[ad_1]
Search engines have come a long way since the early days of the Internet. What started as a way to easily index web pages has grown into a sophisticated tool for finding authoritative and relevant information on virtually any topic imaginable. With Microsoft’s new AI-powered upgrade to Bing Search, called Deep Search, we are witnessing another quantum leap in our ability to leverage search engines.
See more: Bing Chat turns to CoPilot as underlying GenAI to compete with ChatGPT
Understanding the limitations of traditional search
Most traditional search engines rely on keyword matching to deliver results. A user types a search query consisting of a few words or a short sentence. The search engine algorithm then compares these keywords against documents and web pages in its massive index, aiming to find the best match based on relevance rankings.
This approach has served us well so far, but it has innate limitations:
- Matching on keywords alone often misses nuances in the language and fails to accurately reflect the intent behind the search query.
- Ranking systems are not infallible and relevance is still hit or miss for more complex or ambiguous queries.
- There is very little real understanding of context; the search engine has no insight into the bigger picture behind the search query.
As questions become more versatile and searches become more conversational, traditional search methods begin to falter. Can AI and natural language processing help overcome some of these disadvantages?
Introducing Deep Search with GPT-4
Deep Search is Microsoft’s solution for developing search capabilities by adding AI to it. The function uses a technique called generative prior education which researchers have found very effective for natural language tasks.
Generative pre-training works by first exposing the AI model to a vast corpus of text data from the Internet, allowing it to build substantial understanding about concepts, contexts, and the relationships between them based on patterns in the data. The model can then apply this insight to tackle specific tasks or answer questions by generating newly constructed answers.
For Deep Search, Microsoft refined this search technique by pre-training the massive machine learning model GPT-4 on web pages, documents and other content. With its enormous capacity and speed, GPT-4 is ideally suited for this purpose.
Delivering more relevant and complete answers
So how does this increased comprehension translate into better search results for users? Deep Search is enhanced when you enter queries that have a complex intent or that could use more context around them.
For example, if you search for “jobs that help the environment”, traditional search would simply spit back a list of web pages containing those keywords. Deep Search, on the other hand, understands that you probably want an overview and specific examples of professions that fit that description. So an entire passage is put together summarizing relevant jobs and why they benefit the environment.
The function processes queries such as “ELI5 quantum computing” or “new thriller books 2022” in a similar manner, distilling concepts into layman’s terms and providing comprehensive answers. For more subjective topics, it can highlight multiple perspectives.
Also read: How to Use Frosting AI: Unleash Your Creativity
Rank results based on real-world relevance
Another big advantage of Deep Search is its more critical – and dare we say ‘smarter’ – relevance ranking. By understanding the full context and intent behind searches, it can assess the quality of results based on more parameters than keyword density.
For example, a search for “most influential classic rock bands” would likely value size of impact and longevity over just matching keywords. Deep Search is designed to make these types of reasoned judgments among the billions of possible results to better reflect the user’s likely preferences.
Early testing shows that queries need to be rephrased by about 70% to get quality answers. That’s the magic of an AI assistant that understands language and acts as an intermediary between you and the raw index.
Conclusion
Deep Search undoubtedly brings some intriguing improvements to search. Allowing AI to generate written summaries and responses directly addresses the shortcomings previously described in keyword matching and relevance. Tests show it reduces the number of results people click on by more than 40%, indicating it better meets information needs.
However, there are some obvious limitations in these early stages:
- The feature is slow and takes up to 30 seconds to generate responses.
- Accuracy and relevance are still inconsistent for certain types of queries.
- Longer search lengths can reduce coherence and quality.
Nevertheless, the technology is promising. And given Microsoft’s vast computing resources and steady progress in natural language AI, we expect rapid improvements in power and performance.
Just as game-changing leaps like Google’s PageRank transformed web search decades ago, embedding strong language understanding into search indexes could drive another seismic shift. Only time will tell if Deep Search proves to be that turning point, but the potential is undeniably sparkling.
🌟 Do you have burning questions about a “Bing GPT-4 Deep Search”? Do you need some extra help with AI tools or something else?
💡 Feel free to email Pradip Maheshwari, our expert at OpenAIMaster. Send your questions to support@openaimaster.com and Pradip Maheshwari will be happy to help you!