AI Insider - The Weekly Top 5 in AI and Data - Friday June 23rd

‍

‍

Welcome to AI Insider - The Weekly Top 5 in AI and Data! In this episode, we delve into the latest developments in the world of artificial intelligence and data.

‍

In our first story, we explore the findings of a new survey revealing how small business owners are increasingly turning to AI to automate tasks. We highlight the top five tasks they aim to automate, including expense management, invoicing, payroll completion, running financial health reports, and conducting customer communications. Discover how AI is transforming the way small businesses operate.

‍

Next, we bring you an exciting announcement from Databricks. They have launched Lakehouse Apps and expanded their marketplace, allowing customers to harness the power of powerful applications within their Lakehouse instance while maintaining robust security and governance features. Learn how these advancements enable organizations to leverage cutting-edge tools while ensuring data integrity.

‍

In a fascinating development, YouTube is set to introduce AI-powered dubbing in collaboration with Aloud. We delve into the details of this technology, which aims to enhance the quality of translated audio tracks by mimicking the creator's voice, incorporating better expression and lip sync. Join us as we explore the potential impact of this innovation on multilingual content creators.

‍

Our fourth story sheds light on recent revelations surrounding TikTok's data practices. The platform admitted to storing sensitive information, including tax forms and social security numbers of American creators, in China. This revelation has raised concerns about data vulnerability and potential surveillance by the Chinese government. Stay tuned as we delve into the ongoing investigations and scrutiny facing TikTok.

‍

Finally, we uncover how Harvard University is embracing generative AI as an official learning tool in its flagship coding course, Computer Science 50: Introduction to Computer Science (CS50). Starting this fall, students enrolled in CS50 will be encouraged to utilize AI for debugging code, providing design feedback, and obtaining assistance with error messages and unfamiliar lines of code. Join us as we explore the transformative potential of AI in educational settings.

Don't miss out on the latest insights and trends in AI and data! Tune in to AI Insider - The Weekly Top 5 in AI and Data for a comprehensive overview of the most compelling stories shaping the industry. Stay informed, stay ahead!

‍

Full Transcript of Episode:

00:01
Speaker 1
It?

00:03
Speaker 2
No, I just started the recording on.

00:05
Speaker 1
Okay. Are you guys ready?

00:11
Speaker 2
Let's go.

00:14
Speaker 1
Harvard University embraces generative AI in the classroom, adopting its official learning tool for its flagship coding course. Starting this fall, students enrolled in Computer Science 50 Introduction to Computer Science will be encouraged to use AI to help them debug code, give feedback on their designs, and answer individual questions about error messages and unfamiliar lines of code. What are your thoughts from the two alone?

00:44
Speaker 2
I think it's fantastic. That's just how work is going to be done now, and it's just preparing you to work in real life. Real life is not read everything to understand how this thing is working and slowly make progress by reading stuff on, stack overflow it's, get something that's working validly, like as fast as possible. And this just is going to be able to accelerate people being able to write code. Because I think especially with coding, AI takes care of the boilerplate stuff. It still's not at the point where it's going to do everything for you. So it's like a great accelerant, but it's not going to replace people's understanding of what's actually happening.

01:18
Speaker 1
Yeah.

01:19
Speaker 3
Love it. I think a similar debate happened when calculators are invented and should we do calculus with calculators? And that ended in favor of the calculator. Rightfully. I think it's good because, as Paul says, this is how you do things in the real world. And I think it's progressive to educate in that way. I also think it's just practical because I don't think they really have a choice. And trying to regulate against people using it and stop them from using it is going to be like holding back the ocean.

01:44
Speaker 1
Right.

01:44
Speaker 3
And these tools are only going to get better and better. And even now, I don't think we have the technology to really accurately tell if something, especially code, was written by Chat GPT. You hear about these occasional trials where they can identify it and stuff like that we haven't seen anything in production yet. So yeah, I think it's hard for them not to. And if you're not able to regulate it and embrace it, own it. And I think it's great that they're doing that.

02:07
Speaker 1
Yeah.

02:08
Speaker 2
And the only way to tell if in a class like CS 50, if code is written by AI is it's too good if the person just nailed it, like on the head, then it's okay. This is the first time you've ever written code. You probably didn't do it as well.

02:22
Speaker 4
How much of code sorry you go?

02:25
Speaker 3
The other important thing is it'll make people better coders faster. Because you can see what that good code looks like in context. And it's not only just a tool for building code, it's a tool for.

02:34
Speaker 2
Learning how to build code. Yeah.

02:36
Speaker 1
Because it's like default.

02:38
Speaker 2
Things it will try to do will better than someone who has never done it before trying to just guess at what they should even do. It's going to give them enough direction to be like, hey, you should generally split things out this way. That itself really helpful because a lot of times just the staring in a blank text editor is kind of like, how do I even start?

02:58
Speaker 1
Yeah.

02:59
Speaker 4
And how much of CS 50 code is generated through reddit anyways? Or like some other tool.

03:08
Speaker 2
That's right, exactly.

03:09
Speaker 4
This is just like a better way.

03:12
Speaker 1
To do that, right? Yeah. I love how you both said stack.

03:17
Speaker 2
Overflow, because I was like, that's exactly where it's coming from.

03:20
Speaker 1
And also you can do bounties and stuff to solve code on stack overflow. Like, all this crazy stuff. Sweet.

03:27
Speaker 3
Dark secret of software engineering is that everyone's afraid that people are going to realize that everyone's just copying stuff off of stack overflow, but luckily just muddying the waters and it advises a couple more years.

03:37
Speaker 1
Nice. There we go. Sweet. I can run into the next one unless ryan, you got to go.

03:42
Speaker 3
I got to go.

03:43
Speaker 1
Y'all keep on running. Sue sir.

03:45
Speaker 3
What is speaking? Soon.

03:46
Speaker 1
Just yeah. All right, are you guys ready? We got four more. Do it, Paul. You'll love this one so much. Just for you. Databricks has announced the launch of Lakehouse Apps and expanded Marketplace, allowing their customers to utilize powerful applications within their lakehouse instances while maintaining security and governance features. Yes.

04:12
Speaker 2
That's great. This is answer to Snowflake's similar offering. So Snowflake has started to offer, like, different apps that you can run on top of your Snowflake instance. And Databricks and Snowflake are basically just like, sharing punches right now. So Databricks is trying to have an offer that's similar to Snowflakes here, taking advantage of some of the things that Databricks is really good, just running gigantic spark clusters and giving applications the ability to build on top of that. So it's super cool. I think there's going to be a lot of nice niche businesses that are built here because you don't have to manage a lot of stuff. You can just build a lap that basically runs on either Databricks or Snowflake and do something vertical specific that's pretty small and still have a nice income off of it.

04:57
Speaker 1
Does this bring Data to bricks so much closer to Snowflake? Or is the gap between the two.

05:03
Speaker 2
Still very large and they still need.

05:05
Speaker 1
To do a lot of work?

05:07
Speaker 2
I think the gap is still sizable, and the reason for that is that this is playing to Databricks'strength, which is their lake house, kind of. You're basically running on top of this big spark thing. Snowflake strength is not that Flake strength is the SQL based warehouse. Just that is what they do super well. So in my mind, they still have a significant difference. Databricks just crushes that spark interface. SQL interface isn't quite as clean and mature as Snowflakes. Snowflake, on the other hand crushes the SQL interface, but their distributed arbitrary code interface is not anywhere close to Databricks right now. So how do companies have a while to do before they're really equal on those levels?

05:46
Speaker 4
How does their customer profile differ? Who uses databricks.

05:51
Speaker 2
Yeah, Databricks is a lot of, like, machine learning people. So, for instance, if you work at Call of Duty running machine learning models to change things in the environment or something, you're using Databricks because you need to basically process a huge amount of data in relatively real time to train these models, and you need a huge amount of compute for that. So Databricks is great for you. You're writing like arbitrary Python code to train models, run validation, all of this kind of stuff. Databricks is amazing at that. Snowflake, you want to do Bi, you want to do reporting. You really want snowflake? Snowflake is the best in the business.

06:23
Speaker 1
At It, databricks is.

06:25
Speaker 2
So that's the main distinction. Snowflakes analytics. Databricks is machine learning.

06:32
Speaker 1
If we could get Call Duty as.

06:34
Speaker 2
A client, which I think is owned.

06:35
Speaker 1
By Activision or Blizzard, one of the two, I would be the happiest.

06:40
Speaker 2
I know the person who runs Data Science for them.

06:42
Speaker 1
She lives like five minutes away from me. It sounds like. This isn't the work. I would love this.

06:49
Speaker 2
I'll need to hang out with her a few more times and only hang out with her a couple of times. Go.

06:53
Speaker 1
All right. Moving on to number three. According to a new survey, small business owners are looking to automate certain tasks using AI. The top five tasks they want to automate are expense management invoicing completing payroll, running financial, health, and conducting consumer communications. What are your guys'thoughts do you automate all of it?

07:16
Speaker 4
Rippling is so well positioned.

07:20
Speaker 2
Rippling is going to eat it. They're so good. Their whole model is incredible. We use them right now mostly for payroll stuff, but at some point we'll need device management. We can just click plus in Rippling and pay them like a few just pay them a little bit more and everything just works. They'll be able to slide in for we'll use them for expense management, not somebody else, because it's just you click a button and everything already works. Their model is amazing. So they're a juggernaut of a company, and I think they will be the ones that capitalize the most on this.

07:49
Speaker 1
Okay, definitely big fan of them, too. Yeah, they make everything easy. So you think that just one company is going to be the clear winner for this and Rippling is probably going to be It.

08:00
Speaker 2
No, not just one company, other HR, but it's going to be, I think, largely existing providers because I think for something like expense management and stuff like that, expensify can add this feature pretty well in their model without it being like a huge thing. So I think the people who win in this scenario are largely existing providers because I don't think this is like a fundamental architecture change. It's like something that just accelerates the existing products. So you have a little magic button next to approve expense in the same interface that you've got in Rippling or Expensify or whatever tool you're using. So I think it increases productivity. But I think it's something that probably accrues to incumbents, whereas something like bi or maybe a more complicated I haven't thought about this a lot, but maybe something in ERP or something like that more complicated and could represent like a genuine paradigm shift, those ones.

08:53
Speaker 2
A new tool might win because it's just so hard to rearchite the old tool. So I think my heuristic for that is does this fit in as a neat little new feature in an existing product? And if that's the case, then the incumbents are winning for sure or does this completely change, like what's possible? And if it completely changes what's possible, then a new company has a real shot of winning.

09:15
Speaker 4
Yeah, and I think because Riplink can just own that entire process for you and they have the brand behind them that they've done this before and they're doing it with thousands and thousands of companies, that's attractive. If you can have someone just own this entire process and you don't have touch it, you can trust them.

09:31
Speaker 1
Yes. Tough to beat. That nice. Sweet. Let's move on to number four. YouTube plans to introduce AI Power dubbing with the help of the team from Aloud, aiming to improve the quality of translated audio tracks to resemble the creator's voice with better expression and lip sync. Absolutely wild. What do you guys think about it?

09:53
Speaker 4
Walking me through that? What is it doing exactly?

09:57
Speaker 1
YouTube. Paul, do you understand it?

10:00
Speaker 2
I think so. So my understanding was basically that YouTube is going to instead of just doing their normal transcription that they do right now, they're gonna have a dramatically improved transcription service that will be able to way more accurately get what's being said, who's saying what in the conversation to make the transcript just like, way more.

10:18
Speaker 1
High fidelity, which makes total sense.

10:22
Speaker 2
As an aside, I would love to see how much they're spending on that. Just a cost to run that model over the sheer quantity of content that gets uploaded to YouTube is like mind bending, like a billion years of content every I don't even know how often, but just unbelievable how content gets uploaded to YouTube.

10:41
Speaker 1
So I would love to see how.

10:42
Speaker 2
Much they're paying to run the service. Oh, yeah.

10:45
Speaker 1
And I think one of the things that stood out the most to me is the fact that they're trying to dub very similar to the original creator's voice for the new dub, which is so insane. So they would be able to dub over this conversation right now with a Spanish dub, but our Spanish dub would sound very similar. Yeah. So that's a pretty wild thing to get into. Drew, what are your thoughts? Do you want to get dubbed over?

11:13
Speaker 4
All your sales calls? Dubbed over? So since we've been experimenting with fireflies and having that transcription at the end of each of our calls, that's been a big game changer. I think being able to be incredibly focused on the actual conversation that we're having and know that every single note is going to be taken by an AI bot that will send me it, and then I can write, like, a little script that turns that into a really compelling post demo follow up.

11:38
Speaker 2
I don't know.

11:39
Speaker 4
Yeah, I'm very much in favor of all the transcription technologies and how they're influencing my day to day in terms of how it will influence YouTube. I don't know. I don't have as much of an opinion.

11:53
Speaker 1
Got it. I do know that Mr. Beast was the original person to do this with all of his channels, and then he dubbed them out based off of the analysis that most of the world doesn't speak and listen in English. So why wouldn't you serve your video in another language when most of the world can't even understand it? So he's just opened up a whole market for himself by dubbing in different stuff. He did this two years ago.

12:24
Speaker 4
And there's also some interesting multiples that happen. And I don't know how the revenue model works with YouTube and YouTube creators, but I know that the more channels you have with some minimum threshold audience or subscriber base, you get paid significantly more. And I think if you have 250,000 subscribers one channel and again, I don't know how this works, but I think the payout for having 151 hundred on two separate channels is higher than having 250 in one channel. I don't know if that's the case, but it seems like every creator that I've watched has, like, the main channel right, that everyone goes to, and then.

13:03
Speaker 1
They have I don't know.

13:04
Speaker 4
I'll take Mr. Savage, he's a Fortnite player, as an example, but he'll have his main channel. Then he'll have Mr. Savage Raw, which is like, every tournament he plays in just like, the entire clip of the tournament. And then he'll have a Mr. Savage shorts, and those are all, like, just ten second clips on big Kills or I don't know, whatever happens in these tournaments.

13:24
Speaker 1
But yeah, it's cool. Pretty wild. Pretty wild. All right, we are going to the last one. We're going to get a little spicy with this last one. I wanted to give you guys a spicy sandwich. This last one's going to get a little controversial, a little spicy, but I would love your take on the data storage of it all. So TikTok admitted to storing American creator sensitive information in China, including tax forms and Social Security numbers, which has raised concerns about data vulnerability and potential Chinese government surveillance. Senate leaders are calling for investigations and continued scrutiny of TikTok's data practices. What are your thoughts?

14:11
Speaker 2
So I think one thing here is that it might not actually be nefarious. Like, people think it's really easy to, like, just have data get replicated into some data center from some backup. That happens without other context. It's really hard to say if, like, any bad behaviors actually happened here. I think one thing that it's easy to gloss over, especially like when Facebook's had issues with data, when TikTok kernel, these other companies have issues where data shows up somewhere it shouldn't. It is actually really easy to have stuff get replicated, get copied, get put somewhere in some region or some instance that it wasn't originally intended to be. So it takes a lot of work to keep things clean, especially at a company that's growing at the rate that TikTok is. I would definitely say jury still out. I would not read too much into it because from the technical side, it is really easy to have stuff just get copied over.

15:03
Speaker 1
All right, Paul defending TikTok.

15:10
Speaker 4
I'd love to be able to take a trip to Beijing. So with that being said, I'm going to go in Paul's camp and just go plus one there.

‍