AI Weekly Digest #20: Top News Shaping AI in 2023 - Part 3, Agents, AI Ops and Copyright!

Hello, tech enthusiasts! This is Wassim Jouini and Welcome to my AI newsletter, where I bring you the latest advancements in Artificial Intelligence without the unnecessary hype. You can also find me on Medium covering AI related topics.

News Shaping AI in 2023 -
Part 3 , AI Agents !

What to remember about the past 6 months?

***

This is the 3rd part of a retrospective series

Part 1: on the evolution of Large Language Models, the rise of open-source AI, and other key developments in the AI industry in 2023.
You can find it here.

Part 2, on images & videos generation, voice cloning and multimodal models! You can find it here.

***

#1- AI Agents: a versatile AI

What’s an AI Agent?

The landscape of AI has evolved significantly with the advent of autonomous AI agents, a breakthrough that emerged with the release of GPT-4 in March 2023. These AI agents are capable of independently defining, planning, and executing tasks towards a set goal, and can connect with external tools when necessary. This advancement overcomes the previous limitation of AI models, which required close supervision, by enabling autonomous task execution.

A few notable autonomous AI agents include AgentGPT, Auto-GPT, Baby AGI, and Jarvis (or HuggingGPT). These agents can handle complex tasks such as generating a list of different music genres and artists or planning a romantic dinner.

They achieve this by breaking down the main task into sub-tasks, executing them, and utilizing the results for subsequent tasks. Their flexibility allows them to make use of multiple foundation models and to connect to external tools like search engines or calculation tools. These agents, which are still relatively new, represent a big step forward in the AI field, capable of "slow thinking" and tackling multi-step problems.

AI Agents’ Adoption?

AI Agents proved to be very powerful at managing complex tasks. For instance, it was able to play very efficiently complex games such as Minecraft, or simulate a SimCity like universe.

Yet, today, the development of such AI agents requires substantial computational resources, primarily driven by models like GPT-4. This creates a challenge, making the process expensive and time-consuming, thus hindering widespread adoption. However, the future paints a much brighter picture. With the advent of more affordable computing power and the continued advancement of open-source AI models, we can foresee a new wave of AI agent development.

Future of AI Agents: universal colleagues?

Envision a world where specialized AI agents become integral members of our teams - a coder in the software development team, a content creator in the marketing department, or a data analyst in the research group.

Each agent, functioning like a colleague, could be assigned tasks via popular project management tools such as Jira. These AI agents, connected to a team's agile process, would interpret and execute tasks, contributing directly to the team's productivity.

The AI coding agent, for instance, could autonomously write and optimize code, review code commits, or even fix bugs. The AI content creator could draft blog posts, design graphics, or curate engaging social media content. The AI data analyst could process complex data, provide actionable insights, and generate detailed reports. All while interfacing seamlessly with productivity tools, databases, and software environments, making them true teammates in the workspace. This vision not only augments our capabilities but also opens the door to unprecedented levels of efficiency and innovation.

#2- AI Ops

Creating a proof of concept (POC) for an impressive demo is relatively straightforward (just look at the thousands of new AI POCs emerging everyday). However, the deployment of production-grade features that are stable, reliable, and maintainable presents a significantly tougher challenge. This difficulty explains the discrepancy between the multitude of AI announcements and the fewer actual AI releases.

As a matter fact, dealing with

• the models’ limitations (hallucination, context size, security, etc.)

• as well as their potential drift over time: (with a change in behavior and performance). Read this blog post if you are relying on a LLMs for your product: OpenAI Models’ Drift, Fast & Stupid? — New Insights and Consequences.

Makes the design of production grade AI products a real challenge! A challenge for which most engineers don’t have the tools yet!

Personal Note: I do see on a weekly basis companies relying on obsolete models or fine-tuning models hoping to overcome certain limitations, when often simple prompt engineering would have led to better results at a drastically smaller cost.

Conclusion: Training your team and building the right experimental pipelines will play a major role in the deployment and evolution of your AI integrations!

#3- Copyright: Who Owns AI Generated Content?

This is still an open question that needs to be addressed urgently.

There has been a case in the US though, where US Copyright Office refused to protect AI generated images!

Why is it important? the document announces two decisions:

• Images generated by AI, and not significantly altered by the author (via Photoshop for instance) are not protected by copyright in the US.

• Comic book’s text and image arrangement are protected by US copyright since they involved a human effort.

“We conclude that [...] the images in the Work that were generated by the Midjourney technology are not the product of human authorship.” -United States Copyright Office (Source)

Conclusion: It is uncertain how this decision will impact artists in the US, and it remains to be seen whether similar decisions will be made in other parts of the world. The issue of copyright for AI-generated content is still unfolding! We’ll keep you updated ;)!

Conclusion & Perspective!

In wrapping up, we've delved into AI's exciting progress in 2023: autonomous AI agents promising increased efficiency, AI Ops tackling the challenges of creating production-grade AI, and copyright issues concerning AI-generated content.

The future of AI presents exciting possibilities and challenging questions that we must navigate together. So, stay tuned for more insights, and keep exploring the dynamic world of AI.

Until next time, this is Wassim Jouini, signing off. See you in the next edition!

Have a great Sunday and may AI always be on your side!