Microsoft confirmed yesterday the arrival of a new version of Bing boosted with the broad linguistic model and conversational AI like ChatGPT. We have all dreamed of a reinvention of research that CIOs must now understand to better educate all company employees about new uses.Microsoft puts AI ChatGPT in Bing, transforming everyone’s daily life.
Wait immediately. But very soon. This is briefly what we can remember from the press conference given last night by the publisher to formalize the new version of Bing doped with LLMs (Large Linguistic Models) which had been the subject of numerous rumors for a few weeks and especially since the dazzling success of ChatGPT from OpenAI, Microsoft’s preferred partner.
Without returning to the dispute of announcements in which Microsoft and Google have been engaged for a few days, this revelation of a new AI-enhanced Bing allows CIOs to put their finger on a reality of 2023: AI has recently a leap forward thanks to new linguistic models and not only transforms the interactions between man and machine but can become a real assistant in our daily lives, in homes and business. It’s not science fiction anymore.
It’s today! “This technology will reshape just about every category of software we know of,” said Satya Nadella, CEO of Microsoft, in the introduction. Today, we’re launching Bing and Edge, powered by co-driver and conversational AI, to help people get more from search and the Web.
A New Web Experience
Concretely, what does the new Bing do that the old Bing did not do? First, the new linguistic models allow the AI to understand your questions better, better understand contexts, better analyze web content, and ultimately offer more relevant occurrences to the user.
But above all, the AI automatically explores the most relevant links for you, analyzes their content, and presents you with a perfectly written summary (we know the potential of ChatGPT in this area), which summarizes most of the information it could glean. The great originality of this synthesis is that it does not lose the source of the data.
If a point deserves to be explored, click on it to access the original WEB page. Another significant new feature is Bing’s AI can discuss your concern of the day at any time and switch query results to a ChatGPT-style conversational mode with one click, allowing you to learn more, sort the results, filter according to your criteria, or extend the discussion well beyond the initial question.
This conversational mode can be activated without searching, for example, asking the AI which recipe to invent with the two eggs and the slice of ham in the refrigerator. Microsoft demonstrated how conversational intelligence could help you plan your next trip abroad.
Finally, Bing’s AI also has (like ChatGPT, from which it derives) creative power. She can summarize a long-to-read web document, turn your travel plan into an email to share with family, write a social media post in over a hundred languages, turn text into an image, and more.
And to fully exploit the creative potential and this co-piloting aspect of your questions, Microsoft is also launching a new version of its Edge browser with a new panel that allows you to engage in conversations with AI at any time and makes interactions between AI, web content, and user.
How Does BING AI Differ From ChatGPT?
On paper, the AI of BING and ChatGPT are based on similar but not quite identical technologies. ChatGPT relies on a GPT 3.5 model trained to chat with an information base frozen in 2021. Bing AI relies on more evolved models that derive from GPT-4 but are specially tuned for web searches.
These new models (which come with AI filtering and bias confrontation) result from a joint effort between OpenAI and Microsoft Research. They are called “Prometheus Models.” And unlike ChatGPT, they are connected to the Web (and therefore to the news) and regularly retrained.
Known Limits It is still too early to perceive the natural limits of what Microsoft has put in place. For now, the first limit CIOs will face is that to test these innovations; you have to use the development version of Edge, adopt Bing as the default engine, and register on a waiting list.
Microsoft wants to control a gradual and highly monitored deployment of its AI. We will collect feedback from users with early access to Bing to improve the tool before making it widely available. In addition to its temporary limitations (Microsoft promises that within a few weeks, the new Bing will be accessible to millions of people and that the AI extensions of Edge will be extended to other browsers), other limitations are already known:
- AI claims the right to make mistakes. It is up to the user to check the relevance of what it offers by taking advantage of the links to its sources.
- The Bing AI, at least in theory, is designed to “not get involved in potentially dangerous topics.” But it is entirely unknown what the AI currently perceives as a sensitive subject. It is not excluded that, at first, many issues are perceived as such and excluded from AI discussions.
- Not all requests require IA processing or, in any case, will not systematically trigger IA processing. This avoids overloading the Azure infrastructures and limits the environmental impact. But we don’t know which themes will typically trigger the AI or not.
- For now, this new search experience is only available on “Desktop” and not on mobile, although a new version of the Bing Mobile App has already been announced.
7 Of course, there are also questions about safety and relevance. “We have developed a security system designed to mitigate failures and prevent misuse with elements such as content filtering, operational monitoring, abuse detection, and other protections,” says the publisher adding to have designed its AI with a concern for keeping people at the center and demonstrating transparency.
And there are more than ever eternal questions about the American origin of this AI. She judges what is dangerous or inappropriate according to a rather American vision of the world, which is not always aligned with the European idea.
And seeing Google and Microsoft – two huge companies a priori more impactful than a startup like OpenAI – racing against each other also has cause for concern for those who would like a more controlled and mature approach before putting this kind of AI between all the hands to avoid impacts not yet perceived on populations, companies, industries or the media of the Internet (which could, in particular, see the audience coming from the engines decrease). These IAs also raise fears that the European IA Act currently being validated is already out of date before it is even finalized.
For CIOs, A Whole Potential To Be eExplored
Microsoft mainly focused on simple requests in a family setting during its press conference. But it is evident that in business, employees also spend a large part of their time controlling the web browser and search engines. Bing’s AI can therefore disrupt individual productivity. First, by providing more relevant answers and understanding more detailed questions than a simple succession of keywords.
But even more by its ability to explore occurrences to present you with a combined synthesis of their contents. And there, these are hours of exploration that will be won. Microsoft explains that half of the 10 billion daily queries submitted to Web search engines do not find an answer. In over 40% of cases, users click on a link and immediately return to the search page, instantly realizing that the related link does not answer the question.
Finally, this idea of conversation with an AI around any subject but even more around the question posed by the user transforms our way of using the Web and approaching the resolution of the problems which bother us. For CIOs, it will take some time and experimentation to identify the limits of this new approach to research and the Web. Understanding the current limitations of Microsoft Bing’s AI will also be necessary.
But there is little doubt that since the arrival of ChatGPT and now Bing AI, something has changed, and new uses are being born. And this is, of course, only a first step. Other AIs will arrive ever more innovative and ever more understanding of our human needs and moods.
Google Responds Without Replying
Google was holding a conference this afternoon around AI. The publisher wanted to recall that many of the innovations used by others (starting with the Transformer techniques based on ChatGPT) were the fruit of its engineers and that it had already put AI everywhere in its engine. Search in Maps, Google Translate, and in Google Lens (and its visual search).
He took advantage of his conference to demonstrate many AI innovations his teams are working on to make arts and culture more accessible, to transform Maps thanks to 3D views generated from 2D elements, and of course, to change research by adding a conversational dimension. But the publisher was careful not to expand on the subject and did not even demonstrate Bard, its conversational AI, announced earlier this week!
Failing to be able to respond immediately to ChatGPT and Microsoft BING AI, the research leader finally showed above all that he was out of step but that in terms of AI which transforms daily life, the race had only just begun. And the argument which explains wanting to be mainly “responsible” looks furiously like an excuse to justify why everything that has been shown will only be available in “several months”…