In this week's Sage Spotlight, Yash Desai, Machine Learning Manager at Meta, shares his views on the AI / ML market and opportunities for vendors to gain market share.
Key Takeaways
- AI Integration Speed Challenges: The rapid pace of AI development creates challenges associated with integrating the latest research into practical applications within business settings. This cycle of adopting new AI advancements can take three to six months, by which time new research may have already surpassed the previous findings.
- Monetization of AI in Business: A significant emphasis is placed on monetizing AI technologies. The real challenge lies in optimizing the cost and infrastructure required to effectively use AI models for maximizing revenue.
- Importance of Productionizing AI Models: Yash highlights the importance of "inferencing" and the productionization of AI models. It’s critical to not only create theoretical models but also to connect them with real-world data and business problems for effective application.
- Vendor Opportunities and Challenges: There is a strong need for vendors that can navigate the complex integration landscape, which varies from company to company. This includes dealing with custom solutions, third-party services, privacy regulations, and technical infrastructures like cloud or on-premises environments.
- Advice for Vendors Working with Large Corporations: Desai advises vendors, especially those at earlier stages, to focus on understanding and addressing the privacy and legal aspects when pitching to large companies. This understanding can significantly ease the process and focus more on the technological solutions rather than getting bogged down by legal hurdles.
Full Transcript
Yash Desai: So my name is Yash. I'm currently a machine learning manager at Meta, and I lead one of the teams within ads ranking, which is, you know, the cash cow at Meta.
The fun challenge is we're witnessing an AI revolution right now,
and the challenge is keeping up with it, right? So when you're in the industry, you want to take the latest research, apply it in your real life setting, and then seeing the benefit of it.
This whole process, this life cycle can take you anywhere from three to six months. By the time you're done with it, the research has already gone ahead. So how do you keep up with it in a way that you are being able to apply it? But at the same time, also not like lagging behind the the state of the art.
but the real challenge is how do you monetize them? You know, how do you squeeze the juice to get the full revenue out of it?
Because there's a lot of inferencing challenges. There's a lot of challenges around the cost, infrastructure, and actually making it useful in your business setting, right? And that to me is like one of the key challenges which I think industry is talking about more in silos, but I guess not coming together to, you know, figure it out as a whole.
Sahil Khanna: Do you have a perspective on the biggest opportunities for vendors to be gaining market share?
Yash Desai: Yeah, I think the top of mind for me is inferencing. We have some amazing solutions literally with the click of a button to produce some of the best models.
Uh, you can take the code, you can copy paste it, change the conflict, and there you go. Boom in, in like less than a day, you have a model up and running. But now how do you actually. You know, make use of it. So this inferencing, this productionization part is where the key lies, because now you need to connect your, your real data, your real business problem to this, you know, theoretical model.
I think that to me is the, the key area which I'm looking for vendors to solve. And it's a hard problem because every company has its own stack. They have their own custom solutions. They might be using third party. They might have cloud, they might have on prem, their own privacy, uh, you know, their own firewalls.
So it's a really hard problem to solve. And I've actually dealt with some of these vendors, you know, through Sagetap that I'm excited about.
It was just great to know that there are solutions out there, who are almost like mind reading you and trying to figure out your problem space.
Sahil Khanna: Is there any company that, that really stands out to you, that really resonates with you?
Yash Desai (Meta): Yes. the top of mind for me is Wallaroo.ai. It's a pretty unique offering that I found, which was focused on inferencing and model monitoring, which ties back to the original problem that I shared.
We know how to build models. We have solved it, but the life cycle to get them into production and monetizing it is, is much longer. And what Walleroo did is they have like a bunch of off the shelf solutions where they kind of take your model and they use their own service, they use their own tech to basically inference and also optimize on the cost, which is very critical when you get to this larger like, you know, ChatGPT style of models.
And that to me was very unique and something I was really excited about.
Sahil Khanna: For these vendors, especially earlier stage vendors that are trying to work with a company of your scale, what advice would you have to succeed?
Yash Desai: Uh, privacy, uh, and firewalls. Most of the large companies in any industry, you will see that they have a lot of privacy and regulatory issues, right? So it could be healthcare, financial services, or big tech. So you want to, maybe when you're delivering the pitch, you want to make sure that, hey, you know, we understand where you're coming from.
We understand you have your own legal processes involved, and we have the expertise to solve that. And I think that tone from the beginning just takes away a lot of, you know, I want to say that the back thinking, like, especially when I was trying to, uh, you know, make some partnerships, like I just ran into a lot of legal hurdles.
So if someone can tell me that I'll take care of that for you and you only worry about the tech part, that's, that's what I want to hear, uh, because, you know, as a tech person, I really don't want to get into the sales or the legal aspects of things. I really just want to, you know, get the solution.
