What is Sundar Pichai thinking?
There is an obssession for the company. And they are spending a lot of money on it.
Hello all,
As promised, this week will be Google themed. The story for this week will be a birds-eye view of recent Google’s research into AI and Machine Learning. AI and Machine Learning are definitely sectors that will continue to grow exponentially in the future (this is totally not my bias speaking). And since many of you are gunning to get into Google, knowing the direction the company is taking will be very helpful.
Some Context
If you just casually go over the content of Google’s AI research, it won’t really make sense. The company seems to be involved in everything under the sun, from Computer Vision, Math solver, Neural Architecture Search, Deep Learning Theory and so much more (we’ve covered all of this on my YouTube and Medium, so make sure
you’re connected with me there so that you don’t miss out).
And this month, they created an absolutely mammoth protein classification project that could classify a protein correctly out of 18,000 labels (and many other accomplishments).
Their projects have been exceptional, and no doubt have progressed the field of Machine Learning research, but you might be wondering, “What was the point?”. Google is a for-profit company, so they invest in all of this? What are they hoping to gain from this? Is there a method to their madness, that we can decipher?
Why Companies Spent Billions on Deep Learning
Let’s think about how Google makes money. It’s easy to say ads. And that is true. They use all the user data collected through YouTube, Maps, Search, Travel, App Store, and their other services to build up a user profile. Then let businesses and users advertise get to advertise to very specific profiles. This is a much higher ROI than a general ROI than traditional ads like banners and commercials because they reach people who would need something. Google knows I’m a digital nomad (through my searches, purchases, and constant location change). So I get a lot of ads for hotels, traveling services, online services, etc. As a YouTuber, I also get tons of ads for digital assistants, cameras, courses, etc. All of these are products that I am more likely to buy when compared to my mom, who doesn’t need any of this.
To do so, we have to be able to get information from large amounts of data. This is where Machine Learning can really shine. It can take the giant amounts of data that we give it and provide us with some very interesting insights. Coming up with these insights is quite literally my full-time job. The reason companies pay so much for this service is simple, Machine Learning operates at scale. If you can make decisions 1% better, the return becomes massive after 1000 decisions (1.01^1000=21,000). And Google’s AI is taking trillions of decisions regularly.
It would be easy for me to end things here and call it a day. And worse analysis services would do that. But you guys are here because you want better analysis. And that is what I shall give you. So, what exactly is Google hoping to accomplish with their projects?
Deciphering a trend
What do these projects have in common (aside from the fact that Google sponsored them):
AlphaGo: A reinforcement-learning-based AI agent that could play the notoriously hard Go. It beat the World’s best, shaking up the community.
AlphaCode: An algorithm that could read question prompts given in competitive coding challenges and actually solve them. Slightly better than competitive coders can.
A Recurrent Neural Network based AI agent that could create new activation functions for deep learning. Extremely well. Read more about it here.
Multiple iterations of AI, all aimed at cracking the Neural Architecture Search problem (automating the creation of neural networks).
Take a second to think about it. Don’t scroll down to check immediately, that’s no fun.
The answer is that they all solve search problems. Number 1 was able to go through many possible moves to find the best one. AlphaCode generated and tested code samples relevant to the problem. To do so it would have to effectively search over 20,000 choices. Number 3 searched through solution spaces for good activation functions, and 4 searched for good Neural Network configurations.
I’m not making this up. These were all framed as search problems in their formulations (I read all the papers). If you are interested in learning about the technical details, make sure you use the links at the end to connect with me on different platforms (especially LinkedIn, YouTube, and Medium). I won’t go beyond this here since not everyone in this newsletter is into Machine Learning.
So how do these various search problems help with Google’s Business Model? Google has already dominated the search engine, smartphone, and many other important data markets. So it’s not as though they have competition. Time for the final section.
Google’s Masterplan
Google has data. Lots of it. And they already have very refined AI systems that ensure that advertisers and businesses keep bringing Google their dollars. So what is Google trying to accomplish?
There are two ways that a company can boost profits. They can either boost their revenue or reduce operational costs. At this point, adding to their ML will not help them get more customers. All the big businesses are already advertising on Google products either way. So the marginal utility in customer acquisition will be minimal.
However, that doesn’t tell the full story. As they continue to refine different search algorithms they will be entering the next frontier of advertising: hyper-specific advertising. This will be a level of personalization of ads never possible without extreme investment and very refined AI. They will be able to create user profiles at a level of detail impossible before. To do so, the ability to search through large amounts of data and come up with combinations of characteristics will be crucial. Doesn’t that sound like the search problems we discussed?
Once Google gets this rolling, things will really kick into another level. They would be able to cater to their customers at an unprecedented level of efficiency, which ultimately would allow them to bring in more revenue.
Another move that this would facilitate would be an entry into the consulting market. As a creator, and someone involved in a startup, I know how much work things like Customer Discovery and Market Research are. With all the data and insights Google generates, is not farfetched to see how they could use their analysis to consult businesses. Such consultations will give them very fine-grained data by working with different customers, allowing them to refine their products. This would start a positive reinforcing loop between the core business and the consulting verticle.
The SAAS (Software as a Service) market would be another such avenue. Google could start packaging and selling the insights as different scripts so that businesses could leverage insights without paying for consultations. Google’s investments into BigQuery and constant refinement of their analytics platform hint that they are looking at this avenue.
However, it doesn’t end here. Google can also use search to reduce operational costs. This is actually something that I’ve worked on personally. We frame the system (or aspects of a system) in various ways, and then use AI to highlight possible areas of redundancy and find the least/most important aspects. Companies can then figure out how to handle this waste. At Google’s scale, they will be able to run these searches on complex multi-objective functions. This would require the development of very advanced search algorithms that can traverse various search spaces. And shockingly, Google is doing exactly this.
So how can you use this to benefit yourself?
To maximize your chances of leveraging this information, you need to (very ironically) get into the basics. Solidify your Math, theoretical knowledge, ability to frame ambiguous situations, and coding skills. These core skills will allow you to tackle problems at scale (which is what Google is doing). This is why this newsletter focuses on so many aspects, instead of just giving you hundreds of questions+answers (like AlgoExpert, Leetcode, Cracking the Coding Interview, and other services). Their approach will make you better at solving their questions (and thus doing well in your interview), but will not be helpful for too much beyond that. This newsletter will get you good at both.
That’s about it for this email/post. If you come across any interesting stories, feel free to reach out to me using one of my social media links. Would love to talk to y’all about it.
In the comments below, share what topic you want to focus on. I’d be interested in learning.
If you liked this post, make sure you fill out this survey. It’s anonymous and will take 2 minutes of your time. It will help me understand you better, allowing for better content.
https://forms.gle/XfTXSjnC8W2wR9qT9
Happy Prep. I’ll see you at your dream job.
Peak Analyst,
Devansh <3
To make sure you get the most of Finance Fridays, make sure you’re checking in the rest of the days as well. Leverage all the techniques I have discovered through my successful tutoring to easily succeed in your interviews and save your time and energy by joining the premium subscribers down below. Get 20% for a whole year by using the link below:
Reach out to me on:
Instagram: https://www.instagram.com/iseethings404/
Message me on Twitter: https://twitter.com/Machine01776819
My LinkedIn: https://www.linkedin.com/in/devansh-devansh-516004168/
My content:
Read my articles: https://rb.gy/zn1aiu
My YouTube: https://rb.gy/88iwdd
Get a free stock on Robinhood. No risk to you, so not using the link is losing free money: https://join.robinhood.com/fnud75