The Limitations and Opportunities in the AI Space.

Understanding AI, its current state, limitations, and the business opportunities it presents can be complex, especially in a world where we're constantly bombarded with information. We aim to share resources that we find useful, targeting other companies who are either working with AI or looking to integrate AI into their operations. This article lays out the current limitations and opportunities in the field.
Warning: What follows is a summary of key points from talks given by Yann LeCun and Andrew Ng. We suggest setting aside some time to actually watch the videos. They're worth the time investment. We promise.

The Limitations of AI 

In this 57-minute video, Yann LeCun talks about AI and LLM's limitations, and the future of research in the field for the next 10 years. LeCun is super influential in the field due to his pioneering work in deep learning and neural networks in both academia (NYU) and industry (Chief AI scientist at Meta). He won a Turing award in 2018.

1. There's a problem with AI systems

LLMs shine when dealing with text but suck with other tasks. They hallucinate, they aren't super good at math, unless you augment them with tools, which you can, of course, and their intelligence is very limited. Why? Well, because they've been trained with pure text. They lack human-like intelligence. Not all human biological knowledge is in text form. LLMs fall short in reasoning and planning effectively. They also have a limited understanding of the real world.

2. The future of AI: Object-driven AI & Joint-embedding architectures 

LeCun introduces a concept called Objective-driven AI and says it's the future of AI research because it will allow machines to foster a more human-like way of reasoning and planning and will also eliminate the need for extensive fine-tuning of models-.

He also says that, if we want to truly understand the world, it's time to say bye-bye to genAI and hello to joint-embedding architectures.

3. AI should be open source

There's no need for having many models; they are expensive to train. A few open-source ones can do the job, allowing people to develop specialized applications on top of them. Collaboration is key in our AI-driven future internet because at the end of the day, all of our interactions with machines will be through AI agents, and the internet will be like a big knowledge hub. Building something like Wikipedia isn't feasible with proprietary systems. Open-source AI ensures access and progress for everyone.

The Opportunities for AI

In this 37-minute long video, Andrew Ng explains the opportunities of AI in 2023. He's involved with AI Fund, DeepLearning.AI, Coursera, and Stanford, and has significantly advanced AI and democratized its education, reaching over 8 million people globally. His influence is notable, with 1 in a thousand people having taken an AI class from him.

  1. AI is a General-Purpose Technology

AI is like electricity, and there are many use cases and applications yet to be realized.

  1. It will be used in broader industries

The AI community is developing tools that democratize AI customization, shifting from a concentrated focus on tech and consumer software to broader industries.

  1. Collaboration will be key

Launching companies or integrating AI into existing businesses requires a collaborative approach, blending AI expertise with industry-specific knowledge.

  1. Competing in the AI Stack: 


Andrew believes the developer tools layer holds promise with potential mega winners emerging. However, competing in the application layer is better because there are fewer competitors and more opportunities. Especially if there's collaboration.

  1. We must be ethical

    Having AI systems that have our intelligence aka Artificial General Intelligence (AGI) is still far away.  However, we should emphasize the importance of ethical considerations in our current projects, ensuring we use AI for good.

Wrap Up 

AI still needs to improve at some tasks, but it will eventually get there with continued research and development. That's why the job of researchers is crucial, and the emphasis on building open-source solutions is significant. However, even with its current limitations, AI is just like electricity and can be useful in various applications and industries. To make the most out of this, collaboration between AI and subject matter experts is crucial. We are moving forward to explore these opportunities and are looking for partners to join us. If you resonate with this perspective, contact us to discuss and explore possibilities together.