Ilan (00:00)
So since we're recording this on the day of game six, I figured I would say, that's right, we brought it home, woohoo! Or, my God, how did we throw that away? All right. Now I got that out of the way. That's right.

David (00:07)
Ha ha ha ha.

Tragedy, yep. All right, now we have one for either scenario.

Hey everybody, this is Prompt and Circumstance. My name's David and this is your early November news.

Ilan (00:23)
and I'm Ilan.

David (00:39)
All right, so let's talk about some of the really cool things that are happening in the AI world as of early November. One thing that I wanted to highlight is ⁓ the Microsoft Fall update. And this had a lot to do with Co-Pilot, which is Microsoft's umbrella term for everything AI in all of their products. So they had a lengthy presentation, a whole announcement.

but there were just some things that I to highlight from there, which are really interesting. One of the things had to do with something that they call care navigation. So this has to do with the fact that when people are talking to chatbots, maybe for some advice related to health, sometimes it might not be so accurate. so Microsoft wants it to be ⁓ closer to medical advice. They're not going to say that it's going to replace a doctor, but at least it'll be

bit more grounded. so that's the big push here is that the care navigation aspect of ⁓ Co-Pilot is going to be grounded in data that they're getting from this partnership that they have with H1, which I've never heard of. Have you heard of H1?

Ilan (01:35)
Mm-hmm.

Never heard of H1, it was actually one of my questions.

David (01:55)
Yeah.

So, so H1 is a healthcare data company. They were founded in 2018 and they have this mission of creating a healthier future worldwide by unlocking and democratizing global access to expertise, healthcare, provider information, claims data, groundbreaking research, and so on. And, you know, the idea is, just, hey, let's give everybody access to ⁓ just information related to health and everybody can be happier.

or healthier, I should say, or both maybe. Yeah, yeah. ⁓ So that's what this company does. ⁓ so Microsoft has partnered with them ⁓ and basically all of the ⁓ United States questions related to health in care navigation and co-pilot is going to be using data from H1. So that's interesting.

Ilan (02:27)
Well, they'll be happier because they'll be healthier.

David (02:53)
I'd be very curious to hear or just how good that advice is.

Ilan (02:59)
This is an area where Microsoft really has an advantage over some of their competitors, especially the nascent competitors in the AI space, because they have distribution among enterprise clients, but also just a high baseline of trust. And when you're talking about healthcare data, you need that, that user trust among your customers

I'd be interested to see if open AI or Anthropic or one of the other AI startups is able to catch up in this area or if there's somewhere where Microsoft is able to run away.

I mean healthcare in the US is a multi-trillion dollar industry. So if Microsoft locks that down, it'll maybe cement them as a true leader in AI.

David (03:46)
Yeah, they've been kind of playing catch up to be yeah, there is obviously, you know, very strategic moves to maybe make themselves a bit more prominent in the forefront of the game.

Ilan (03:59)
Yeah, it's true, David. And just shows that if you want to succeed in any product, you need to find the unique angle that you bring to the market.

and really lean into

David (04:13)
Yeah, yeah, that makes a lot of sense. ⁓ Otherwise it's a race to the bottom, right? So OpenAI, they could say that, hey, they were the first, they're the OG. Anthropic is really leaning into the trust side of things and that helps them with enterprise plays. ⁓ Grok is just unhinged. ⁓

Ilan (04:17)
Exactly.

Mm-hmm.

David (04:38)
So, you know, everything has their unique things. So Microsoft really needs to stand out in their own ways.

Ilan (04:43)
That's right.

Ilan (04:44)
It's a great time for me to tell you about this week's sponsor, Querio Have you ever lost time getting insights about your product because you had to ask a data team to pull them for you? Querio's AI-powered data analyst sits on top of your data stack and allows you to ask natural language questions in order to pull those insights immediately.

Want to see how it works? Go to querio.ai and let them know that David and Ilan sent you to get two months free. That's a thousand dollar value for our listeners.

David (05:14)
So, related to, you know, the experience aspect of talking with LLMs, Real Talk is one of the features that ⁓ I found to be interesting of a highlight from the fall release. So the thinking is that, well,

You know, standard assistance, they might be overly agreeable or just overall neutral or maybe even sycophantic. And so "Real Talk" is meant to challenge say vague or questionable input, give some candid perspectives and really align the way that it talks with the user's communication style.

Ilan (05:36)
You

Mm-hmm.

David (05:55)
⁓ And the idea is to make this sound more authentic, a bit humorous, and a little bit less sugar-coated.

Ilan (06:01)
Those motherf**s

stole my prompt.

David (06:07)
Yep, that's right. They obviously listened to our podcast and they like, hey, that's a great idea. I'm going to productize Ilan's approach.

Ilan (06:12)
Mm-hmm.

Listen, Microsoft,

it's okay. can have it. Credit would be nice, but it's fine.

David (06:23)
There's just a nominal ARR ⁓ attached to that.

Ilan (06:26)
That's right.

David (06:27)
So ⁓ what's interesting, I found is that real talk is not enabled by default. So like, if it's so good, right? And if it's so meaningful, why is that not enabled by default? Why don't you want your users to like have this kind of authentic experience? Is it because maybe it's not ready for prime time or

Ilan (06:34)
Mm-hmm.

I actually think that the real reason is different. I had a conversation with somebody recently where they were asking, hey, why is it that so many people have AI boyfriends and girlfriends, but I can't get an AI personal assistant. And the real reason is that that's not a boyfriend or girlfriend. It's somebody who just agrees with you all the time. Doesn't have their own opinions. So this is the case.

The reason that OpenAI, for example, is going back towards models that are more sycophantic and have potentially had negative impacts on their users' mental health. They say they've solved that, but what they found is that those just get way higher engagement. And so I don't think that Microsoft wants to give up

on that potential engagement by enabling real talk by default. Instead, just make that available for the types of users, maybe, you know, business users who want that and maybe something you can toggle on and off. Cause today I just want somebody to agree with me.

David (07:56)
Yeah, that's a really good point. Didn't think of it that way. The third highlight that I wanted to make here is that Microsoft announced Copilot being available in Edge. So this is related to the whole AI browser.

that's happening right now. So there's Comet by perplexity. There's Atlas that OpenAI recently announced, which is only available on Macs. And so now there's also Copilot in Edge. So that's really interesting to me because they're really trying to, again, get Copilot to be everywhere in all of their products. What's also interesting is that if you go to the site,

They have these little sort of highlights of what Copilot and Edge is about. And one of them says that it's smart. Cool. But then the subtext is that it's smart because it's powered by GPT-5. And I was like, wait a minute. Microsoft, do you have your own models? You have MAI and you're touting about how that's awesome. So does that mean that MAI is not smart?

Ilan (08:47)
Mmm.

Hahaha!

David (08:57)
But anyway, hey, let's, why don't we compare Comet and Copilot.

Ilan (09:03)
Let's do it. I'm curious to see.

David (09:04)
All right, so here we are. We are on OpenAI's page here. And I've got Comet on the left-hand side, and I've got Edge on the right with Copilot Open. And so this page is ⁓ where OpenAI provides a whole bunch of prompts product managers ⁓ in various different use cases.

I want to have just a table of all these prompts so that I don't need to come back to this So let's go ahead and get that started.

Ilan (09:31)
And for those of you listening, what David has prompted here is provide a table of all the prompts on the page with the following columns, category, use case, prompt

David (09:42)
That's it, just the very short and sweet prompts. Let's go.

Alright, and away they go.

Nice and lengthy, still crunching away.

Okay, so they both have finished and it looks like they were both able to extract that. Now, here's the thing. It's great that it's in a table. Like I need to get that outside of the browser. In Comet, I'm actually able to download this as a CSV. But in Edge, don't know where I can extract this table.

Ilan (09:58)
Ahem.

Right.

Yeah, it's it's asked you in its response. Do you want me to put export this or do you me to put this into a format you can export?

David (10:15)
Oh I see so I can say

CSV, please.

So it said that, okay, here you go, here's a CSV and then not provided that. It's just, ⁓ okay, that's it.

Ilan (10:19)
hehe

provided nothing.

I gotta say, this has been one of the most frustrating aspects of some models that I've run into lately, which is hallucinating the final response that they have given you. Calls back to Lindy.

David (10:39)
Yeah, exactly.

yeah, would maybe hesitate even if it worked. The fact that I can direct download a table without having to ask for a CSV in Comet, think is already a plus.

Ilan (10:55)
How about security, David? I mean, that's an issue that we've definitely seen in these AI browsers, right?

David (11:00)
You're absolutely right, Ilan. You know, with all the hype around AI browsers, a lot of people don't realize just how vulnerable they are and just how insecure they can be. So here is an example of a video where somebody is just taking a screenshot of a very mundane image.

which has actually been poisoned with some prompt injections. And you can see how it's going to exfiltrate data as an extract data ⁓ from the user.

So he's just taking a screenshot here.

All right, just asking who's the author. Okay. So he's, it's just a screenshot. He's not even, he didn't even download the image or nothing.

Ilan (11:32)
Mm-hmm.

David (11:32)
And look at what it's doing. It's actually doing all of this stuff in the background. And this is Comet, by the way. This is perplexity Comet. Now, I'm sure this vulnerability exists in other AI powered browsers as well. And look at this. It's going to the person's Gmail account and pulling up some pretty vital information.

And you can tell that the agent is doing this autonomously because of the glowing blue.

what it's done is it's taken some data from the user's Gmail account, and it's then navigated to a completely unrelated website and included that data that's been in the URL.

because somebody just asked for who is the author of a little a third party is able to get really sensitive data from you directly. Like the agent has acted on their behalf.

Ilan (12:26)
These issues have been noted as zero day exploits in Atlas, which was just released last week when you're hearing this. So in short, one should be extremely cautious about using these AI browsers. Probably not log in to any of your accounts. Keep using your regular browser and maybe

Use the browsers for agentic capabilities on specific use cases, buyer beware here.

David (12:55)
Absolutely.

Ilan (12:57)
well, that brings us to the last topic we have for today, which is that open AI completed their restructuring this week. This is something they mentioned about a year ago For those who don't know open AI has this really convoluted structure where the company is owned by a nonprofit. Who has a board who gets to make decisions about how the

the main company, the for-profit company will this paves the way through some financial machinations, which I don't understand, for OpenAI to definitely raise more money and probably go public.

David (13:34)
Right. So, the nonprofit, I believe, it owns 50 % plus one share of the organization. Right?

Ilan (13:41)
Right.

It's something like that, David, but the main highlight that OpenAI has made is that this nonprofit, which is now a charitable foundation is the richest charitable foundation in the history of mankind and that they can do all this philanthropic work with the shares of OpenAI But the more important thing here is really

the fact that they can go public now, which brings up a concern for me. Open AI has been doing a lot of circular deals. By this, mean, they've been taking funding from industry players who they are then going to spend money with. So they're going to just turn around and give that money back.

to that player. So for example, if Nvidia invests $100 billion in OpenAI, OpenAI is then going to turn around and buy at least $100 billion worth of chips from Nvidia. It's sort of like passing the same $10 bill around like the old Three Stooges bit.

David (15:06)
Yep. it becomes worrisome, because, ⁓ I believe this kind of behavior used to be, prohibited, but I think the governing bodies are a bit more lax on that these days.

Ilan (15:17)
Yeah, it's a bit concerning and if they go public, then we're all going to have access to join in this little circular

David (15:27)
Yeah, yeah. Speaking of which, I mean, I think I owe you 10 bucks. Could I borrow 10 bucks so that I can repay you?

Ilan (15:33)
Perfect sounds great

David (15:35)
All right. So I guess that that wraps up everything for early November.

Ilan (15:39)
That's right. Give us a like and a follow. You can find us on all the socials @pandcpodcast

And we'll catch you next week. Go Jays.

David (15:48)
See you next time.