration, Gemini Advanced is designed to be a powerful tool for a variety of complex tasks. It's available through the Google One AI Premium Plan, which includes additional benefits like 2TB of storage. Costs around same price of $20 as GPT-4.

The image rings a bell? Much like most YouTubers, i conducted a deep study and a lot of what we saw was, well, not true.

here’s a video about good and bad of the model.

Top AI Products from this week

• Gemini - Google's Gemini AI is a cutting-edge, multimodal model adept at processing text, images, audio, video, and code, setting new standards in AI generalization.

• Wondera - Sing One Song to Create Your AI Voice

• Data Analyst AI - Receive AI-generated eCommerce performance reports directly in your inbox for quick and easy analysis. 

• Retell AI - API that enables developers to build human-like voice agents 

• Heyday - turns your conversations, documents, and articles into quotes, shareable content, and a queryable database.

• Rizzle AI - Rizzle helps you create copyright-compliant videos from text, podcast audio/video, or long-form video.  

• Capitol AI - Capitol is the most powerful way to create content with AI.

• Jan - Jan runs 100% offline on your computer, utilizes open-source AI models, prioritizes privacy, and is highly customizable.

• Open Love - AI girlfriend - In Open Love improve dating skills with the help of an AI girlfriend

•  3DAiLY Beta - 3DAiLY, powered by Generative AI, enhances asset creation and monetization for game studios and 3D artists with tools like a 3D Editor, SDK, CMS, Store, and on-demand asset creation.

• GPTs App - The Best Third-Party GPT Store, powered by GPT-4 Turbo and Pinecone. Discover custom ChatGPT apps to boost productivity and innovation in your business.

Handpicked Video

Generate Professional Headshots and AI Avatars, full Guide:

This week in AI

• Google Chrome's AI Features: New AI enhancements including tab organization, personalized themes, and an AI writing assistant in the U.S.

• AI-Generated Image Concerns: Study reveals issues with fraudulent AI images, prompting OpenAI and Microsoft to bolster safety measures.

• AI Trends and Governance: Emphasis on adapting to generative AI trends and upcoming EU regulations for AI oversight.

• Google's 'Gemini' AI System: Introduction of 'Gemini' for AI-powered subscription services.

• AI in Journalism: Semafor's AI-driven news feed 'Signals' enhances story editing and perspectives.

• Apple's AI in Image Editing: Launch of MGIE, an advanced AI model for text-driven image editing.

• Deciphering Ancient Texts with AI: AI helps uncover secrets from a 2,000-year-old scroll, awarded $700,000.

• Rise in AI Scams: FTC probes into AI voice scams and cybersecurity threats among tech giants.

• Deepfake Scam Alert: A firm loses $34 million to a deepfake scam, underlining AI security concerns.

• Meta's AI Labeling: Introduction of technology to label AI-generated images for authenticity.

• AI Job Market Growth: Steady increase in AI-related job listings, with the UK leading slightly over the U.S.

That’s all folks, ciao