ChatGPT

ChatGPT’s Integration into Microsoft Bing: A Game-Changer for Search?

Bing is rumored to get a significant upgrade with ChatGPT integration: here’s what it means for users.

In the last few days, there has been a lot of buzz about a potential collaboration between Microsoft and OpenAI. The rumor is that Microsoft is planning to integrate ChatGPT, a free chatbot powered by the powerful GPT-3 large natural language model, into its Bing search engine. If this rumor turns out to be accurate, it could be a game-changer for the search engine industry, and maybe Google should seriously start to worry.

While this news is making headlines, it should not be forgotten that the two companies had already signed an agreement for the exclusive use of the GPT-3 model for commercial purposes. In fact, it has been a year since GitHub, now owned by Microsoft, integrated the GPT-3 model into its service, offering developers a way to get assistance while coding. This is done through a variant of the model trained specifically on coding-related tasks. The feature has been well-received by many, and has also raised questions about the future impact such assistants may have on developer jobs. In one of my articles, I recently explored this area with some specific tests to gain insight into the potential of these tools:

Now, returning to the rumor, it seems that the earlier deal has been expanded to a whole new level — integrating a modified version of the ChatGPT model into the Bing search engine. While language models like those from OpenAI being integrated into search engines in some form is certainly something everyone expected in the future, no one could have predicted it happening so fast. On the other hand, the possibilities opened up by these tools are clear to everyone. Just a few weeks ago, I wrote an article on how these models fare in typical search engine tasks:

What has come out of those tests is something that many have observed while experimenting with ChatGPT, which can be used for free through the official website https://chat.openai.com/chat at least at the time of writing this story. And that is, that while being extremely powerful at understanding complex queries thanks to its language model, it falls behind in instances where access to fresh information is needed. In fact, one of the peculiarities of these language models is that they are trained with a static set of information, the training dataset. This is done once, in the case of ChatGPT it happened in 2021, and requires a huge amount of computational power. Once done, it is difficult to make the model learn new information as it is being used, and in fact, a large part of today’s research is focused on this issue. Until now, this process is limited only to partial introduction into the model dataset of new information, such as through fine-tuning techniques. But it remains to be seen how the continuous injection of updated information needed for a traditional search engine to work would be worked out.

What this all means is that the model cannot answer, for example, to questions regarding an event that happened after it was trained, making it always outdated in its answers compared to a traditional search engine.

But this does not mean that models like ChatGPT are useless. As I mentioned in my article comparing ChatGPT and Google, one of the areas where models like this can really stand out is in formulating answers to complex queries — queries that are not limited to simple information seeking, but are more akin to the actual articles we’re used to searching for on Google. One of the areas where people have already started to appreciate its capabilities, for example, is related to development and coding. Until now, developers were accustomed to turning to Google for simple queries related to coding issues, often finding a solution on StackOverflow and copying and pasting it into their code. But now, many are turning to ChatGPT, at least experimenting with it, and in some cases replacing Google in their daily workflow for these particular tasks. The fact that ChatGPT is a conversational model makes it possible to present a follow-up question to the assistant in order to better shape its answer to our needs, such as asking it to clarify some aspects that were not clearly explained. These are things that are simply not possible with the typical Google search workflow.

Of course, you can find many other tasks performed by the model for which experiments have been done by people from every background, but these tasks may yet not be enough, meaning they might not provide everyone with added value to make it worth integrating into a traditional search engine.

Conclusion

That said, the growing number of users of ChatGPT seems to suggest that many are appreciating at least some of its capabilities. So, it’s only a matter of time before we see what the future Bing will offer. It will be particularly interesting to see how Microsoft will be handling the accuracy of the information given in output from the model, since at the moment there’s no way for the user to check the reliability of the provided answers. Anyway, it remains to be seen whether or not Microsoft will follow through with this rumored integration, so we’ll have to wait and see if and how it will pan out.

Further reading

Thanks for reading through. If you liked 👏 this article you may enjoy reading through some of my other articles. Oh, and don’t forget to subscribe! 🫵