* This blog post is a summary of this video.

4 Levels of Search: How Bing is Transforming Mobile Phones

Author: AI ExplainedTime: 2024-02-04 02:30:01

Table of Contents

Introduction: Bing App Now Available on Mobile with Voice Recognition

Bing app now available on Android and iOS is about efficiency over extravagance. It's on your phone, so it's not about essay writing, compiling code, or making funny poems in French. It's about taking search to the next level. I'm not sponsored by them, but I think it's mind-blowing. There are four levels of search increasing in power and requisite intelligence, and Google is pretty much only able to do level one and even then not always. Bing can do all four. Let me show you in this video exactly what I mean, taking you through all four levels of search. By the end, I honestly think I'm gonna persuade you that smartphones just got upgraded permanently.

Just quickly, how does it look on mobile? If you open up the app, you will see at the bottom a Bing button on any web page just like you can in the Edge browser. You can press on the Bing button and open up Bing chat. You then have two options: you can ask it questions via the microphone or via the keyboard. For example, you could ask who is currently seventh in the Premier League.

Voice Recognition Makes Bing Faster Than Typing

That's another fascinating difference with Bing on mobile - it actually speaks to you. I wouldn't call it cutting edge text-to-speech, but we're just getting started. You're probably wondering how does this transform search? How does this upgrade our smartphones? Well, this is just a level one search. We are retrieving one bit of information, and even on this front, Bing does better than Google.

Bing Understands Context Better Than Google

The exact same question into Google gives a generalized formula about the Premier League and what seventh place means. Bing understands exactly what I want and gives the correct answer. I will admit, if we were just comparing level 1 searches, Bing wouldn't be that much of an upgrade. You could always click on a link to a table and see for yourself where Brighton are. Maybe you're saving a few seconds with Bing, but not really a big difference. But just wait until we get to level 3 and even level 4 and beyond searches.

Level 1 Search: Basic Information Retrieval

Now you guys are ready for level 2 searches. And what do I mean by that? This time we are retrieving two bits of disparate data, but we want to do it at the same time, not doing two separate searches. By the way, I'm typing these searches into Bing desktop so you can see them more clearly, but of course it would be even quicker to ask them with my voice into Bing on mobile.

Bing Directly Answers Premier League Table Question

I asked what are the ages of the Eiffel Tower and the Empire State Building, and look at the difference. I can clearly see the two results on the left in Bing, but on the right I'm gonna have to click on at least two links and scroll through the data. You can just begin to imagine the number of searches this could apply to, and we're only on level two.

Google Gives Generalized Response

The next example of a level 2 search would be to retrieve a bit of information and do something to it. For example, I asked both Bing and Google if Microsoft's market cap doubled, what would it be? There are two stages to that question: First it has to retrieve Microsoft's current market cap, then it has to double it. Bing gets the answer, Google doesn't even show the market cap, not immediately at least.

Level 2 Search: Retrieving Disparate Data

Even here I know some of you will be thinking I could just type in market cap, find the source, get out a calculator, and double it. What's the big problem? Yes maybe you save 30 seconds, but what's the big deal? Well we haven't even gotten to level 3 or level 4 searches yet. So what is an example of a level 3 search?

Ages of Famous Buildings

Imagine you're on your smartphone and you're considering the Adobe Creative Cloud. And imagine you wanted to know just how much more expensive it would be over say a couple year time period than DaVinci Resolve. You could press the Bing button and ask this: 'According to this page, if I got the individual account for two years, how much more expensive would that be in pounds than the one-off payment for DaVinci Resolve 18?'

Company Financial Data Calculation

Now as I've talked about in my other Bing chat playlist videos, it understands the context in which you're asking the question. It knows you mean this Adobe Creative Cloud individual plan. It correctly multiplies this for two years and then compares the total to DaVinci Resolve's price. Now initially I thought it made a mistake because when I quickly checked on Google, DaVinci Resolve costs 255 pounds, but then when I click to buy it, it adds on VAT which makes it 306 pounds in the UK. So in a way, that's another win for Bing - it understood about adding on VAT.

Level 3 Search: Comparison, Analysis, and Reasoning

But what makes this a level 3 search is it did all of that work - it retrieved the two bits of information in context, and then compared them, then subtracted them, giving us a difference of 941 pounds in the price over two years. And of course you could carry on this conversation with Bing about the pros and cons of each package.

Software Subscription Cost Comparison

For some of these searches, I am not even going to try it in Google because it would completely flop. Level 3 searches are about much more than this though. Imagine you're standing on the Houston Road and you want to get to Central London. You could conduct a level 3 search using Bing. The question might be: 'How much longer would it take to get the Underground from King's Cross to Piccadilly Circus than from Houston to Oxford Circus?' Or 'How much longer would it take to go from King's Cross to Hyde Park Corner than from Houston to Victoria?'

Travel Time Comparison for Specific Routes

These are all journeys that I make on a regular basis, and I can confirm that the results are accurate. Why is this level 3? Because it had to retrieve the information about one journey, then the other, and then make the comparison.

Animal Size Comparison and Explanation

Our level 3 searches all about addition and subtraction? No, check this out. You could ask how much bigger are polar bears than brown bears and why. Google would have to do three things and it just isn't up to it. You'd have to find the size of the average polar bear, the size of the average brown bear, and then do an analysis of why polar bears are bigger - not just a mathematical comparison, but an understanding of comprehension, an explanation of the why's.

Level 4 Search: Complex Multi-Step Reasoning

Think of level 3 as adding a when, where, why, and how to level 2 searches. The answer, by the way, is quite interesting. So polar bears can weigh up to 1,700 pounds versus 1,320 pounds, but we didn't just want to know that. We wanted to know the reason why. And apparently the reason why is, and I can believe this, is that they need more body mass and fat to survive in the cold Arctic environment. They also have a bigger skull and larger teeth to hunt seals. So now I've got more of an idea - not just how much bigger they are, but why through evolution they ended up being bigger.

Ancient Site Age Comparison in Human Lifetimes

But we have waited long enough, what about level 4 searches? Well, think about a complex, interesting search like this: How much older is Stonehenge than the Coliseum in Rome expressed in human lifetimes? Bing has to do four things: find the age of Stonehenge, the age of the Coliseum, the difference between them divided by the average human lifetime. It does this successfully, and we have a really interesting result that is about 38 human lifetimes older if we take the older date for Stonehenge. That is an insane level of intelligence for a single search.

Sports League Standings Extrapolation

Now we're genuinely talking about saving a minute or more compared to using Google. That is a big enough difference to really matter. And that's not the only example I could give you of a level 4 search. I'm sure you could tell me hundreds of ideas in the comments, but try this one: Going back to the Premier League, I could ask 'If Arsenal beat Leicester and Man City draw Bournemouth, how many points ahead would Arsenal be?' I didn't have to specify Premier League. I didn't have to say fixture. And of course I didn't have to tell it the rules of the Premier League about 3 points for a win, etc. It knew exactly what I was asking, calculated the two results, found the league positions, and then found the difference.

Conclusion: Smarter Assistants Transform Mobile Experience

Now you don't have to be into sport to know that that's an amazing answer to a complex search query. Think about how this applies to your domain, your career, your interests, and come up with some level 4 searches that you could ask, which brings me on to my final point: The question for every smartphone user will be, is the small but real risk of hallucinations more meaningful to me than the additional seconds and minutes required to perform multiple searches? For me the answer is already yes, but clearly what we decide will depend on the topic. Right, for sport it's fine, for maybe a life-changing house decision, maybe not.

Risk of Errors vs Time Savings

By the way, with those new modes that Bing is debuting this week that I'm going to do a video on, where you can pick precision over creativity, soon you might not even need to make a choice between hallucination versus efficiency. Very much looking forward to doing a deep dive into those modes, by the way.

Other AI Assistants Playing Catch-Up

And yes, what you may find is that Bing when you do voice recognition makes the occasional mistake in what you're asking it. That certainly happened to me. OpenAI, who are the partners to Microsoft, are responsible for Whisper, which I have downloaded locally and tried out. It is phenomenal at voice recognition, as good arguably as human transcribers. Now I don't think Whisper is powering the current Bing voice recognition, but because of Microsoft's partnership with OpenAI, I wouldn't be surprised if by the end of the year Whisper doesn't power Bing, and that should make voice searches almost flawless. I know what some of you are thinking, but what about the conversation limits? Well, according to my sources, they should be ending soon too. Now of course we are days or maybe weeks away from Google Bard's system being released, but I've got a feeling that Google doesn't want those multiple searches to go away anytime soon. Bing, on the other hand, doesn't care. So I'm not sure there's much incentive for Google to maximize the efficiency of Bard. Early indications about the size of the Lambda model that's powering Bard and the leaked rumors that every Google employee is spending two to four hours a day to improve the output of Bard suggests that my feeling might have some basis. But maybe Google will surprise us all, or Facebook - they just released LLaMA, a large language model that outperforms even PaLM and therefore likely Bing on some metrics. I'm hoping to use it soon. And Amazon, well, they just released a model that outperforms GPT-3.5 by 16%, so they're not out of the game either. By the end of 2023, who knows who will be the king of search and the king of your smartphone? But it will definitely be a smarter smartphone if you agree, or even if you disagree, let me know in the comments. Please do leave a like, and thank you very much for watching. Have a wonderful day!


Q: How does the Bing mobile app upgrade search?
A: The Bing mobile app can handle multi-step searches at level 3 and 4, delivering results much faster than manual lookup.

Q: What are some examples of level 3 searches?
A: Level 3 searches involve comparisons, analysis, reasoning such as comparing travel times, software subscription costs, or explaining why certain animals are larger.

Q: What are some examples of level 4 searches?
A: Level 4 searches require complex, multi-step reasoning like comparing ancient site ages in lifetimes or predicting sports league standings based on hypothetical match results.

Q: How accurate is Bing's voice recognition?
A: Bing's voice recognition is very accurate thanks to Microsoft's partnership with OpenAI, though occasional errors can still occur.

Q: Does Google's upcoming Bard have multi-step reasoning?
A: It's unclear if Google Bard will match Bing's level 3 and 4 search abilities out of the gate. Bard may focus more on conversational pleasantries initially.

Q: What are the downsides to AI assistants like Bing?
A: There is a small risk of errors or hallucinations. Users must decide if the time savings outweigh this risk for important decisions.

Q: How could Bing improve further?
A: Integrating OpenAI's Whisper voice recognition should make Bing's speech interpretation nearly flawless. Removing conversation turn limits would also help complex searches.

Q: How quickly is AI progressing?
A: Very rapidly. Google, Facebook, Amazon and others are all working on large language models. The leader in AI assisted search could change by end of 2023.

Q: Will smartphones get smarter?
A: Yes, AI assistants like Bing will transform smartphones by enabling faster information retrieval and multi-step reasoning while on the go.

Q: Should I switch to Bing on mobile?
A: For casual search, Bing provides a streamlined experience. For high-stakes decisions, compare results across multiple sources.