Chaperon ChatGPT Hand Holding

Building Bots to Chaperon ChatGPT

As ChatGPT and AI continue to fill headlines, the conversations have shifted entirely towards productivity and automation—a surprise. As we continue to think of new ways to use ChatGPT, and new automation, there is also an unease about the information that the AI is accessing, and how this information might be used or dispersed.

When we talk about bots chaperoning ChatGPT, there are ways that we might interpret this. One is to have a bot hold ChatGPT’s hand while it completes a task for a certain context or a certain audience. Another would be making sure it doesn’t step out of line ethically or with sensitive information. Lastly, we might also be talking about a more extended idea of automation—truly letting the AI run by itself. Let’s cover each of these possibilities.


Personalizing a Chatbot

First things first, how do you get ChatGPT to work for you? For example, it’s becoming more and more common for businesses and brands to have a sort of “AI Chatbot” on their website to help answer certain questions for customers. A big problem with these bots is that they feel aggressively dehumanized—you’re clearly talking to a bot that’s been told to use emojis. Getting a bot on your site, your app, or technology that has the same conversational ability and even charm as ChatGPT can be a game-changer for user experience.

Fortunately, this is absolutely something that can be pulled off. OpenAI offers the ChatGPT API and other API’s for a monthly fee. This allows developers to integrate the ChatGPT technology into their own products, websites, and apps. The model can also be trained on private information, which would enable it to answer questions specific to your organization. You could direct users to instead, but the general ChatGPT model probably won’t have too much information about how your website or company functions—especially given its timestamped data restriction.

Several services like Custom GPT have been formed to do this dirty work for you, but ChatGPT itself can also be used as guidance to building your own bot. For example, this Twitter bot that Rakshit was able to construct in a single day, without knowing too much programming in the first place. With a few prompts and API licenses, you can be on your way to making sure ChatGPT is working for you, on your website, with your style, and with your branding.


Setting and Protecting Boundaries

A significant consideration related to the previous point is that the model has to learn from your data. This is something that has to be done with care. Chances are, within your private datasets, there is sensitive information that probably should not be accessible to just anybody for a variety of reasons. If you set the AI model loose on all of your data, there’s always a chance the corresponding failsafe will fail if someone tries to coax that information out of your bot. That’s why taking strict measures to protect sensitive information is an important part of implementing automation safely.

One of these measures is taking care of which data the model has access to in the first place. Before training AI on your data at any level, you should take time and put good thought into what information is absolutely non-public, and which can be very useful for a bot to know. We’ll talk about another failsafe in a bit, but in the worst case where no failsafe works, you don’t want your AI bot to be spouting off company credit cards. That’s why, at the very least, being a chaperon to ChatGPT has to include not telling your model everything.

Another safety measure can come in the form of another bot. Just like the ChatGPT API allows individuals and organizations to create AI chatbots tailored to their contexts, there are also AI bots that can help filter sensitive content before it causes harm. These are common in forums and other public platforms and are usually put towards censoring harmful language. However, trained to these cautions, these “censorship” bots can be used to make sure your ChatGPT bot doesn’t release sensitive information, even if the sensitive information was cleared by the AI’s filter initially. An important tradeoff to consider, though, is these kinds of filters can be computationally and often financially expensive to run continuously. Additionally, they might make interaction with the bot a little less quick. This means you ultimately have to decide between how quickly you want users to access information, and how safely that information is being guarded. Either way, though, it’s not something to overlook.



Finally, what about the “automatic” part of automation? Most of these bots automate a big chunk of work, but they also depend on specific prompts and guidance. Whether it’s using the ChatGPT API or other ways to make tasks fully automatic, and create a bot to ‘chaperon’ the model. This enables it to perform your tasks whenever the bot fires off the AI.

Once again, a basic example is this Twitter bot designed to identify and interact with certain tweets completely autonomously. This way, you can build AI tools that don’t necessarily require human input to function, and that can continue to function outside your supervision. When using ChatGPT itself as a guide, the barrier to entry to this sort of bot building is lower than ever before. With realistic goals about what you want to achieve, you’re probably able to use ChatGPT to create all sorts of bots like this with little-to-know actual programming experience.

We should make no mistake, there is still quite a way to go before we deal with fully autonomous, fully functional AI models. The automation we can program now isn’t always very complex and can be quite computationally expensive. Most of the impressive AI models come from proprietary code, meaning that the actual “thinking” that the model does is a bit of a black box. This greatly limits the extent to which ChatGPT and similar models can truly be autonomous, or even truly need to be chaperoned. That said, there are certainly useful and powerful automation applications of this software, given that they are taken with the usual set of caution and precautions. 

Living Pono is dedicated to communicating business management concepts with Hawaiian values. Founded by Kevin May,  an established and successful leader and mentor, Living Pono is your destination to learn about how to live your life righteously and how that can have positive effects in your career. If you have any questions, please leave a comment below or contact us here. Also, join our mailing list below, so you can be alerted when a new article is released.

Finally, consider following the Living Pono Podcast to listen to episodes about living righteously, business management concepts, and interviews with business leaders.

Leave a Reply

Your email address will not be published. Required fields are marked *