Microsoft has a new version of Bing in testing powered by OpenAI’s GPT artificial intelligence. However, the answers that Bing offers are very different from those of ChatGPT. It is important to understand the main differences between the two tools.
Microsoft offers those interested to try the new version of Bing with a simple registration. We now have access to it and have started asking it questions to test the new search engine.
We quickly noticed big differences with the ChatGPT responses. However, the engine has often been presented as a simple ChatGPT integration. How to explain these differences? Microsoft didn’t just integrate the OpenAI tool.
How Microsoft created the new Bing
Here’s brick by brick what makes up the new version of Bing.
The OpenAI engine and a new version of GPT
The basis of the new system is a next-generation language model. ChatGPT is based on GPT-3.5, but the model used by Bing is ” much much more powerful according to the signature Microsoft has invested heavily in OpenAI and is the startup’s preferred partner. This model would have been optimized for the particular research.
The Prometheus model
Beyond this language model, there is the “Prometheus model“, a set of proprietary technologies and techniques developed by Microsoft. It is about interrogating the OpenAI model in a more intelligent way, both during the training phase of the AI and during its execution.
It is to this brick that Microsoft begins to infuse the indexing of the web operated by Bing. This would allow the system to be more accurate and reliable, but it could also annotate responses and provide sources to users.
It’s also thanks to the Prometheus model that Microsoft is able to keep its AI up to date with the latest data from the web. This results in sometimes annotated responses to very recent tweets or articles.
Without going into details, Microsoft explains that it applied the Artificial Intelligence data model to the Bing index to improve the relevance of search results. According to the firm, this is the biggest improvement in web search relevance in two decades.
Works ? our test
We wanted to put the new Bing to the test in order to verify Microsoft’s claims, and in particular, compare the responses with ChatGPT. If it were a simple integration of the latter, the answers should be very close, if not identical.
To illustrate the limits of ChatGPT, Marie Turcan published an article on January 31 wherenumerallaunched the queryName 5 women who have been Prime Ministers in France“. This is a misleading request as only two women have been Prime Ministers in the history of France.
The question posed to ChatGPT // Source: Numerama
Answer from Bing // Source: Frandroid
This is a landmark question where ChatGPT completely failed when Bing’s conversational AI was able to detect the trap and answer correctly, with the correct sources. This simple test already demonstrates the difference between the OpenAI tool and the one from Microsoft.
Bing is still limited
However, today it is still very easy to blame Microsoft Bing. Microsoft’s AI is able to confidently assert false things, despite questioning the user. We were able to see this with several of our requests on personalities and this is also what Nicolas Lellouche was able to notice in his tests.
Let’s hope that Microsoft comes soon to make corrections to the new Bing systems. In the current state, if the progress against ChatGPT is significant, the firm also exposes its service to the general public a lot more. Where ChatGPT is a beta service with a limited audience (100 million users worldwide) where the user quickly wants to understand that they should not trust AI responses, Microsoft Bing comes across as a reliable tool.
In this article we have focused on Microsoft’s solution, but keep in mind that the competition is also preparing as Google Bard.
Do you want to join a community of enthusiasts? Our Discord welcomes you, it is a place of mutual help and passion for technology.