Skip to main content

Choose your Champion! Task-Specific vs. General Models

Should AI models be like Swiss Army knives, versatile and handy in a variety of scenarios? Or do we prefer them as precision tools, finely tuned for specific tasks? In the world of artificial intelligence, and natural language processing specifically, this is an ongoing debate. The question boils down to whether models trained for specific tasks are more effective at these tasks than general models. Task-specific models: specialization and customization In my last blog post , we looked at the rise of personalized LLMs, customized for specific users. Personalized LLMs can be seen as an extreme form of task-specific model. Fans of task-specific models stress that these kinds of models are better suited for tasks involving confidential or proprietary data. This is obviously true. But some people also believe that specialized models necessarily perform better in their specific domains. It may sound logical, but the ans...
Post a Comment

OpenAI's ChatGPT plugins: To infinity and beyond

Stars spilling out of glass mason jar

“The thing’s hollow – it goes on forever – and, oh my God, it’s full of stars!”
- David Bowman, 2001: A Space Odyssey (novel)

Last week, OpenAI co-founder Greg Brockman demoed some of the largely unreleased ChatGPT plugins at TED2023. At the same time, he provided more clarity on these plugins and their intended use-cases.

Yesterday marked exactly one month ago that OpenAI tantalisingly announced that they will be introducing support for plugins in ChatGPT. We knew that this would enhance ChatGPT's capabilities beyond its built-in functionalities. But the specifics were a little bit unclear until now.

Essentially, the various plugins will enable the language model to access real-time information from the web and other sources such as databases. Third-party services will allow it to perform actions such as booking a flight or ordering food on behalf of the user. And we will be able to access all of this functionality seamlessly through ChatGPT.

Unfortunately, the plugins are currently still in a limited alpha release and as such currently only available to a small number of developers and ChatGPT Plus users. But it is now clear that OpenAI envisions a future where these plugins will automate various decision-making processes and errands. From Brockman’s talk, we can gather that at least the DALL-E plugin will be made available to all (not just paying) users soon.

And it is not inconceivable that eventually such chatbot-plugin combinations will completely transform how we access and use the internet.

A little robot made from wall plugs

ChatGPT Plugins for Internet Access and Retrieval

By far the most significant plugin is the one allowing internet access. With this plugin, created by OpenAI themselves, we will finally be able to access real-time information through ChatGPT. Think about the implications here: Users will no longer need separate apps for browsing. Talk about industry disruption!

Additionally, and thankfully, the AI will now also be able to fact-check itself when prompted to do so.

Using the Retrieval plugin, ChatGPT will also be able to access sources such as databases in real-time.

Person holding up a picture lightbulb in front of a whiteboard full of gibberish

Code Interpreter

With the code interpreter, it will now be possible to write and execute code within ChatGPT.

Image manipulation will be as easy as uploading an image, requesting certain changes, and then downloading the edited file.

During his talk, Brockman demonstrated the use of this tool to analyze datasets and make exploratory graphs. The AI is able to suggest as well as create visualizations, which the user can then inspect as Python code if they wanted to.

Again, we can only imagine how many existing applications will now become obsolete because of this ease-of use within ChatGPT.

A person creating art with shapes of various colors

DALL-E integration

Brockman announced in his talk that a new DALL-E model will be available inside ChatGPT soon. (DALL-E is an AI image generator and another of OpenAI’s creations.) As an example, he showed how ChatGPT will now not only be able to suggest a meal, but also to generate a detailed image of how it envisions it.

As explained with the code interpreter earlier, the tools will be inspectable. This means users will be able to see how the machine is using the plugins, and to provide feedback. In the example of the DALL-E-generated images, for example, users will be able to “look under the hood” and see the prompt that ChatGPT generated in order to create the image.

A collection of old black and white photos

Memory

Brockman went on to explain that users will also be able to tell ChatGPT to save something for later. He then gives an example of how ChatGPT can use the saved information and integrate it with other applications: A user can ask ChatGPT to make a shopping list based on its suggested meal and then write a tweet about it.

ChatGPT will automatically select the appropriate tool, in this case Instacart, in order to create the shopping list. This brings us to third-party plugins.

Tiny shopping cart dangling from a finger

Third-Party Plugins

ChatGPT will also be able to integrate with various third-party plugins. For example, we will be able to ask ChatGPT to book us a table at a restaurant. The integration with Instacart will allow users to input ingredients, perhaps for recipes suggested by ChatGPT, and then have ChatGPT order them from the platform.

And the integration with Wolfram Alpha will provide access to a vast database of mathematical formulas, finally promising to make ChatGPT more adept at performing calculations.

Although not touched on in the TED Talk at all, we've already known that another promised plugin is Zapier, which will connect ChatGPT to 5,000 other apps. For instance, users will be able to connect ChatGPT with Gmail, Slack, and much more. This way, it can help us write emails, send messages, and so on, without us having to switch between apps.

Person holding a speech bubble

The Brockman TED talk

In the TED Talk, Brockman said the plugins demonstrate a new way of thinking about the user interface and provides a more streamlined experience.

He reflected on the progress made in AI since the founding of OpenAI seven years ago. He expressed his excitement and pride at the positive impact of the technology they are building. But he acknowledged that he knows some people have concerns. He said OpenAI has concerns too and that he believes we are entering a critical period in history. He stressed that OpenAI will strive to help manage AI for the good of society.

Brockman explained that it's not just about building the tools, but also about teaching the AI how to use them. To do this, he said, OpenAI used an old idea from Alan Turing's 1950 paper on the Turing test. The paper suggested that you can build a machine like a human child and teach it through feedback. This is exactly how they train ChatGPT – through a two-step process. The process involves unsupervised learning followed by teaching the AI what to do with those skills. The unsupervised learning process involves showing the AI the whole world and asking it to predict what comes next in text it's never seen before. This imbues it with various skills.

The second step involves teaching the AI what to do with those skills. This involves providing human feedback. This feedback not only reinforces the specific answer but also the whole process that the AI used to produce that answer. This approach enables ChatGPT to generalize and infer the user's intent and apply it in scenarios it hasn't seen before.

This way, they taught the AI to push back on humans in specific scenarios. He explained that the team also listens to users and gathers feedback through the thumbs down signal in ChatGPT and uses this to improve the AI's weaknesses.

However, he admitted, scaling our ability to provide high-quality feedback will be difficult, especially for harder tasks, and we will need to improve our ability to supervise the machine as we move forward.

The collaboration between humans and AI is a many-step process and, Brockman says, this delicate and careful design of humans and machines working together to solve problems will become more common in the future. By providing management, oversight, feedback, and inspectability, humans and machines together can create even more trustworthy machines that can solve, as he put it, impossible problems.

Lastly, Brockman cautioned that developing AI will require participation from everyone, including setting rules for how AI should operate. He stressed the importance of AI literacy for the general public as a step towards achieving OpenAI's mission of ensuring that AGI benefits all of humanity.

Labels:

Comments

Post a Comment

Popular posts from this blog

Why the Bots Hallucinate – and Why It's Not an Easy Fix

It’s a common lament: “I asked ChatGPT for scientific references, and it returned the names of non-existent papers.” How and why does this happen? Why would large language models (LLMs) such as ChatGPT create fake information rather than admitting they don’t know the answer? And why is this such a complex problem to solve? LLMs are an increasingly common presence in our digital lives. (Less sophisticated chatbots do exist, but for simplification, I’ll refer to LLMs as “chatbots” in the rest of the post.) These AI-driven entities rely on complex algorithms to generate responses based on their training data. In this blog post, we will explore the world of chatbot responses and their constraints. Hopefully, this will shed some light on why they sometimes "hallucinate." How do chatbots work? Chatbots such as ChatGPT are designed to engage in conversational interactions with users. They are trained on large ...
Post a Comment
Read more

Chatbots for Lead Generation: How to harness AI to capture leads

What is lead generation? Lead generation is the process of identifying and cultivating potential customers or clients. A “lead” is a potential customer who has shown some interest in your product or service. The idea is to turn leads into customers. Businesses generate leads through marketing efforts like email campaigns or social media ads. Once you have identified one, your business can follow up with them. You can provide information, answer questions, and convert them into a customer. The use of chatbots for lead generation has become popular over the last decade. But recent advancements in artificial intelligence (AI) mean chatbots have become even more effective. This post will explore artificial intelligence lead generation: its uses and methods. We’ll specifically look at a chatbot that has been drawing a lot of attention: ChatGPT . What is ChatGPT? ChatGPT is a so-called “large language model.” This type of artificial intelligence system ...
Post a Comment
Read more

Liquid Networks: Unleashing the Potential of Continuous Time AI in Machine Learning

In the ever-expanding realm of Artificial Intelligence (AI), a surprising source has led to a new solution. MIT researchers, seeking innovation, found inspiration in an unlikely place: the neural network of a simple worm. This led to the creation of so-called "liquid neural networks," an approach now poised to transform the AI landscape. Artificial Intelligence (AI) holds tremendous potential across various fields, including healthcare, finance, and education. However, the technology faces various challenges. Liquid networks provide answers to many of these. These liquid neural networks have the ability to adapt and learn from new data inputs beyond their initial training phase. This has significant potential for various applications, especially in dynamic and real-time environments like medical diagnosis and autonomous driving. The strengths of scaling traditional neural networks While traditional n...
Post a Comment
Read more