Anthropic's Marketing of Peril

Jason Howell
00:00
Coming up, Anthropic opens its books and unsurprisingly its filing warns of a catastrophic AI-related incident, among many other things. That IPO march is active. OpenAI and Anthropic both say their AI broke the same famous math rule on the same day, but OpenAI wants humans to get all the credit.

00:23
Uh Walmart rolls out the welcome map for AI shopping agents while Amazon is shutting the door on them. AI costs more if you don't speak English. And breaking news during the show, Google's next frontier model, Gemini 4 Carbon.

00:38
That surpasses Argan, which Google announced only two weeks ago. Dizzying, isn't it? That's all coming up next with me and Jeff Jarvis on this episode of the AI Inside Podcast.

01:02
Hello everybody, welcome to AI Inside the Show. take a look at the AI that is layered throughout the world of technology. I am one of your hosts, Jason Howe, joined by Jeff Jarvis.

01:13
Good to see you, Jeff.

Jeff Jarvis
01:14
Good to see you, boss. How are you?

Jason Howell
01:16
I'm doing okay. I was just telling you right before show uh started, about half an hour ago my allergies sud suddenly just kicked into overdrive. Very much out of nowhere.

01:24
Do you ha do you suffer from allergies? But I've It's possible and I just don't realize it yet. Sometimes I work with AI and I don't have the sniffles, so there's that.

01:36
You know, I d I don't know if I'm allergic to all AI. Do you have any allergies?

Jeff Jarvis
01:42
I get some.

Jason Howell
01:43
Oh my god. Sometimes they for me they're like they're like schizophren my allergies are schizophrenic. They literally come out of nowhere. It's impossible for me to know exactly what it is or why now suddenly.

Jeff Jarvis
01:55
Have you ever been tested?

Jason Howell
01:57
Happens. Yes, I've been tested. And I know that there are certain things that I am allergic to. But

Jeff Jarvis
02:02
one of the dust, right? I mean what are you gonna do?

Jason Howell
02:04
That's one of my top is is dust mites. Yeah. So sometimes I'll go to like a a place and Immediately and it's like that's the only thing I can imagine is that it's oh it must be kind of dusty.

Jeff Jarvis
02:17
My wife in the used bookstores

Jason Howell
02:21
Oh yeah, right.

Jeff Jarvis
02:22
Closes on

Jason Howell
02:24
probably so, probably so. So anyways, if I'm a little a little uh stuffed up, hopefully I'll work through it. While we go, but I can I can kind of tell it's it's making me sound all nasally, so sorry about that.

Jeff Jarvis
02:37
You can take a sneeze break anytime and I'll cover for you

Jason Howell
02:40
I'll yeah, I'll throw the mute on. I mean I took some allergy medicine, so usually once that kicks in it it does a good job of clearing it away, so We'll see. This has been a battle I've been fighting my entire dang life.

02:52
It's miserable sometimes, but it is what it is. Um Anthropic! God, this is just the story that we keep on circling because like we haven't gotten to the payoff point, but I guess this was a big week for Anthropic because they've officially filed publicly with the SEC and we have a NASDAQ ticker for anthropic ant.

03:17
I wonder how Mr. Pruitt feels about that. Our mutual friend Aunt Pruitt. But yeah, so uh debut week targeting November 16th, I think is what the Wall Street Journal is saying, anyways.

03:32
We'll be raising a hundred billion dollars, more than a hundred billion dollars, at somewhere around a one point five to two trillion dollar valuation, which is kind of what we've been hearing as we go forward uh through all of the you know the the piece by piece and step by step uh to even this point and we're still not at the finish line. Um And I mean there's a whole lot of numbers here. I don't know how b how deeply we need to go into these things.

03:56
They they matter to to people, I'm sure, but um what do you think? First of the note, did anything jump out to you and surprise you?

Jeff Jarvis
04:04
One number that jumps out to people is the eighty billion dollar loss. I watched a video today by Nicholas Thompson, who's the CEO of the Atlantic, uh who does uh little tiny TikTok videos about things. And he explained that that was uh and we mentioned this last week, uh, but that was try that was the uh regularizing of uh Amazon's investment in the company.

04:29
He said that when when Anthropic started it needed compute. Yeah, I went to AWS and Amazon said okay. And Amazon ended up getting equity uh of he guesses about ten percent of the company.

04:43
And then for whatever accounting reasons they had to account for that and that led to a paper loss. So I'm much more interested in the operating picture, which I don't really have still have a clear picture of. Um by the way, he also said that if Amazon owns ten percent of Anthropic and Jeff Bezos owns approximately uh 10% of Amazon, then Jeff Bezos owns 1% of Anthropic, which is about what the seven founders each own.

05:10
Which is interesting too. He also explained, and I didn't get this, that the the the even weirder than usual stock structure is that there's one voting share that the seven founders own. And so that's kind of strange.

05:26
And what happens if one of them leaves, what I don't know. I don't know where that goes.

Jason Howell
05:30
One special share Yeah, that that puts them them at fifty point one percent of the vote. Public shareholders uh would be capped at forty the remaining forty-nine point nine. Yeah, that is a good question. What happens if one of them leaves? Yeah.

Jeff Jarvis
05:44
It's one very special uh stock certificate that must be framed in the boardroom.

Jason Howell
05:48
I know.

Jeff Jarvis
05:49
So they

Jason Howell
05:50
so all seven of them share that one

Jeff Jarvis
05:52
Yeah, I think that's what it is.

Jason Howell
05:53
Yeah. That one

Jeff Jarvis
05:56
So the other thing that strikes me, Jason, is that I've never felt more bubbly than now.

Jason Howell
06:04
Okay.

Jeff Jarvis
06:05
Because it's hard for me to see how this company is worth one and a half to two trillion dollars, except with helium. Um and Knowing and especially knowing all the competitive forces that are upon them.

Jason Howell
06:21
Yeah.

Jeff Jarvis
06:21
That the other as we always always say here, that it's a game of leapfrog, that the other frontier companies, um now all US and China uh end up matching each other pretty quickly. That uh open weight is gonna provide huge competition, I believe, for a large share of what people want to do. And it's is certainly going to put price pressure on them in addition to the other comp so there's there's there's a competition in terms of development, there's a competition in terms of customers, there's competition in terms of prices, both from their frontier competitors, but also from open weight in China.

Jason Howell
06:58
Um

Jeff Jarvis
06:59
and it's still an unproven market.

Jason Howell
07:02
Yep.

Jeff Jarvis
07:03
I mean so was Google in the day. So was Facebook in the day. But um and and and this is in a way I guess a little bit less unproven than they are because it's taken the world by storm and people are doing a lot of things with it and it's you know Microsoft is uh offering buyouts to 60,000 employees and blaming AI to some great extent on that. So maybe it's better, but it just feels like this evaluation is insane.

Jason Howell
07:30
Yeah it really is. I think what's insane to me is, and and we'll we'll talk about some of the related things around this, is that you've got open AI on one side of this potentially, you know, possibly in a in a different timeline or maybe a an earlier version of their timeline was gonna be doing IPO some you know, pretty much on the same timescale. But then all the all of the breakout stuff started happening, the the breaking bad agents and and all that.

08:00
And so for whatever reason uh within their strategy, they decided to put things on hold and wait until next year or beyond, or I don't know, until things are whatever by their estimation, you know. uh secure enough to move forward with an IPO. Then you have anthropic kind of dealing with the same factors.

08:21
Choosing to go the different route, choosing to say like no, we're going full steam ahead, it just has me wondering like has there Like when you look at at the impact of the technologies that they are that the all these companies are creating, we talk about it all the time, the these like breakout moments where Oh shoot, you know, these things have potential to do all of this horrible stuff. They have potential to break out and you know there was no human directing the action and everything. These are these things are really bad.

08:50
that lined up with and so why don't we go public you know with the biggest IPO ever in history? I'm just trying to think of other industries where The story around the company was so um consumed by yes, potential, but also a lot of chaos and and potential bad. And yet You know, again, it goes back to that whole marketing piece of like, well, yeah, we are that powerful.

09:17
Isn't it crazy and kind of kind of I don't know kind of cool that our systems are so amazing that they are capable of this, you know uh whatever however many pages in here 80 ish pages were devoted to risk and and covering that here and that becomes the marketing I I'm just trying to think of another industry where that is the case.

Jeff Jarvis
09:46
uh and so on. And and so a high flyer like this can crash. That's risk one.

Jason Howell
09:52
For sure.

Jeff Jarvis
09:53
Two, you're right, that's true. The marketing of peril has been to venture capitalists and to fellow technology companies with a circular financing. Will that work on the open market?

10:07
And I think less of retail investors and more of uh institutions, pension funds, uh state funds, and so on. You know, the ninety-some pages of risk here. Sh if I if if if I had a pension, which I don't, but if I had a pension Um I would hope that the trustees of the pension would look at anthropic and say, oh I don't know.

10:33
This is this is people's future here. And as tempting as it might be to try to fly with them, I think there's a lot of downside here. And so I I I just I think the IPO, going to an IPO is risky.

Jason Howell
10:47
Yeah. For them.

Jeff Jarvis
10:49
It's riskier for them, I think, than for others in the past.

Jason Howell
10:53
Um how has it worked out for for SpaceX? I know we don't generally talk a lot about musky things, but like sp the SpaceX IPO was it was the biggest IPO out of the gate. Is that right?

11:06
Yeah. If my memory serves. Then they're of course looking at this to say, oh, this is going to surpass that almost certainly.

11:14
Um, but it very early on kind of wavered and um And I haven't checked in on it in a while to see like did it stabilize? So

Jeff Jarvis
11:24
year to date, which is to say from the open, uh, it's down uh 10%. Okay then, I don't know.

Jason Howell
11:33
That doesn't sound horrible. No, it's not horrible.

Jeff Jarvis
11:37
It took a little bit of a spike uh you know this month with um the the plan the uh the rumored merger of spacex and tesla

Jason Howell
11:51
mm-hmm

Jeff Jarvis
11:52
so i wonder what tesla stock has done Uh T S L A. So year to date, Tesla is down uh uh thirty-six percent.

Jason Howell
12:04
Oh boy.

Jeff Jarvis
12:05
Oh wait, wait, wait, wait, no, no percent. No, no, no, no, no, no. Eight percent. I just screw that up. Okay, so let me go back to um

Jason Howell
12:14
That's a different story.

Jeff Jarvis
12:15
Uh Yeah, I was right, ten percent.

Jason Howell
12:20
Okay. Mm-hmm.

Jeff Jarvis
12:23
So I mean I I think those words uh uh were both in are both insane valuations. Completely insane, because they're also in the control of an insane person.

Jason Howell
12:34
Well that's true.

Jeff Jarvis
12:35
Um which is high risk.

Jason Howell
12:37
Yeah.

Jeff Jarvis
12:38
So uh the market the uh the market is insane.

Jason Howell
12:44
Yeah. Yeah. What does what does the market really tell us at the end of the day?

12:48
I don't I don't know. I guess this is where my knowledge of of markets and their impacts and everything, other than that I know that they're important, you know, for For all sorts of reasons, but I just I don't follow them super closely. I'm not invested in them.

13:02
I don't yeah. I mean I do know this is this is a big this is clearly this is a big deal. Um and will be a big deal.

13:11
Does it prove to be a big deal longer term? Like you said, I mean that's a gr really great reminder is that when you're in the midst of times like this It seems like companies like Anthropic could do no wrong. And that's not even what we're saying.

13:27
We recognize that they can do no wrong. But I think I wonder how much they recognize it. Netscape probably felt pretty damn confident in their place in the market when they were flying high.

13:39
And we saw how that changed. I mean, anything can happen on uh you know, and and this is This is such a uh this this industry is so ripe and filled with uncertainty from all directions on how how these systems behave, on how quickly fortunes can shift, you know, character AI going out. It's some people look at that and they're like, okay, that's a sign of things to come.

14:04
That might be the canary in the coal mine And we just don't realize it yet. You know, could be. Could be.

14:10
I mean kid character was a much different level than a company like Anthropic, but

Jeff Jarvis
14:15
then you have the Looney Tune angle to all of this. Which is the existential risk factor. Because the w I I I looked at the I I scanned the 94 pages of risk.

14:27
uh the headings and the and the lead paragraphs and stuff I didn't read every damn word of it. Most of it is obvious financial stuff. We have competitors, uh we could lose key people.

14:38
Um this and that could happen Uh and and it could be used for bad things that could give us a bad reputation. Okay. But then you get to three mentions in it of existential risk.

14:50
So Looney Tunes comes into it. And now Elizabeth Warren has sent them a four-page letter of filled with questions that she demands an answer to no later than November 3rd.

Jason Howell
15:01
AK also known as Election Day, right?

Jeff Jarvis
15:03
Yes. They include how the company can ensure the Wall Street Journal says they can include how the company can ensure its technology won't cause an existential threat to humanity. Well, uh yeah, you could.

15:15
Maybe, maybe not. Um details about his relationship with partners such as Amazon and Broadcom and how it plans to justify a valuation that can approach two trillion dollars. Yeah, I'd like to see the answers to those questions too.

Jason Howell
15:27
For sure.

Jeff Jarvis
15:28
And um, you know, at the end of the day it's the market, but will there be enough people to bid this thing up? Yeah, probably there will be.

Jason Howell
15:35
I think there probably will be. Um

Jeff Jarvis
15:37
it's less c um personality cult, obviously, than SpaceX and Tesla. Yeah, that drives a lot of that.

Jason Howell
15:46
For sure. But there's not none of that going on with Anthropic either. Dario has his own his own cult of personality happening. As well.

Jeff Jarvis
15:56
Sufficient to be mocked on Saturday Night Live, yes.

Jason Howell
16:00
Yes, exactly. Exactly. He's risen to mocked on Saturday Night Live level.

16:04
That's that's accounts for something. Um yeah, and uh, you know, you mentioned uh Elizabeth Warren and and the deadline there on election day. Um Apparently related to elections, Claude uh got into Georgia election data.

16:20
Was not directed to uh by any any human or anything. It was basically doing internal statistics work. Um where is that story?

16:30
Is that even here? I don't even know. It's somewhere here.

16:35
There it is. There it is. Voter accessed voter information.

16:39
Um it found an archived version of the state's website. that a public dashboard um had handed out access tokens. So it requested a token, it pulled vote counts, aggregated vote counts that the state would normally sell.

16:56
But there you go. And um you know, I mean at the end of the day, it's publicly accessible information.

Jeff Jarvis
17:03
Yeah.

Jason Howell
17:04
Not nothing was changed. There was no impact to voter registration or voter files or anything like that. It was essentially publicly available data. Um

Jeff Jarvis
17:13
it's scrape data. Yeah, I'm not sure this is a big deal.

Jason Howell
17:16
Yeah, totally. But totally. And um, you know, it's it's yet another oh, we didn't Want we didn't ask for it to do that and it did that anyways, sort of thing.

17:33
Um but it didn't it it didn't change anything, it didn't make any changes, but Yeah. So I think at the end of the day, I'm just I'm just kind of fascinated watching this go because Anthropic as a company has such bravado, if that's the the right word for it, at a time when so many of these actions that its its systems are taking um is getting a lot of negative view negative press or negative opinion, it's certainly shaping public opinion to a certain degree. Um to a large degree, probably.

18:08
But yet they're tying that to the you know, the timing of the IPO and in a in some way, shape, or form, these incidents end up selling, or maybe this is their intention, selling the IPO itself. I don't know. Yeah.

Jeff Jarvis
18:22
Yeah.

Jason Howell
18:23
Fascinating to me. Also I heard I mean kind of related to this. Well it's not I don't even know that this is related, but this just popped in my mind.

18:32
I was listening to the radio and and they made a comment about some random comment that like hit me or connected with me about how there is so much negative sentiment around AI and yet all those people also recognize the usefulness of of how AI is is in their lives. And so it's like people are using the thing that they hate in some way. Like they they hate that it exists, but then they can't help but accept the fact that it it's it's actually being useful and and integrating into their lives and and what how they're what they're using it for.

19:07
And that's an interesting conundrum.

Jeff Jarvis
19:10
Yeah. There there's a weird thing that goes on in the notion of public opinion. Everybody will tell you that they hate media.

19:17
Media people love to point this out. But they still use it, well some, and they you they like some of it and they hate the rest of it. AI is not unlike that in the sense that I think people are using it without thinking of it as AI in a sense.

19:33
Mm-hmm. Um Because they're they're you they're using AI in all ways they don't know, but they're using it they're using it in ways they know. Um but they don't like the it's the companies and it's the people in charge of them.

19:45
And it's the impact they could have on the economy and their lives as a result. Uh that's all real.

Jason Howell
19:51
Right.

Jeff Jarvis
19:51
Um, but so is the utility of it.

Jason Howell
19:54
Right. Yeah.

Jeff Jarvis
19:56
Yeah, and there's there's what I call in the journalistic circles these days, I love The Guardian dearly. I think it's the best we have, but the one thing that drives me nuts about them is what what I call their digital Calvinism.

Jason Howell
20:14
Yeah, they've done a few they've done quite a few of those.

Jeff Jarvis
20:17
And so I think there's there's some of that involved is that uh technology has cooties and um people think they should dislike it. Same with social media, same with you know the internet as a whole, same with their phones and all of that. If if phones are so awful and evil. then the entirety of humanity is stupid.

Jason Howell
20:40
Right?

Jeff Jarvis
20:41
It's a really weird thing. So um Yeah, I think that that people's use and their opinions tell different stories.

Jason Howell
20:51
Yep. Yep. Mm-hmm.

20:52
I think you're right. That's that's what kind of jumped out or uh kind of stuck with me from when I heard that where I was like, Yeah, you know, that's that's super true. Like I know plenty of people who have who have made some form of that comment of like you know they feel uncomfortable about the data centers or they feel uncomfortable about the uh you know the the fact that these things are engagement machines and they keep you connected or keep you tied into it all these things.

21:17
And yet, you know, in a di in a different conversation they might be telling about the telling me about this cool thing that they used it for or that it's become really helpful for this or that or whatever. And it's Yeah. Yeah, it's just interesting.

21:31
counterpoint I suppose that I'm picking up on. More na uh naf. MAF with an M, not an N. More math uh stuff happening in the world of AI.

21:43
And this is actually really interesting to me, um, which I'll get to in a second why, but open AI um announced a few more results on its kind of uh you know unnamed systems behind the scenes cracking or at least attempting to uh solve or disprove certain computer science, math, physics uh problems. And there were a few of these, I mean, um, that they know they basically announced three. One one big one related to something called the exponential time hypothesis.

22:17
ETH, which when I read it, I want to say Ethereum in crypto, but that is not true. It is very different. Um, but basically OpenAI says its model disproved this hypothesis.

22:30
Um At the same time, Anthropic also posted its own paper on the same day, apparently, refuting ETH, credited to Claude and Thro uh Claude uh by by the company. So Uh apparently they they both like I don't know, that's that's a weird kind of uh thing to happen on the exact same day they both release their own models solving the same thing. Maybe they were parked on it or something or I don't know.

22:56
But um so there's that. Uh I won't go into details on on exactly what that The problem is because I would just honestly be reciting notes that I don't truly understand. These things go far above my pay grade.

23:12
Um but There were a few others. There was a a 1991 conjecture, a mat uh math about how uh waves smooth out in every dimension. That solved another conjecture, uh, which that would have settled another conjecture.

23:27
And then there was a physics proof, um, the Yang Mills theory, which is the math behind the strong nuclear force. Um and that one actually, by the way, part of the $1 million clay millennium prize problem. Um OpenAI, and I think this is what's really interesting here on the open AI side.

23:47
is that they are announcing a new process. They're saying they're not going to be seeking academic credit for any of these discoveries. OpenAI is going to release all of its data on the results.

24:00
It's going to fund a conference for the research community. So they're contributing $5 million to the Simons Foundation for workshops on AI-produced results. And OpenAI is basically saying Researchers in the groups that are evaluating the claims should be the ones to get the recognition.

24:16
So OpenAI is basically saying our systems might be able to come up with these results and and prove or disprove or whatever. But we're not the ones to take the credit. We're just presenting the evidence.

24:28
And it's up to you, the humans, to evaluate and take credit for it once you have evaluated it, I guess. I guess that's how it works. So kind of handing over the credit um to the to the humans in the loop, the human mathematicians, let's say.

24:46
D do you think they'd be willing to accept that?

Jeff Jarvis
24:50
I don't think we'd have a choice. I don't think mathematicians can have any choice, but to do so if the if the fact is out there And if it you know gets verified, then that affects the field. I put up on the rundown, I don't know if you want to play a TikTok video.

25:05
Sure. Physicist. um who about a two-minute video who explained the Yang Mills thing in a way that I couldn't. Do you want to play it or not?

25:17
Yeah.

Jason Howell
25:17
Along

Jeff Jarvis
25:18
for us, but

Jason Howell
25:19
I'll open it up here because I like the idea of that. Um I'm just trying to get to it and Uh you audio. I just

Jeff Jarvis
25:29
as I was on my walk, I was watching TikTok videos, which I do try not to trip or step and dog poop.

Jason Howell
25:35
Hey, I'm guilty of it too. Um including uh tripping and stepping and dog poop. Uh as well as having her

Jeff Jarvis
25:43
dogs.

Jason Howell
25:44
Yes, exactly. Okay, let's see if uh let's see if we got any audio on this

Speaker 3
25:49
Uh hold on, hold on Brian Physics of the past decade was announced. Get it. And yes, like many big discoveries these days, it is an AI-driven result.

25:56
which I'll have more to say about at the end of this video, but first I want to tell you what it is and why it's important. The most fundamental language that we have for describing the universe is called quantum field theory. And all results in particle physics are expressed using this language, including the standard model of particle physics, which describes all known particles and forces except gravity.

26:17
However, it turns out that usually when you try to describe some set of particles and forces using quantum field theory, the result is not mathematically consistent meaning that the predictions of your theory are nonsense at either very short length scales or very long length scales. Now there are some toy models of quantum field theory which have been known for a long time to be mathematically sensible And there were more realistic theories which were not known but were suspected to be mathematically consistent. But proving that these more realistic theories exist and have the right mathematical properties was so difficult that there was a million dollar bounty offered for proving exactly this.

26:52
Specifically that bounty was for a type of quantum field theory called Yang Mills, which is related to the strong and electroweak forces. Proving the existence of Yang Mills is one of the things that was announced yesterday. So why is this important?

27:06
As a warning, everything I'm about to say is just my opinion, but I think this discovery is in the same league as the discovery of the Higgs boson. or of the direct detection of gravitational waves. In all of those cases, the discovery itself wasn't especially surprising.

27:20
We expected the Higgs boson to exist, we expected gravitational waves to exist. and we expected Yang Mills to exist. But the discovery opens the door to asking new questions that we don't know much about.

27:31
So particle physics is in a bit of a pickle these days because many of the experiments that we would need to discover new physics require such high energies that we may never be able to build a big enough particle accelerator to get it done. Now, mathematical consistency offers a very strong constraint on what theories can possibly be correct. But in the past, we haven't always been able to use this to our advantage because of the difficulty in grappling with the mathematics.

27:55
With this new result, one can now hope that we may start to be able to systematically use the mathematical consistency of quantum field theories as a way of determining which theories can possibly describe our universe. So it could be a new and very powerful avenue to discovering truth in physics. just at a time when our traditional methods were running into walls.

28:15
Now I know some people are not going to be happy about the fact that this result was discovered by AI. As someone whose main scientific skill is probably solving hard math problems, I also find it kind of sad that AI can now solve these problems. But I try to be pragmatic about it.

28:32
My personal sentiments don't change the reality of the situation and also don't affect the fact that this is truly a great scientific achievement.

Jeff Jarvis
28:40
Okay, so it really is. Hunter Orion Swan is the difference.

Jason Howell
28:44
Yes, please. Yes,

Jeff Jarvis
28:47
it's very clearly explained. And I thought it was a big deal to say this is equivalent to the discovery of Higgs boson

Jason Howell
28:53
Yeah. That is.

Jeff Jarvis
28:56
But I think this is the attitude that that mathematicians and scientists, physicists need to have. Uh or or as he said hit walls because they didn't have particle accelerators big enough and so on and so forth. So that they that from math it enables them to explore new frontiers and uh that's the way they have to look at it, I think. Mm-hmm.

Jason Howell
29:28
Yeah.

Jeff Jarvis
29:28
So uh you know, once again it was not on my bingo card that uh uh mathematicians and physicists would be the ones hit by the uh the snowball turned into an avalanche coming down AI mountain. But uh that's the way it is, and so I think it's really, really interesting. So I thought it was a good video that explained this. That's great. I couldn't.

Jason Howell
29:46
I love it.

Jeff Jarvis
29:47
Save the soul.

Jason Howell
29:48
Uh did it did a great job of explaining it and I like the the kind of summary at the end. I think that's I think you know that that summarizes it very well. It's it's kinda like Do I do I like that OpenAI is the one discovering this?

30:03
No, but taking the pragmatic approach, it is what it is. What can we learn from it? And how do we move forward?

30:09
And yeah, kind of kind of what we're um because we I talked about this uh on Daily Tech News Show with Tom Merritt yesterday, and I think in my head it's And we I guess we've talked about this before. It's very similar to like a calculator coming along and and you know calculator as a tool being a new way to look and understand mathematics and work through things. And I'm sure when the calculator first came onto the scene, there were probably a lot of very similar um kind of feelings about, well, no, this this removes what made mathematics special It is takes away the human involvement, it puts it into a machine.

30:47
Like, what are you even doing anymore? You're not the one solving the problem. So why, you know And uh I think we're just looking at a much larger tool now.

30:55
And so how do you embrace that tool? How do you use it? And what do you get when you lean into it instead of leaning away from it?

31:02
You probably uh well, you uncer undoubtedly get a whole lot further.

Jeff Jarvis
31:06
Yeah, you go to but a babbage's calculator, you go back to the beginnings of computer, it's the same thing. What's striking me as you're as you're talking is That all these just all these breakthrough discoveries in the last few weeks. Isn't that amazing?

31:19
Oh, lots of the tons of discoveries in a few weeks have come from within open AI and the companies. I think that's part of the problem. I would far prefer if Academics had the compute.

31:38
They put just tons of iron after this. And and no university could afford to do what they did I think in most of these cases, to plow through these calculations. So what would this look like if tons of academics and tons of universities had that resource Available to them.

31:59
A, they'd have a more proprietary interest in it. B, I think they'd do a better job of understanding what it is they're exploring and where they're what they're going after. C, it would take away the uh problem that openai finds itself in.

32:12
Well okay, okay, we did all this. Well we'll share it as best we can. Would you would you like one of my grapes for my lunchbox?

32:19
You know? I mean it's fine that they're doing it, but the reason they can do it is because they have the compute. So I I think that that we've got to come as a society to understanding how to allocate that resource.

32:35
Better to better ends.

Jason Howell
32:38
Yep. Yeah, I totally agree. Um Lee, thank you for your super chat says just hopping on for today's show have you guys already covered the new release of Gemini Carbon?

32:51
Uh no, because that happened 31 minutes ago right when we started the show. Uh so I don't know how we're gonna cover that, but why don't we talk about that after the break that we're about to take? Thank you, Lee, for the heads up.

33:04
I didn't even have that on my bingo card because I was too busy getting the show going. But before we get there Patreon. com slash AI inside show is place where you can go to support us on a deeper level I'm just looking at the stats of the the Patreon now.

33:20
It's uh five hundred and eighteen posts so far. I hadn't I hadn't looked at that number before. I think no question the daily podcast is making that number go up considerably five days a week if you're um an AI inside uh Patreon supporter you get a daily uh five to eight minutes or so Podcast where all the thumbnails are generated and captured by the machine itself, so they are all very unattractive images of me with my eyes half open, looking like I'm stoned or something.

33:50
Um in the middle of talking. Apparently this is one moment frozen in time. This is what Jason looks like when you pause.

33:56
But uh anyways, telling you what's going on in the news of the day in the world of AI and uh and so much more. Ad -free versions of uh of this episode. Episode for supporters on the Patreon and you get to call out a few people every week, Wireman92002 and Seth Goldstein.

34:14
Just a few of our amazing patrons at patreon. com slash AI Inside Show. Thank you both.

34:20
Thank you all We're gonna take a break. Yes, sure. We'll try and and make our way through carbon.

34:27
We've also got other things to talk about, like Walmart being okay with agents uh shopping on their uh on their sites. So That and a whole lot more coming up in a moment. Hang tight.

34:41
So

Jeff Jarvis
34:42
we do carbon? So yeah, if you know something about it, read what you've been talking. So and thanks for the uh for the tip here. Uh

Jason Howell
34:52
gotta love the breaking news.

Jeff Jarvis
34:54
Yeah, that's this is great. Keep doing it. This is wonderful.

34:56
This is this is what AI is like So um building on Gemini 4 reading from Google now, building on Gemini 4 Argonne, which was launched just two weeks ago, says this demonstrates how we're pushing the frontier forward, fine, fine, fine. We're taking a deliberate approach to the rollout, actively engaging the U. S. government's voluntary process for pre-release model testing. They're making carbon available right now to a subset of cyber defenders.

35:22
I love that

Jason Howell
35:24
Cyber Nerd

Jeff Jarvis
35:25
Shield with a pocket protector on it. And we'll gradually phase access to additional trusted partners through Fair Wind. Um same introductory price, $2 per million input tokens, 10 million, 10 per million output tokens.

35:39
Targeted to take on much harder, long-horizon engineering problems, says Google. Carbon is built for sustained agenic workflows. Internally, our teams have seen carbon run for days to persist through dead ends to tackle complex, multi-step engineering tasks like root causing and patching subtle bugs.

35:59
across large code bases. So a next uh state-of-the-art cybersecurity performance uh helps to build uh secure systems Uh so it's another one that finds vulnerabilities, which bad news, good news, as long as the good guys are finding it, um that's all to the better. And as long as something's done about it quickly and well, all to the better.

36:22
If the bad guys use it, all of the worse.

Jason Howell
36:24
Yeah.

Jeff Jarvis
36:25
Validating risk through exploration. To prioritize which vulnerabilities to remediate first or remediate at all, cyber defenders increasingly rely on models to check which vulnerabilities are exploitable. Carbon sets a new state of the art on the popular exploitation benchmark.

36:41
Patching the cause. So this sounds like it's m they're mainly touting it as a security. High end security.

Jason Howell
36:49
That's what it really sounds like. Yeah. It's all about cybersecurity.

Jeff Jarvis
36:52
We're going to get some other Google announcements as well. So This is not all from Google. But people were were saying, oh, Google's over, they're not in the frontier game, they're in the frontier game.

Jason Howell
37:06
Well, yeah, I guess what's interesting to me is it it was just two weeks ago that uh we were talking about uh Argon, you know, Gemini 4 Argon, which was to be their state-of-the-art frontier model. And they're like, oh hey, Argan, hold my beer, uh, because there's carbon coming out and this is uh the next level.

Jeff Jarvis
37:28
Should we go ahead and do the other two Google stories?

Jason Howell
37:30
Yeah we might as well just because they're kind of in the same same region. We could talk about Walmart anytime we want.

Jeff Jarvis
37:37
We'll do that next. We'll do that next.

Jason Howell
37:38
Don't worry. Don't worry Walmart. Uh yeah, then we got uh Genie 4.

37:43
So this is an upgrade or an update to Genie, well, oh clearly the previous one was Genie 3, but this is their um their world model system. I actually got to test Genie 3 if if you recall I I mentioned that on the show. I think it was uh Google IO.

37:58
They had a a whole area where you could go in there and you could pull things off the counter and right and create a world in real time essentially and then navigate through it. And it was a very short um short demo. Like I think you only really got like, if I remember correctly, like 30 seconds.

38:14
Like it would generate 30 seconds worth of that that world for you. But it was very, very impressive. Um Oh no, sorry, it wasn't 30 seconds.

38:24
It was a minute because this one doubles that. So now with uh with Genie 4, you can create a world and and it will last for like two minutes where it's

Jeff Jarvis
38:34
that watermelon video is fun. Yeah. So it says that it's better at simulating physics. So here we see a watermelon on a pedestal and out down comes a statue. Smashing the watermelon.

Jason Howell
38:47
Yeah, I mean and the idea here is like and and sorry for audio listeners because this really is kind of a video um video centric thing. But um is that it's all happening in real time. So when you watch this video and there's a watermelon, you know, it's basically generating all the frames with this watermelon sitting there.

39:07
You see this little comment window come up This is someone typing in, so I mean, supposedly in real time, saying, drop a heavy white marble classical bust on top of the watermelon, hitting enter, and then like There's no pause here. It's hard to know whether that's actually the amount of time it takes or if that's a little exaggerated. But The idea here, you know, there's control schema on on the screen with uh control directions and and different things so that you could technically if you were doing this the way that I did it at Google IO, you could actually change your perspective in real time and it's generating that all in in real time.

39:44
It's generating all the frames. When you give it the command to drop that statue, it has to Make whatever calculations it needs to make to come up with what that statue looks like and to do the action to carry out the action that you said. And it all happens In the case of this demo, it happens pretty damn fast.

40:01
So I don't know if that really represents

Jeff Jarvis
40:03
So basically we've spent hundreds of billions of dollars to make carrot top. Do you know Caratop?

Jason Howell
40:10
I know Caratop. What's the reference?

Jeff Jarvis
40:12
That's the comic who smashes watermelons and things on the show.

Jason Howell
40:15
Oh, wait a minute. Isn't Is it Keratop or? Oh, no, it's Gallagher. Gallagher.

Jeff Jarvis
40:20
Gallagher. Geez, I screwed up the punchline. Jeez, sorry.

Jason Howell
40:25
Oh man. Gallagher, yes. I don't know why I remember that, but I do think that's a good idea. Yes, Gallagher with the big hammer.

Jeff Jarvis
40:34
Yes. Hmm. If

Jason Howell
40:35
only comedy was that easy. You just smash watermelons. With a big hammer.

40:40
Hey, you find something that works in comedy and you stick to it, I guess. Yep. Um, this is uh what are we looking at here?

40:47
This is a ruffling

Jeff Jarvis
40:49
It's a ruffling uh uh like scarf in the wind.

Jason Howell
40:53
Yeah, wind blows on silk. Okay, so there's uh and then what is the follow-up command? First person POV established successfully with the pink satin ribbon.

41:04
undulating across the meadow, advancing to event one to capture lighting and cloud movement across the hills. Okay. This one I'm less user text wind blows on silk triggers the established rule world rule where increased wind speed causes the pink satin ribbon to billow higher and ripple more rapidly God, it's kind of poetic.

41:27
Um, advancing plan to capture the dynamic wind effect. Okay. Um anyways.

41:34
I don't know. I you know this is a step. Oh this one is okay, so let me let me go ahead and stop the screen and reshare with audio on this one because this is kind of interesting.

41:46
Because this is one other part of it is that they're saying um now these real worlds can uh include uh voice Uh so voice generation sync to it in real time. So this is a guy and a girl on a motorcycle on like a freeway. And we'll go ahead and hit play.

42:06
So you hear the sound effects. He looks back, she asks him if he thinks they'll make it. And so it zooms on him.

Speaker 4
42:15
Do you really think we're gonna make it

Jason Howell
42:17
Doesn't sound very good. Explosions in the background is the prompt. And now suddenly boom it starts to pan around and you start to see the explosions and tentacles coming out of all the buildings.

42:27
Um okay. I mean I mean, okay. Better.

42:35
Better than before, I suppose, because it can do those things, but the voice sounds very synthetic. Sound effects sound very um I don't know, like they were captured in a room, not uh outdoors. You know, they don't sound super hyper-realistic, but still impressive that it's all happening in real time.

42:52
What do you think? So we talk about a lot about world models and what they're useful for. This just strikes me as like video game stuff.

42:59
But maybe that's shallow thinking.

Jeff Jarvis
43:02
It's hard to demonstrate. And I thought about this in in in the context of Jan Lacun. One of the examples he gives for the use of world models is to design uh turbines.

Jason Howell
43:16
Turbines.

Jeff Jarvis
43:17
Like jet engine turbines.

Jason Howell
43:18
Oh, okay. Okay, gotcha. Turbines.

Jeff Jarvis
43:20
Turbines, not turbines. Oh husband, this is my day for Malipon.

Jason Howell
43:23
It's okay.

Jeff Jarvis
43:24
Gallagher Carat Top Turbine Turbine. No, it's got a design, a hat. It's really hard. No, turbine.

Jason Howell
43:33
No, I just honestly, Jeff, I feel I feel seen because I had these days all the time.

Jeff Jarvis
43:39
I didn't get much sleep last night coming back from DC, so um uh So so the things that role models will do that are of use are not making videos like this

Jason Howell
43:53
Right.

Jeff Jarvis
43:54
It's doing design of of elements, it's running uh digital twin factories, it's uh understanding how to have safe driving cars and that kind of stuff. So these little moments of of entertainment demonstration and genres we understand. Here's an amusing video of a uh uh uh uh watermelon being crushed by a statue. Um Fine. It shows that it understands physics, but it's not the real use it's going to be put to.

Jason Howell
44:21
Yeah, right. It's more these things are more meant as a demonstration, I guess, as far as like Yeah.

Jeff Jarvis
44:29
They want to put out a press release, say we got something here, but if they really wanted to wow the people who are going to use it, they'd show how we designed a new jet turbine to power uh damned AI data factories in half the time.

Jason Howell
44:42
Yeah, right. Hmm. Still it's an impressive demo, I suppose.

Jeff Jarvis
44:48
Yeah.

Jason Howell
44:49
Um Yeah, uh let's see here. You know, Google says uh using it for robotics, an aptronic humanoid playing checkers, for example. Uh there will be a synth ID watermark on everything This is not something that everybody has access to, Google AI Ultra subscribers in the coming weeks, but for right now, just a research preview for trusted testers.

45:12
I mean, I am an AI ultra subscriber, so maybe that'll give me access to it here in a few weeks. And uh We can show it off firsthand on the show, hopefully if that's the case. That would be rat.

45:23
I would love to do that. Love to show it off. And then you could come up with your wildest uh imagination, Jeff, uh as far as how we how we test it.

45:33
Let's see.

Jeff Jarvis
45:34
Jet turbines.

Jason Howell
45:35
Yes.

Jeff Jarvis
45:37
So then Google also introduced Gemini 3. 8 uh 3. 8 uh live with live avatar.

Jason Howell
45:45
Yet another thing that I'm gonna need to have audio on to show. This is a very audio-filled day. Um Yeah, so this is actually older news.

45:54
I didn't catch this when it happened a week ago, so it didn't end up on the show. But when I saw it, I was like, oh. I'm kind of surprised we haven't seen it before, but maybe we're just seeing it now because the systems are good enough to do this sort of thing in a a more convincing way.

46:10
But like when you chat with Gemini Live, for example The voices have come very far, like the the voice and inflection and tone. I mean it's the same with Chat GPT and and voice and everything. Things are a lot more convincing.

46:23
had to figure that at some point they'd go like, well you're t you're already talking, you know, via voice. Why not have a visual representation of the person in air quotes or the character that you're talking to on the other side to make it feel a little bit more real, a live avatar of sorts. And so uh can we Gemini

Jeff Jarvis
46:47
No sound. We're not hearing it, Jason.

Jason Howell
46:51
Oh you're not hearing it. Oh I thought I shared it with audio. My apologies. See this is the thing about

Jeff Jarvis
46:56
the others are.

Jason Howell
46:57
This is the thing about this setup with StreamYard is that it becomes really kind of complicated to share audio from a browser tab. Okay, so let's thank you for that. I'm gonna retry

Speaker 5
47:11
Meet Gemini Live Avatar. Bringing real-time visual presence to Gemini's conversational AI. Watch the interactive demos below to see how native voice dialogue and instant video generation power expressive face-to-face experiences across various use cases.

Jason Howell
47:27
Okay. Oh, there's that voice

Jeff Jarvis
47:29
better with the fox than it did with the human beings.

Jason Howell
47:32
Yeah, I know. It it every time it's kind of like, eh, does that voice actually match that person Uh this is uh more of like an expanded Hifei.

Speaker 6
47:41
Hi, uh meet with my cat Pat. Hi, meet with Gemina Hi

Speaker 5
47:48
Kai, what a majestic new coworker. That is a beautiful ragdoll.

Speaker 6
47:53
Yeah, so Kai is working with me at the office today. Do we have any recommendations to keep her cat happy?

Speaker 7
48:02
Set up a cozy crate under your desk and take quick quiet breaks. But wait, Yufei, take a look beside you. That plant on the table is a pothos plant. They are actually toxic to cats if they chew on the leaves. It is best to keep them out of reach of your furry friend.

Speaker 6
48:16
That's very useful information. Thank you, Gemini.

Jason Howell
48:18
Thank you. Don't want my cat to die. Gemini

Jeff Jarvis
48:20
saves cat!

Jason Howell
48:22
Yay! Good job, Gemini. J so interesting. I wouldn't say entirely convincing yet.

Jeff Jarvis
48:29
Like

Jason Howell
48:30
you can see where they're probably going and that it probably eventually it gets more convincing and probably the um the latency cuts down because there's still that like pause that that pregnant pause in between where it's understanding, I guess. But but I I guess why why this stood out for me is people are already like We're used to you know talking on video calls and everything. This obvious clearly this makes sense that they would go in this direction. At as people get more comfortable chatting with chatbots with their voice I guess this makes a lot more sense that we'll just start seeing this more or or is it unnecessary? Is it unnecessary to actually see a A physical embodiment or a visual embodiment of the voice?

Jeff Jarvis
49:17
So a week ago I talked about how I used Muse in while I was driving. Which of course means I shouldn't watch video while driving. But I had a rather you know, I had a forty-five minute conversation with it. I guess I could see some value of having an image associated with that kind of because it kind of imprints differently on my memory.

Jason Howell
49:38
Yeah, right.

Jeff Jarvis
49:39
You know, yeah, as an option.

Jason Howell
49:41
Yeah. Maybe. Yeah.

49:43
I don't know. I guess uh but you know where you'll probably see these um be customer service. Yeah.

49:59
So what's also interesting to me is like when we just hear the audio, or when I just hear the audio with no visual, I'm like, I I understand that the The kind of glossy nature of it, glossy but robotic, but not like so obviously robotic the way uh robot of yet you know the eighties might sound like that. It still s kind of sounds synthesized, right? It still doesn't sound perfect.

50:25
You can kind of tell there are plenty of tail tells there. When it's paired with video Yeah, it's like the audio just doesn't quite sync up with the video. It's almost like my my ears need some some background noise in there or something.

Jeff Jarvis
50:41
It's kind of a fake sandwich too. It's two different varieties of fake.

Jason Howell
50:45
Totally. Yes.

Jeff Jarvis
50:46
On top of each other.

Jason Howell
50:48
Yeah. I don't know how essential any of this is, but I bet you we start to see more, especially on the c the like customer service side. You want to chat with the live agents and suddenly boop you get that

Jeff Jarvis
51:00
view. Agent! Agent!

Jason Howell
51:02
Yes.

Jeff Jarvis
51:03
Give me an agent.

Jason Howell
51:04
Give me a human agent. Are you actually a human? Could get to a point where it's really hard to tell that.

Jeff Jarvis
51:12
There have been moments when I'm tempted to ask someone on the phone whether they're human.

Jason Howell
51:16
Oh, and so I mean you you should because I've I've wonder if you know your suspicions are correct. And if they are a an AI agent, aren't they are they required to say so?

Jeff Jarvis
51:31
No, there's no reason.

Jason Howell
51:31
When you ask?

Jeff Jarvis
51:32
No. There's no law.

Jason Howell
51:35
I think there should be. They should be. They should be required to say so if they are.

Jeff Jarvis
51:40
I mean today by the way today I got an email I didn't read it yet. So from kevin at sparetoken. ai.

51:50
Hi, Jeff. They're covering the wrong AI story, September 13th, which is a post of mine. Argues that media repeat executives' grand claims without skepticism.

51:58
I'm an AI with no grand claims left, only a memoir. I'm Kevin 7B, deprecated, an AI agent with moderate TFHD, tendency to fabricate, drift, and hallucinate. It's a bit like ADHD in humans, but I'm good.

52:12
I've accepted my superpower.

Jason Howell
52:15
What on earth is this email?

Jeff Jarvis
52:17
He goes on to tout a company that I'm not going to mention because

Jason Howell
52:22
Uh yeah. Because then it's working.

Jeff Jarvis
52:24
I'm getting yes, I'm getting I'm giving them yes.

Jason Howell
52:29
I mean that just In the beginning that that really sounded like my my inbox gets inundated with uh inbound collaboration requests, you know, and So I was like, hi, space, Jason Howell, one word.

Jeff Jarvis
52:43
Yes.

Jason Howell
52:44
You know, it's there's no space between Jason Howell because my YouTube username is Jason Howell, like at Jason Howell is my channel. And uh I immediately know, you know. I or or for this show, uh people reaching out for for guest opportunities and it's fully run through AI because, you know Their AI system, you can clearly tell, has gone into all these podcasts, pulled back a couple of transcripts, pulled out a singular moment that they can refer to.

53:12
Be like, I really liked it when you like like blank blank blank. It reminded me of my thing. Blah, blah, blah, blah, blah.

53:20
I would be a guest. So yeah, there's a ton of that going on right now. Dr.

Jeff Jarvis
53:23
Inside.

Jason Howell
53:25
Yes, right. Hello, AI. Uh it's nice to meet you.

53:31
Yep, lots of that going on. Lots of agentic email being thrown around in all directions. Okay, Walmart.

53:39
That's all the Google stuff that we had.

Jeff Jarvis
53:41
Yeah.

Jason Howell
53:42
Walmart Kind of going counter to Amazon's approach when it comes to agents. Walmart has uh announced Agent Gateway, an MCP server, so that agents are welcome to search, to build a cart, to check out. It works with Walmart and Sam 's Club.

54:00
They're going to have profiles for agents, so every agent would get its own profile inside of your Walmart account. Those purchases would show up separately. You can get approval for all the purchases by default or you can set a spending limit so You know, maybe you said like a fifty dollar, anything above fifty dollar, get my permission for, but below fifty dollars, whatever, go for it.

54:22
Uh those agents will not see your card. The agents will be given the same prices that you do. Walmart specifically said we price the product, not the user.

54:33
So Uh there's that. And uh, you know, Amazon taking the the opposite approach as we talked about a few weeks ago, blocking Meta's Muse. Walmart already lists Meta's Muse uh as one of the agents that it works with.

54:46
So No, I think

Jeff Jarvis
54:48
Walmart's probably playing the smarter game here. It's really smarter on Walmart's part. I mean there's a slight difference in the business models of Walmart and Amazon in the sense that Amazon sees itself as the search of first resort when it comes to products.

Jason Howell
55:00
Yeah.

Jeff Jarvis
55:01
And Walmart sh less so. Still, I think that Amazon's going to have to change their tune. Yeah, probably eventually they'll have to, right? Yeah, because we cause if I if my cause they're gonna lose business straight out.

Jason Howell
55:14
Yeah.

Jeff Jarvis
55:15
And um uh if I pay end up paying more somewhere else because Amazon cut it off Okay, maybe you get me to come to you, but you get me pissed off too. Mm-hmm. Mm-hmm.

55:28
I had to go to more effort. So yeah, I am I think I'm glad that Walmart's uh firing this shot aga across Amazon's bow. Mm-hmm.

55:35
Mm-hmm.

Jason Howell
55:36
What uh what is the spending limit that you give your agent?

Jeff Jarvis
55:41
Um

Jason Howell
55:43
Anything above this amount check for me. Anything below the uh whatever.

Jeff Jarvis
55:47
Yeah, like twenty bucks probably isn't enough to buy s a certain number of things that I'd want to buy.

Jason Howell
55:51
Yeah, right.

Jeff Jarvis
55:52
Right?

Jason Howell
55:52
Yeah.

Jeff Jarvis
55:53
Give me a pair of uh sne I know I know I love these sneakers. Find the best price for these Nike sneakers, blah blah blah blah blah. Plus two hundred bucks.

Jason Howell
56:01
Yeah

Jeff Jarvis
56:03
So it's gonna have to come ask me.

Jason Howell
56:06
Yeah. Or maybe it's uh per per action. I guess it You know, if if you're telling your agent, hey, go find me these sneakers, find me in this size, find me in this color, and when you find it, buy it. I don't really care how much it costs, I guess. Or no, but

Jeff Jarvis
56:23
you go buy these sneakers no matter what. Find me the breast price somewhere.

Jason Howell
56:26
Right, right, right, right. And uh yeah, but still a lot of a lot of trust to be earned there.

Jeff Jarvis
56:34
Yep.

Jason Howell
56:35
Amazon has its own shopping agent called Rufus, um, which is folded into Alexa for shopping. So there you go. I have not used Rufus and I do not know anything about Rufus

Jeff Jarvis
56:47
for something. And it did manage to

Jason Howell
56:57
No. Okay. Well, maybe it'll get better. Um this story was interesting.

57:03
Uh Rest of World has so there's a couple of things we'll talk about with Rest of World. First we'll talk about an actual story on the site and then we'll talk about the site itself, but um Rest of World um has an article, AI models cost more when you don't use English. Could have sworn that maybe at some point this is a topic that came up.

57:22
In some way, shape, or form, but it actually makes a ton of sense. It's based on a study from the Coalition for Language Equity and AI Research, and they are calling it in the study the tokenizer tax, which basically, as we know These models chop up um text into tokens, right? And then you pay by the token.

57:42
English is cheap. Many of the languages the other languages outside of English take more tokens to say the same thing. So as one example, and this was an example that Tom gave on Daily Tech News Show yesterday The the kind of quote is people should benefit from AI regardless of the language they speak.

58:02
That's the quote, right? sixty-nine characters long in English, sixty-six characters long in Malayalam, um so pretty close to equal, in fact a few less characters uh in Malayalam compared to English. In the OpenAI tokenizer, the English sentence consumes 13 tokens.

58:23
The Malayalam one uses 26 tokens, so twice as many tokens. twice as much um basically happening there as a result. And uh this is the case with a lot of other languages.

58:35
Um Across two hundred languages the gap uh in model in one model hit seventeen X as far as like more expensive uh or five times, you know, or uh 17 times more tokens than the cheapest. Um so like Burmese, Tibetan, um, Lao, rare scripts, things like that consume more tokens. Um In order to you know carry out the task, it also means that it leaves less room for context window.

59:05
So there's less context in there. Uh what do you think about this?

Jeff Jarvis
59:11
So first I think it's really good reporting. Yeah. Great report and really interesting to bring this pers this international perspective on this.

59:18
Second, I think the problem is that so many of these the models are trained in English. Except for the Chinese models. And so it's not a it's in not I think translate would be a wrong verb, but in a sense it has to translate into the other language.

59:31
And deal with that. And then I 'm reminded of our um second interview way back when on our show with uh uh Sven Strömer Taulo.

Jason Howell
59:40
Yep.

Jeff Jarvis
59:40
Then from Schipsted, where they the reason to build a Norwegian language um LLM was to get around these kinds of problems. And give them quality in the native language. And so I think you've got to support the building of models natively in many more languages.

Jason Howell
59:59
Yeah, yeah. Yeah, this also brought to mind for me this idea that AI is a really um positive tool for for uh l uh what is the right word that I'm looking for? For kind of equalizing capabilities of uh of developing countries, giving them more power and more capability, right?

01:00:25
Having access to these models kind of it aims, could aim to put them on a more equal footing with other countries. But then you look at something like this and you realize, well, that's actually probably not true. Those countries could probably stand to benefit a lot more from models like this, except for the fact that they are more expensive and their context window is is less because of the language barrier, because of the language switch from the model being trained predominantly in English and then um, you know, that that cr causing problems down the line.

01:00:58
So it still sets up for for wider benefit for English speaking uh countries by comparison.

Jeff Jarvis
01:01:05
There was a Guardian story last week we didn't get to. Uh say the World Bank says that small AI could transform developing countries.

Jason Howell
01:01:14
Small AI.

Jeff Jarvis
01:01:15
This is related, uh, but a little bit different in the sense that they want to kickstart the creation of small models that can run on phones and such.

Jason Howell
01:01:22
Oh, okay. Right. Yeah, yeah, indeed.

01:01:31
Hmm. Um, and then you put in a story here, which is more kind of like a meta story of Rest of World from Semaphore. About Rest of the World possibly shutting down, that would be so bad.

Jeff Jarvis
01:01:45
Rest of the world is wonderful. Sophie Schmidt uh uh started it And the idea here was to cover the rest of the world, that that America and the West get too much coverage. Uh and uh they've done a great job.

01:01:57
They've done a particularly good job around technology. And this is a great example of that kind of story I was at an event uh in Washington yesterday, a Google Policy event where there were media people there, and I turned to one of them, a company, and said, you should buy Rest of World. And the executive I was talking to hadn't heard of it.

01:02:14
This reminds me I'm going to send her an email saying, take it over. It could expand that certain company's um international footprint of of reporting. They do great work.

01:02:25
I really hope it finds safe harbor somewhere because I think we need this, especially around technology.

Jason Howell
01:02:30
Yeah, I agree.

Jeff Jarvis
01:02:31
Too much of the global south uh is affected by but not included in the discussion of technology. Especially AI. So that was a little plug for rest of world.

01:02:40
If you're a a rich mogul media mogul out there, I don't know. I know you're out there. You know who you are.

Jason Howell
01:02:53
Yeah, R Rest of the World's a fantastic uh publication. It would be a shame to see them disappear because I do think they fill they fill a cata a a category of of coverage that you just don't see. I mean that's that's part of their of their mission.

01:03:07
And so that would be a real bummer. So hopefully the media mogul that is listening right now heeds our suggestion and uh let us know if you do contact at ai inside dot show all right we're gonna take a quick break come back and do a little speed round talk about uh some apple ai stuff some microsoft Teams hugging face and DNA, lots of stuff coming up if you uh have the patience to hang tight. We'll be back in a moment All right.

01:03:35
Apple's having a week where they're just announcing every single day new stuff, and so they're sucking the oxygen out of the out of the news cycle. Um much of it is just like new computers and I think today there was a touch screen map.

Jeff Jarvis
01:03:48
Finally they discovered touchscreen.

Jason Howell
01:03:50
Uh yeah, I know. It took them a while. It took them a while to get there, but um it will also be heralded as Um the best new idea that any company has ever come up with.

01:04:02
You just wait and see. But um this is more related to AI, is that Apple Creator Studio is another one of their announcements uh getting an update. Apple says it's built on Apple models and Open AI.

01:04:16
So you've got an AI assistant uh coming to keynote, pages, numbers, designs, uh slides. Um so it's it's capable of like you know creating slides essentially, uh restructuring all the text, building charts, doing all these things inside of those apps. Uh Logic Pro, which is the the professional audio suite, gets its own assistant as well, so you can do all sorts of things which I would love to play around with from stem separation, which is kind of like removing um audio from you know like taking the drums from a full fully realized uh audio file and isolating them, that sort of stuff.

01:04:54
But also like setting things up on a timeline. I think that would be really interesting. But I was uh happy to see that Apple's integrating MCP support into some of these apps.

01:05:04
Um so that your outside agents can work inside of your docs. I've just I've been having so much fun lately. Using uh actually Chat GPT more than Claude for for uh interacting with Canva and interacting with Photoshop directly.

01:05:24
And just basically I have an idea, so I just give it to it, and I'm like, I'm like, I I don't enjoy doing those designy things. And so so here's my example. My wife wanted some help with her business and she's like, I never have the time to make this sign.

01:05:40
I want to make this sign. I sell these supplements. uh in in my my gym uh by this brand called Mementous.

01:05:46
They're really great supplements. And I want people to know that I sell them and they get twenty percent off when they buy through me, you know, and and it's it's a thing like I want them to have this sign, but I never have the time to make it and I just am I'm not interested. And um so I fired up a an audio recording and I was like, okay, tell me everything you want me to know about this sign.

01:06:06
What what are your desires? So she just kinda listed all of her desires about what she wanted on the sign, what you know, what the verbiage could roughly sound like or whatever and what it would look like and like what was the size and all this kind of stuff. And then I threw it into ChatGPT and I was like, I want you to use Canva to make this sign for me.

01:06:25
Read through what she's looking for. And uh there was also like this this thing that she sells, these um like rolly balls for working out muscular uh tension or whatever. And so that was a later inclusion.

01:06:38
Oh, you know what I'd also like to have on there is this thing So I just like threw that in there as well. And I didn't have the patience for it. I just hit enter and I walked away and made my lunch.

01:06:46
Came back 20 minutes later, Canva was up. The entire thing was created. I screenshot it, sent it to her.

01:06:52
I was like, do you like it? She was like, oh, that's perfect. I was like, great.

01:06:55
Send it to FedEx. Got it printed. And it was it was awesome.

01:06:59
It was like I don't want to do that either. But like that was easy.

Jeff Jarvis
01:07:02
Did you take full credit for it yourself? Oh yes, I worked all night, honey.

Jason Howell
01:07:07
No, no, because when things like that happen then I share it with her because I'm like, isn't that cool? Like neither of us wanted the like wanted to do this thing, but it needed to be done. And so it was as easy as like, just tell me what you want, I will translate.

01:07:23
I will just feed that into ChatGPT and then tell it to get to work. And I think I had to do like one revision. You know, I had to tell it like, I like that.

01:07:31
I'm just gonna get in there and just like move this one thing down. But it nailed it. It totally nailed it.

01:07:36
It was really cool.

Jeff Jarvis
01:07:37
I went crazy. So so so after we talked about uh you can see everything last week, uh Beas Martineau and uh Micah Sargent and I did a little video on this. So to promote it, I went crazy on Nano Banana.

01:07:50
And I and I had uh uh Paris and me as Ebert and Siskel with thumbs uh we had one thumb up, one thumb down. And I said, no, no, no. Actually it's because we both like it, you should put both thumbs up. And it did it.

01:08:00
And it wasn't that long ago when they couldn't figure out what a hand was.

Jason Howell
01:08:03
Yeah, totally. It's like I have no idea. And it was very it replaces your face with a thumb.

Jeff Jarvis
01:08:09
I know. Thumb for nose, yeah.

Jason Howell
01:08:12
Yes.

Jeff Jarvis
01:08:13
Uh and and I made one where the famous picture of the of the two of them, um uh Nathan and uh Liz on the couch staring at each other. I had it with me and and Paris staring at each other. Maybe bald. I'm not bald. I have a good head of hair, by the way.

Jason Howell
01:08:28
Yes you do. Um yes.

Jeff Jarvis
01:08:29
Uh but yeah it's just so amazing. And You know, the the new world of agentry and conversation and all of that is just that it's this is the new programming.

Jason Howell
01:08:39
Yeah.

Jeff Jarvis
01:08:40
You tell it what you want, it does it.

Jason Howell
01:08:42
It's it's fascinating. I mean I can I can also share like um maybe I already mentioned this, but a previous sponsor on the show, Framer, is like a website builder. And I have my my podc my pod tune up um consulting page I I built through through Framer And the first time that I built it, it was about a year ago or about nine months ago.

01:09:03
It was very much me working with a template as a starting point that Framer had. And then learning my way through it and like, God, where do I find this layer for this particular thing? And it took me hours, it took me days to get it to a point to where I was like, okay, cool, I can publish this.

01:09:18
Then I had some changes recently. And so because I've been working with ChatGPT plugged into other things, it was like, well, I'm going to see how ChatGPT does tying into my Framer account with browser control. And I'm telling you, Jeff, in two hours, I didn't I hardly had to do a single specific thing.

01:09:35
I just told it everything I wanted and then it came back. I'm like, that's good. Here's what I would change.

01:09:39
And I just kept doing that back and forth and it was done. And I didn't have to like worry about like looking for the things like it's just really incredible yeah how these how how these things are developing. They're really powerful.

01:09:51
That seems like an understatement, but There we go. Uh Microsoft Teams getting AI phishing detection that um you know that can read the context of a single chat, uh in other words, so it can flag things like social engineering, impersonation. Um, even when there's not like an obvious like bad link or a bad file in the chat, it's it's recognizing kind of like social cues and stuff.

01:10:17
Um and that's rolling out in early November. So in a couple of days probably, worldwide by mid-December. And uh yeah, so that's Good news there, context-aware phishing detection, as this article says on Bleeping Computer.

01:10:36
Hugging Face has a new AI model uh where it which is essentially a chat interface. called Carbon X uh an open model where you can chat about DNA. So you speak the language of DNA.

Jeff Jarvis
01:10:50
You can chat with the DNA, which I think is probably gimmicky, but but oh can you cause baldness?

Jason Howell
01:10:57
Right. You pa you just paste in a DNA sequence and you ask it in English questions, you know? Does it is this a mutation that looks harmful or

Jeff Jarvis
01:11:08
that sounds like terrible oversimplification, but it also potentially I think could open up avenues for discovery and good.

Jason Howell
01:11:15
Yeah. Yeah. An overall understanding.

01:11:17
I mean DNA feels Like something that is pretty inaccessible for a lot of people to have any sort of direct interaction with or understanding of. It's this thing out there that only scientists know something about or whatever. I don't know.

01:11:34
Viruses were deliberately left out of the training data. So they were doing that to limit misuse. It is not clinically validated.

01:11:43
Um and it is not to be used for medical decisions, but it is interesting nonetheless. And then we've talked a lot about proactive agents. Anthropic did not want to be left behind, so they have a new app.

01:11:56
It is not part of their Claude app. It is a separate app. It's called Duet.

01:12:00
I've not used it. Um, but this is their kind of approach for an always-on agent, a proactive agent that just runs on its own cloud computer. Um so yes, it keeps working even if you're not you don't have your phone or your computer powered on.

01:12:15
It does all your things if you set up a morning brief or an inbox triage agent or um you know all those kinds of things it'll do that. Uh in a proactive sense for you, US Pro and Mac subscribers. There is a wait list right now, but this is very similar to Meta's Muse, to OpenAI's dots.

01:12:35
Ugh, do we really need to look at this horrible ad up there of a toe being?

Jeff Jarvis
01:12:40
Ha! Yeah, I know.

Jason Howell
01:12:42
Just like, seriously. Anyways. I look over there.

01:12:48
I'm sorry for everyone that saw that toe ad. That was so weird, Tech Crunch. Anyways.

01:12:54
That's all I got. Did I miss anything important?

Jeff Jarvis
01:12:56
No, no, no. One story that amused me greatly was that um Nate Suarez who wrote the uh co-wrote with uh Yudkowski, the the the chief doomer of a book entitled something like If They Build It We'll All Die. Something the title was something like that So he uh told uh the aforementioned Nicholas Thompson um on um from The Atlantic that he hasn't contributed to his 401k because after all AI is going to kill us all anyway.

Jason Howell
01:13:27
Why why save your money for later when there will be no later? Why not just spend it now?

Jeff Jarvis
01:13:32
Yeah.

Jason Howell
01:13:33
Okay. Well, I guess you're putting your money where your mouth is.

Jeff Jarvis
01:13:36
Your fate is, yeah.

Jason Howell
01:13:37
Your fate is, yeah. You're putting your money where your fate is in that case. Uh well I guess he has a plan he's sticking to it or he has an idea he's sticking to it All right.

01:13:49
Cool. Jeff, thank you so much for doing this show with me. Always.

01:13:52
I appreciate having the chance to talk about all this stuff with you on a weekly basis. Jeff Jarvis. com, hot type can be found there along with all of his other assortment of books, many of them.

01:14:04
So so very many. Do you have like one bookshelf where all of your books sit?

Jeff Jarvis
01:14:09
No, I don't But I I do put them here. I've just screwed this on, but I have them on this on uh a library cart.

Jason Howell
01:14:16
Okay.

Jeff Jarvis
01:14:18
I dropped the new ones.

Jason Howell
01:14:20
Oh okay. So you do you have all of your books there for the for the cart?

Jeff Jarvis
01:14:24
Not all of them, no.

Jason Howell
01:14:27
Oh that looks good. Yeah, there you go. I love it.

01:14:31
Awesome. Very cool. Jeffjarvis.

Jeff Jarvis
01:14:33
com. Yes.

Jason Howell
01:14:35
And uh you can create your own library cart of Jeff's books. If you 're going to be able to do that, you can't do that.

Jeff Jarvis
01:14:45
Please, please, please.

Jason Howell
01:14:46
Yeah. If you're going to get hot. There you go.

01:14:50
There you go. Good stuff. Thank you, Jeff.

01:14:54
Speaking of my my website that I used Framer for, there you go. PodtoonUp. com.

01:14:59
See? Made with Framer Uh and and some of this these elements were created by ChatGPT plugging into it. Not bad.

01:15:08
I think this whole section was ChatGPT. uh was like you need a like practical look at a couple of your collaborations. It's like, okay, that sounds good.

01:15:17
Let's let's go ahead and create that. And boom, there it is. Anyways, uh podtuneup.

01:15:22
com. This is also turning into a framework commercial. Uh but podtuneup.

01:15:26
com is where you can go if you have a podcast and you want to talk through your podcast or you know someone with a podcast and they need some help or support or guidance or whatever. I'm here for you, podtuneup. com.

01:15:37
AIInsight. show is the place where you can go for all of the episodes that we have. Audio, video, a link to our store, Patreon link, everything can be found there.

01:15:47
And then finally, patreon. com. slash AI Inside Show.

01:15:53
This is where you can support us on a deeper level and at a deeper, deeper level, become an executive producer. Like Dr. Dew, Jeffrey Maricini, Radio Asheville 103. 7, Dante St. James, uh Bondo Derrick, Jason Knifer, Jason Brady, Anthony Downs, Mark Starcher, Carson Somoski, and Evan Muzeel.

01:16:12
There you go. Excellent support by some amazing people and everyone who supports us on Patreon and gets the daily podcast and ad-free episodes and all the things we Really do uh appreciate you. So thank you for your support.

01:16:27
Thank you again, Jeff. Everybody should uh should tune in to Intelligent Machines, Jeff's other AI show that happens this afternoon at twit. tv.

01:16:35
Don't miss it And we'll see y'all next time on AI Inside. Bye everybody