When GPTs communicate with each other 🔁

Today’s top AI Highlights:

1. OpenAI is Rolling Out “GPT Mentions”

2. Microsoft’s Report on How LLMs have Impacted Work in 2023

3. Meta-Prompting: Enhancing Language Models with Task-Agnostic Scaffolding

4. AI Copilot, Everywhere in your Mac OS

& so much more!

Read time: 3 mins

Latest Developments 🌍

Bring Another GPT Directly in your ChatGPT 📲

Ever felt you want to use a GPT in the middle of your chat with ChatGPT? OpenAI quietly released this new feature called “GPT Mention” Just type @ followed by the name of the GPT to bring it directly into your conversation.

The New Future of Work 2023 💼

In their latest endeavor to understand and navigate the evolving landscape of AI in the workplace, Microsoft has released the comprehensive "New Future of Work Report 2023". This report, a synthesis of research conducted by Microsoft and its global partners, delves into the multifaceted impact of LLMs like GPT-4 on various work-related aspects. It draws from a wide array of studies, experiments, and real-world observations, offering a nuanced view of how AI is shaping current and future work practices.

Key Highlights:

1. Productivity and Task Efficiency: The report indicates a substantial improvement in productivity for information workers using generative AI-based tools. In a study involving BCG consultants, it was observed that AI tools led to more than a 40% increase in the quality of outputs on a simulated consulting project. However, it's worth noting that increased speed sometimes came at the cost of moderately lower correctness, with BCG consultants making more errors when the LLM made mistakes.

2. Impact on Different Skill Levels: The report sheds light on the varying impact of AI across skill levels. For instance, a study involving GitHub Copilot suggested that developers with less experience benefitted more from AI assistance. In contrast, in a task involving BCG employees, those in the lower half of the skill spectrum showed a 43% improvement in performance when using AI, compared to a 17% improvement for the more skilled top half.

3. Critical Thinking and Creativity: AI's role as a 'critical thinking provocateur' is highlighted, emphasizing its ability to stimulate deeper analytical thinking and creativity in complex tasks. This is particularly evident in fields like software engineering and education, where AI tools are not only aiding in routine tasks but also encouraging a more thoughtful and innovative approach to problem-solving and learning.

4. Human-AI Collaboration and Co-Audit: The report emphasizes the evolving dynamics of human-AI collaboration, highlighting the importance of co-audit tools. These tools, such as AI-generated spreadsheet computations, assist users in understanding how AI translates their instructions and helps in inspecting the outcomes, thereby enhancing decision-making accuracy.

5. Workplace Integration and Sociotechnical Considerations: The integration of AI into the workplace is not just a technological shift but also a sociotechnical process. The report cautions against over-reliance on AI and suggests that effective integration requires consideration of organizational culture, user perceptions, and workflow compatibility. This aspect is underlined by studies on tools like Microsoft’s M365 Copilot, which showed significant perceived time savings by enterprise users, yet also necessitated an understanding of its fit within existing workflows.

Smarter LMs with Task Agnostic Prompting Technique 😎

Researchers at OpenAI and Stanford propose a prompting technique to enhance the capabilities of LLMs like GPT-4. Meta prompting transforms a single LM into a multi-functional entity, capable of breaking down complex tasks into smaller, manageable subtasks. These subtasks are then assigned to different instances of the same LM, each functioning as an "expert" in a specific area. The LM, acting as a conductor, manages these experts, integrating their outputs to produce a comprehensive and accurate final response. This process allows a single LM to function simultaneously as a holistic orchestrator and a diverse panel of specialists.

Key Highlights:

1. Meta-prompting has demonstrated its efficacy in a range of tasks, notably improving performance in challenges such as the Game of 24, Checkmate-in-One, and Python Programming Puzzles. By breaking down complex tasks into smaller, manageable subtasks, the technique enables a single language model to efficiently handle multiple aspects of a problem, leading to a more accurate and refined output.

2. One of the standout features of meta-prompting is its task-agnostic nature. Unlike traditional methods that require detailed, task-specific instructions, this new approach utilizes a universal set of high-level instructions. This simplifies the user experience significantly, as it removes the need to provide specific guidance for every distinct task, making the technology more accessible and user-friendly.

3. The research highlights the seamless integration of a Python interpreter within the meta-prompting framework. This addition not only broadens the applicability of the technique but also allows for more dynamic responses, further extending its potential to effectively address a wide array of complex tasks and queries.

Tools of the Trade ⚒️

1. VectorShift: An AI automation platform that allows teams to use AI through a no-code interface or Python SDK. It enables the creation of automated workflows, including document generation, knowledge base searches, and deploying chatbots and assistants.

2. Scade: Create instant AI apps in minutes with nocode AI platform for B2B & B2B2C businesses. Scade automates business processes and makes creating full-fledged products as easy as creating a PowerPoint presentation.

3. Maika AI: Maika AI helps you become an expert in content marketing. Write fast, multi-platform content in just a few clicks

4. Omnipilot: An AI that can use your computer with full context on what's on your screen and make decisions.

Hot Takes 🔥

1. The skills needed for developing new ML techniques have little overlap with the skills needed for applying ML effectively. A bit like how chip design has little overlap with software engineering. ~ François Chollet

2. I think assuming you’ll get gpt-4 or even gpt-3.5 level quality in a sub 7B LLM is a mistake. Problem is it might do well on benchmarks but then you start using it and realize there’s all these edge cases it doesn’t handle well - does not generalize as well as the larger models ~ anton

Meme of the Day 🤡

That’s all for today!

See you tomorrow with more such AI-filled content. Don’t forget to subscribe and give your feedback below 👇

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!!

PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!