Former Tesla director and OpenAI expert Karpathy: I was distracted by autonomous driving, AI agents are the future!

Recently, Andrej Karpathy, co-founder of OpenAI, former director of TeslaAI, and now returning to OpenAI, shared his views on AI agents at a developer event.

Seven years ago, the time to study AI agents was not yet mature

He first talked about his early work at OpenAI (in 2016 ), the industry trend at that time was to study how to use reinforcement learning methods to improve AI agents.

Many projects are making AI players based on games like Atari.

Picture

What he wanted to do at that time was a product with a wider range of applications.

However, due to the technical limitations at the time, the results were not good, so he and OpenAI changed direction and began to develop larger language models.

Of course, I was distracted by the automatic driving during this period.

But now, five years later, AI agents have become a very promising direction again.

Because there are now new technical means to study AI agents, the situation is completely different from 2016.

The simplest example is that no one is using reinforcement learning methods to study AI agents anymore like they did in 2016.

The current research methods and directions were unimaginable back then.

AI agents represent a crazy future, although it may be a bit far away

Because in the future, if AGI can appear, it will Give full play to the capabilities of AI agents.

The AI ​​agent in the future may not be a single individual, but there will be many AI agent organizations or even AI agent civilizations.

Picture

This could be a very exciting, even crazy, future.

But at the same time, everyone must stay awake and calm.

Because some technology trends may be easy to conceive and envision, but it is difficult to implement products.

Many technologies fall into this category, such as autonomous driving.

Technological visions are easy to envision, and demonstrations of cars driving around the block are easy to create, but it can take 10 years to create a product.

Similarly, VR is also the same situation.

Picture

AI agents may also belong to this category of technology. The application scenarios are easy to imagine and the prospects are exciting. But it requires long-term technological development and accumulation.

AI agents need to draw inspiration from neuroscience

Just like the early development of deep learning, the development of AI agents may Get inspired by neuroscience.

It is interesting to think about the relationship between AI agents and neuroscience.

Especially now that many people regard large language models as part of their AI agent solutions.

But how to build a complete digital entity that possesses all human cognitive abilities?

Obviously, we all agree that there needs to be some underlying system for planning, thinking and reflecting on what we are doing.

This may be where neuroscience can help.

For example, the hippocampus is a very important part of the brain.

But what in the AI ​​agent plays the role of the hippocampus to store memories, implement tagging and retrieval, etc.?

We have a general understanding of how to build the visual and auditory cortex, but there is still a lot we don’t know about what exactly it means in an AI agent.

For example, what does the thalamus, the seat of the subconscious, correspond to in AI Agents?

These are very interesting questions.

I specially brought a book on neuroscience, "Brain and Behavior" by David Eagleman. I found this book very interesting and enlightening.

Just as early AI research did when designing neurons, drawing interesting inspiration from neuroscience may be a direction we should try again.

Everyone here is at the forefront of the industry

You may not necessarily know it, but the AI ​​agents built by everyone present today are already at the forefront of the industry. The cutting edge of AI agent capabilities.

I don’t think any of the institutions currently working on large-scale language models, such as OpenAI, are at the forefront of this field.

The forefront is all of you here.

For example, OpenAI is very good at training Transformer language models.

If a paper proposes a different training method, then everyone in our OpenAI internal Slack group will discuss:

"I tried this method two and a half years ago, but it didn't work."

We are very clear about the ins and outs of the method of training the model.

But when the paper on AI agents comes out, all of us will be very interested and think it is amazing.

Because our team has spent the last five years elsewhere.

Picture

We don’t know more than you in this field, and we stand on the same level of competition as everyone else .

This is why I think everyone here is at the forefront of AI agents. This is very important for the development of AI agents.

The above is the detailed content of Former Tesla director and OpenAI expert Karpathy: I was distracted by autonomous driving, AI agents are the future!. For more information, please follow other related articles on the PHP Chinese website!