WEBVTT
00:00:00.029 --> 00:00:02.942
Mike, you're back in your familiar setting of your home office.
00:00:02.962 --> 00:00:04.232
I'm back in the UK.
00:00:04.312 --> 00:00:13.682
We're no longer at Trust Con, but that doesn't mean we haven't been perusing the internet for fun, fun things that we, you know, might take our mind off of the current moment.
00:00:14.300 --> 00:00:14.710
Yes
00:00:15.108 --> 00:00:16.562
what have you been kind of perusing this week?
00:00:16.815 --> 00:00:25.315
so mine, mine relates to a sort of pop culture reference, for people who grew up in the 1980s, which is, uh, in the US for me.
00:00:25.315 --> 00:00:30.138
So I don't know, did the, cartoon show Scooby-Doo make it to the UK?
00:00:30.138 --> 00:00:30.518
Okay.
00:00:30.847 --> 00:00:31.457
Yeah.
00:00:32.027 --> 00:00:32.747
What do you think we are?
00:00:32.747 --> 00:00:33.167
Kind of
00:00:33.258 --> 00:00:34.527
I, I don't know.
00:00:34.618 --> 00:00:42.927
The, every once in a while you hear random things and they're little cultural things that, you know, did or did not leave the US, but okay
00:00:43.317 --> 00:00:48.356
this is a bit like when Elon Musk says that London is full of pedophiles and rapists, isn't it?
00:00:49.097 --> 00:00:56.746
It-- I don't know who's, I don't know who's told you that we didn't get Scooby-Doo, but they need to, they need to course correct what they've said.
00:00:56.796 --> 00:00:57.046
That's
00:00:57.249 --> 00:00:57.749
Okay.
00:00:58.088 --> 00:00:58.558
All right.
00:00:58.558 --> 00:00:59.029
Fair enough.
00:00:59.069 --> 00:01:01.209
All right, so then you'll, you'll get this reference.
00:01:01.209 --> 00:01:01.389
The...
00:01:01.668 --> 00:01:07.784
So, in Scooby-Doo, they drove around in the Mystery Machine, the famous brightly colored van.
00:01:07.899 --> 00:01:11.671
So I found, there have been a few pe- I know this has been done.
00:01:11.671 --> 00:01:17.698
This is probably not the only one, but people have made their own exact replicas of the Mystery Van.
00:01:18.069 --> 00:01:22.349
and there is one that is currently for sale based in Oregon.
00:01:22.599 --> 00:01:27.371
It is, currently for sale for, uh, has a bid of$7,000 as we speak,
00:01:28.602 --> 00:01:29.322
Let's buy it.
00:01:29.811 --> 00:01:31.882
looks amazing.
00:01:32.266 --> 00:01:33.025
Really?
00:01:33.091 --> 00:01:34.191
Like you, you...
00:01:34.191 --> 00:01:41.064
I, I will put the link in the show notes, and you need to see this, not just for the outside, because the outside looks exactly as you expect it to.
00:01:41.194 --> 00:01:42.594
It looks like the Mystery Machine.
00:01:43.105 --> 00:01:45.305
The inside, though, that is the story.
00:01:46.078 --> 00:01:46.849
Really?
00:01:47.004 --> 00:01:48.495
have kitted out this thing.
00:01:48.745 --> 00:01:50.875
It has Scooby-Doo headrests.
00:01:51.295 --> 00:02:04.561
The entire interior of the van is in brightly colored plush stuff, and the back of the van looks like, you know, m- my childhood dream of what the back of a van would look like.
00:02:04.831 --> 00:02:07.662
It's got a giant TV, video games.
00:02:07.852 --> 00:02:10.692
There's Scooby-Doo memorabilia everywhere.
00:02:11.051 --> 00:02:11.891
Oh my God.
00:02:12.032 --> 00:02:13.072
It is astounding.
00:02:13.622 --> 00:02:14.092
Yeah.
00:02:14.391 --> 00:02:15.951
This, the Ctrl-Alt-Speech mobile.
00:02:16.876 --> 00:02:18.836
I, I think, I think it may need to be.
00:02:19.635 --> 00:02:25.175
It even, it even, it even has customized license plates that say,"Zoinks."
00:02:25.944 --> 00:02:27.974
Oh, nice touch.
00:02:28.365 --> 00:02:29.754
Really nice touch.
00:02:30.465 --> 00:02:32.305
don't know, I don't know why we're still recording.
00:02:32.395 --> 00:02:34.764
You, you should be, you should be heading straight to Oregon.
00:02:37.995 --> 00:02:38.455
Okay.
00:02:38.528 --> 00:02:38.688
you?
00:02:38.688 --> 00:02:40.149
What, what, what fun thing did you find?
00:02:40.812 --> 00:02:46.272
Well, I'm gonna ask you, about something that I'm pretty sure hasn't made its way to the US, at least not yet.
00:02:46.665 --> 00:02:47.639
what do you know about cricket?
00:02:50.135 --> 00:02:52.756
I do, I do actually know a little bit about cricket.
00:02:52.756 --> 00:02:57.742
I'm a huge baseball fan, and cricket is obviously sort of a, related sport.
00:02:57.752 --> 00:03:07.733
And at times I have spent time to figure out how cricket works, and then I remember for a few months, and then I tend to forget some of the rules as compared to baseball.
00:03:07.763 --> 00:03:10.272
But I, I'm, you know, g- generally aware of it.
00:03:10.527 --> 00:03:11.168
Okay.
00:03:11.168 --> 00:03:11.228
Okay.
00:03:11.228 --> 00:03:15.485
Well, there has been a, cricket story this week that has broken out of the UK.
00:03:15.485 --> 00:03:16.534
It's gone global.
00:03:16.544 --> 00:03:23.770
There's been reporting about this story all over the world, and that's because it concerns a very local cricket team called Saltburn.
00:03:25.288 --> 00:03:40.170
an amateur club, which has got under a bit of, gotten into hot, some hot water because it has been accused of cheating by, wait for it, clicking the fingers of one of the fielders to imitate the batter hitting the ball.
00:03:41.161 --> 00:03:52.038
And, and so what's happened is that they, this, team has done incredibly well by, having lots of catches taken by the wicket keeper, the guy who stands behind the stumps,
00:03:52.461 --> 00:03:52.872
Uh-huh
00:03:53.064 --> 00:03:55.427
which are actually not legitimate catches.
00:03:55.467 --> 00:04:02.508
They are somebody standing close by pretending clicking their fingers that the batsman has hit it.
00:04:02.737 --> 00:04:11.634
And, you know, we, we like a rule in the UK, But rules, cricket rules are even more, important to us as a nation, and this has, this has gone big.
00:04:11.694 --> 00:04:12.844
This has gone incredibly big.
00:04:12.974 --> 00:04:25.845
And it's, and it's all because, it's all because people have been looking very closely at live streams of this amateur cricket team and spotting where this one particular fielder has been clicking their fingers.
00:04:26.471 --> 00:04:26.480
Uh?
00:04:26.774 --> 00:04:35.404
we have a, a kind of entire, group of internet sleuths who've been going back through old matches and spotting who this offending
00:04:35.555 --> 00:04:37.475
this is what the internet is for
00:04:37.845 --> 00:04:38.574
Exactly.
00:04:38.605 --> 00:04:41.954
Bringing cricketers, offending cricketers to justice.
00:04:42.259 --> 00:04:42.928
There we go
00:04:44.375 --> 00:04:45.144
Let's get cracking.
00:04:53.642 --> 00:05:02.012
Hello and welcome to Ctrl Alt Speech, the podcast where we make sense of the major debates shaping online speech, platform power, content moderation, and the future of the internet.
00:05:02.452 --> 00:05:05.262
It's July the 30th, 2026, and this week.
00:05:05.322 --> 00:05:16.629
We're talking about the race to regulate AI, Europe's latest attempts to bring big tech platforms to heel, and in our special segment for Patreon listeners, how xAI is challenging lawmakers in Minnesota.
00:05:16.910 --> 00:05:25.079
Ctrl Alt Speech is the weekly coping mechanism of Mike Masnick, founder and editor of Techdirt, and me, Ben Whitelaw, founder and editor of Everything in Moderation.
00:05:25.560 --> 00:05:27.139
Mike, we're back from Trust Con.
00:05:27.449 --> 00:05:30.930
You seem like you've recovered from three intense days.
00:05:32.648 --> 00:05:34.028
Let's say, let's say that.
00:05:34.088 --> 00:05:36.411
Uh, it's, you know, it's been a lot.
00:05:36.528 --> 00:05:39.728
I can hear in the background that, that your son is happy that you're home.
00:05:41.165 --> 00:05:46.966
Yep, that is, that is, uh, that is my son who is doing well and making his presence felt.
00:05:47.894 --> 00:05:48.694
Yeah, it's good.
00:05:48.704 --> 00:05:49.223
It's good.
00:05:49.223 --> 00:05:52.153
We always, we always enjoy when, when he says hello.
00:05:53.579 --> 00:05:59.319
This is the glamour of recording a podcast on a Thursday evening in the UK right at bedtime.
00:05:59.519 --> 00:06:00.720
There's, there's nothing better.
00:06:01.139 --> 00:06:04.959
Uh, you know, you could argue that I'm getting out of it, but, yeah.
00:06:05.216 --> 00:06:08.326
let's cr- It's better to be here than it is there, I'll say that.
00:06:10.586 --> 00:06:11.665
Um, yeah.
00:06:11.665 --> 00:06:13.766
Wh- what did you come away from Trust Con thinking, Mike?
00:06:14.060 --> 00:06:14.569
What was your takeaways?
00:06:15.665 --> 00:06:16.156
It was great.
00:06:16.165 --> 00:06:16.836
It's always great.
00:06:16.855 --> 00:06:18.086
There's always lots of people to talk to.
00:06:18.086 --> 00:06:23.232
It's always good to get a, feel for what's happening and what people are thinking about and the different perspectives.
00:06:23.262 --> 00:06:29.851
Obviously, a lot of AI talk, I think, was sort of the theme in sort of thinking through sort of...
00:06:30.101 --> 00:06:34.041
this was a conference that historically focused more on other areas of the internet.
00:06:34.041 --> 00:06:42.059
I mean, you know, when it started, the AI boom wasn't really happening, and now it is, and so everyone's sort of figuring out how those things play into each other.
00:06:42.418 --> 00:06:43.048
I thought it was great.
00:06:43.319 --> 00:06:49.869
lots of excitement in thinking about how, how the world should work and how internet speech works in, in the age of AI.
00:06:50.500 --> 00:06:50.971
Yeah.
00:06:51.040 --> 00:07:00.550
And, and for those who, who listened to the live episode that went out on the feed last week, we had a brilliant attendance for our live recording of Ctrl Alt Speech.
00:07:00.550 --> 00:07:04.478
We had You know, if, if I was Donald Trump, I'd say maybe there was millions of people.
00:07:04.548 --> 00:07:05.298
Uh,
00:07:05.370 --> 00:07:06.079
something like that.
00:07:06.079 --> 00:07:10.279
It's the largest, the largest crowd ever, I think, you know, undeniably.
00:07:10.536 --> 00:07:12.168
couldn't see the end of the crowds.
00:07:12.423 --> 00:07:16.374
but yeah, the people who came were fantastic, and we got some brilliant questions.
00:07:16.663 --> 00:07:25.767
we got some great response, and thanks to everyone who, who found us, who, came and talked to us, and actually who, who also got one of our special cards, that we were giving out at Trust Con.
00:07:26.151 --> 00:07:27.021
As promised
00:07:27.398 --> 00:07:30.038
this was We, we do deliver on Ctrl Alt Speech.
00:07:30.295 --> 00:07:37.480
the people who did come up to us, they, got a special 90% discount on the first month of our Patreon, supportership.
00:07:37.851 --> 00:07:43.630
And we're actually gonna be extending that to everyone who listens to the podcast now in celebration of the live recording going very well.
00:07:43.901 --> 00:07:56.240
So if you're listening to this and you want to you've been thinking about becoming a Patreon supporter, you want to kind of upgrade, you want to get the longer episodes and the earlier, launch date of the podcast, then now is the time to try it.
00:07:56.250 --> 00:08:08.398
You can use the discount code TRUSTCON, T-R-U-S-T-C-O-N, all caps, no space, and that will get you a basically free first month of our Patreon.
00:08:08.807 --> 00:08:09.367
And
00:08:09.656 --> 00:08:10.286
Close to free.
00:08:10.326 --> 00:08:11.526
Let's not overpromise.
00:08:11.555 --> 00:08:14.286
It is a 90% discount.
00:08:14.315 --> 00:08:21.435
I don't want, I don't want the FTC coming down on us for misrepresenting and promising free when it is not free.
00:08:21.435 --> 00:08:24.615
It is a heavily discounted month of our Patreon
00:08:24.648 --> 00:08:29.187
I, I'm sure the FTC are listening and are very concerned about our practices.
00:08:30.432 --> 00:08:31.622
The, you do not know.
00:08:33.744 --> 00:08:35.888
and that, that's, a great opportunity.
00:08:36.217 --> 00:08:46.033
We're actually g- we've got a great segment, for the episode today, really diving into how you regulate AI and maybe some missed opportunities that one particular US state has missed.
00:08:46.371 --> 00:08:49.464
Before we do that, Mike, well, let's jump into this week's stories.
00:08:49.464 --> 00:08:51.104
We've got a lot to, to go through.
00:08:51.504 --> 00:08:59.063
a lot of the conversation, as you say, at Trust Corn and elsewhere has been about AI regulation, and it seems increasingly that there is a demand for it.
00:08:59.583 --> 00:09:06.124
What this week has told me more than anything is that not necessarily everyone agrees what that looks like.
00:09:06.323 --> 00:09:09.124
And there's a few stories that kind of have brought us to that conclusion
00:09:09.595 --> 00:09:10.735
Th- there are a whole bunch of stories.
00:09:10.735 --> 00:09:16.434
I mean, uh, we're sort of entering the battlefield of what is, AI regulation going to look like.
00:09:16.434 --> 00:09:25.626
And, and, you know, Cat talked about this on the podcast a little bit last week and mentioned that an article th- that she had co-authored was, was coming out.
00:09:25.658 --> 00:09:44.725
and that is now out, and is absolutely worth reading, sort of looking at how can you, design an AI regulator properly, and it walks through the various challenges and the suggestions that some of the industry folks are making and where they could go wrong and how to correct them.
00:09:44.765 --> 00:10:00.279
And it is, you know, this is a very thoughtful piece, which is somewhat rare in this space, seeing actual thoughtful pieces on, on how to actually do this, and sort of recognizes that it is extraordinarily complicated to do well and do right, and it would be very easy to get it wrong.
00:10:00.279 --> 00:10:03.729
So I hope that more people will read this particular piece.
00:10:03.863 --> 00:10:20.436
Now, at the same time, you also had, Mark Zuckerberg sort of YOLO-ing into, to his version of, of what the regulation space should be because you know, we've had, Sam Altman give his suggestion for how AI should be regulated.
00:10:20.436 --> 00:10:23.342
We've had Dario, Amodei give his version.
00:10:23.408 --> 00:10:27.041
We had, Demis Hablis, from DeepMind give his version of it.
00:10:27.376 --> 00:10:29.229
but we hadn't really heard from Mark Zuckerberg.
00:10:29.239 --> 00:10:47.282
So, unlike the others who mostly did it on their own corporate blogs, Zuckerberg gets the op-ed pages of The Wall Street Journal, which shows a little bit of how these companies are, different, which ones are a little more established and able to commandeer the op-ed pages of The Wall Street Journal.
00:10:47.662 --> 00:11:41.589
And he gives his position of basically complaining about the other AI companies talking about doom and the end of the world and everybody losing their jobs and saying, and points out how ridiculous that is, and basically makes the argument that the best path forward is to allow AI companies to innovate as quickly as possible, not put in place all these other things that will slow, slow them down, s-sort of, you know, look sideways at the various proposals which may very well come to pass that AI models will have to get pre-vetted over a period of time by either the government or some sort of industry consortium that reviews them for safety issues, points out that, even that delay can be problematic, and then finally comes out in favor directly of allowing open weight models to be available.
00:11:41.961 --> 00:12:01.152
And this was also sort of piggybacking on a separate story, which was that late last week, there was a, an open letter that came out, I think officially put together by NVIDIA, but signed on by Most of the various AI companies.
00:12:01.591 --> 00:12:32.525
Notably absent when the letter came out was it did not include OpenAI, Anthropic, or Google, which are sort of the three biggest frontier model producers, which was notable, talking about the need to allow for open weight models and the importance of having those open competitive models to the frontier models in order to allow for more exploration, especially on the safety front, allowing people to test things, and for there to be more competition to drive innovation.
00:12:32.785 --> 00:12:42.755
It is notable that in the days immediately after this, both Google and OpenAI did join the letter and, put their names on it, but they were a few days late.
00:12:42.975 --> 00:12:50.975
And then on top of all that, and I will take a pause after this to-- because I know I'm just running through a whole bunch of these different things, but they all sort of play into this.
00:12:51.241 --> 00:13:10.869
Dario Amodei at, Anthropic put out a thing basically explaining why they refused to sign this letter, but claiming that they are supportive of open weight models, even though they are also the only company really that hasn't released one, and that, they don't think that they should be banned.
00:13:11.328 --> 00:13:23.591
Even as there have been rumors that the US government was considering banning open weight models, most of which are the, the most successful ones are coming from China, and are seen as very competitive to the, American companies.
00:13:23.922 --> 00:13:49.730
But rather, he suggests that there are other things that should be banned, and that includes distillation, of the frontier models, which is a method by which many of the open weight models are being made, which is basically, building a model, testing it against a frontier model, and adjusting it in the training process to make it make the results closer to that frontier model, and sort of a shortcut to making a model much better.
00:13:49.980 --> 00:14:02.053
And also, he suggests banning sending chips to, you know, the, the highest end chips to China because he's, you know, as he presents it, at least he's very worried about how, Chinese leadership in AI will, will develop.
00:14:02.394 --> 00:14:13.634
So there, there's a whole bunch of different pieces here of all of these companies sort of trying to jockey for position in, in how AI regulation should come about.
00:14:13.775 --> 00:14:14.206
Yeah.
00:14:14.296 --> 00:14:16.332
And, and there's a lot to get into there.
00:14:16.352 --> 00:14:35.174
I wanna start with Zuckerberg, who from the kind of way you described it and from, you know, his Wall Street Journal op-ed and his-- he's also given an interview to, the Financial Times talking about Chinese open weight models, resembles that kind of meme in which"Stone Cold" Steve Austin kind of like walks into the ring.
00:14:35.184 --> 00:14:41.804
You know, that's that kind of famous one where he's like kind of bowling it in as if, as if he kind of owns the place.
00:14:41.995 --> 00:14:44.835
That's, that's how I imagine this to, to have played out.
00:14:45.445 --> 00:14:54.620
Um, what do you make of him making this media push now, how do we think about his intervention relative to these other CEOs?
00:14:54.650 --> 00:15:02.765
Because as you mentioned, Meta is, behind in terms of development of AI models relative to those other, those companies.
00:15:03.035 --> 00:15:14.251
And he, whilst being the CEO of the world's largest social media platform, doesn't have the same kind of clout, I would say, as Amodei or Altman or Habis in, in these kind of conversations.
00:15:14.261 --> 00:15:15.792
So how do we, how do we think about that?
00:15:16.253 --> 00:15:16.643
Yeah.
00:15:16.663 --> 00:15:29.020
I mean, it is interesting because Meta is definitely far behind these other companies in terms of their AI, aspirations and, product and, and the company is now under pressure.
00:15:29.159 --> 00:15:33.169
In fact, there's been a lot of concern from investors that they're spending a ton of money.
00:15:33.429 --> 00:15:37.200
They spent a ton of money on this, uh, the whole, uh, you know,
00:15:37.384 --> 00:15:37.393
VR
00:15:37.770 --> 00:15:49.263
uh, virtual VR world stuff, which appears to have, you know, flopped in, in all sorts of ways, and now they're spending a ridiculous amount of money to not be a major player in the AI space.
00:15:49.613 --> 00:15:58.091
Now, I think, if you just take it at face value, I actually think that Zuckerberg's interview and the op-ed are not wrong.
00:15:58.110 --> 00:16:12.994
I mean, I think he actually makes a lot of reasonable and accurate points about, how open innovation works and how it should work, and the problems of concentrating power in the hands of a few, which is something, you know, I've talked about for years, and he addresses that.
00:16:13.283 --> 00:16:24.384
It's kind of funny to hear that coming from Mark Zuckerberg, because in the social media space, that's not how he acts at all, and he's very much about concentrating power and, and taking control of things.
00:16:24.638 --> 00:16:27.168
But the context here is that they're behind.
00:16:27.177 --> 00:16:34.245
You know, they're, they may be, in fifth place, arguably, maybe sixth or seventh place in the, in the AI race, and they're way behind.
00:16:34.245 --> 00:16:54.486
And so for him to say,"Yes, we need, like more faster innovation, and we don't want it to become too concentrated," is, very much self-serving for him and his position in this race, which, is certainly an important economic race, especially given how much they've spent on it to, you know, not have very much impact.
00:16:54.756 --> 00:17:01.620
So, I think he's saying things that are true, but you should definitely read it through the lens of, you know, because they're losing.
00:17:01.904 --> 00:17:02.345
Yeah.
00:17:02.524 --> 00:17:03.085
Yeah, yeah.
00:17:03.134 --> 00:17:03.784
Makes sense.
00:17:03.835 --> 00:17:12.856
The open weights conversation as well has come about because of the Hugging Face OpenAI story that we touched on in the live recording last week, right?
00:17:12.856 --> 00:17:29.893
Where, in the triage of, the kind of now famous hack that the OpenAI agents made of Hugging Face, it was not possible to kind of triage that situation without having an open weight model that didn't have guardrails and that could kind of then, be used to, decipher what had gone on.
00:17:30.280 --> 00:17:36.371
That seems to be the reason why this kind of letter from NVIDIA has come about and why there is this kind of discussion about that.
00:17:36.837 --> 00:17:48.025
how do we kind of, again, parse out whether companies are advocating for or against open weight models for their own self-interest versus for a kind of broader societal benefit?
00:17:50.171 --> 00:17:50.371
Yeah.
00:17:50.401 --> 00:17:52.861
I mean, you know, these are companies, right?
00:17:52.861 --> 00:17:56.520
They're always advocating for their own self-interest in, in some form or another.
00:17:56.877 --> 00:18:08.622
but it is an interesting discussion, and I think that the fact that almost everybody other than Anthropic eventually signed on to this letter, there is this sort of recognition that open-weight models are not gonna go away.
00:18:08.762 --> 00:18:22.693
and, I think the discussion, you- you're right that, that the discussion sort of ramped up because of the Hugging Face situation And Hugging Face's use of open models and, and Hugging Face hosts a bunch of the open weight models.
00:18:22.693 --> 00:18:24.334
That's where you can download many of them.
00:18:24.593 --> 00:18:40.920
but I think the bigger catalyst for why this was a concern is because, Anthropic in particular, but also the other leading companies have gotten worried about the quality of the Chinese open weight models in particular.
00:18:41.170 --> 00:18:52.950
And so there was this suggestion of should the US government take on this industrial policy of effectively banning Chinese open weight models in some form or another?
00:18:53.230 --> 00:18:56.980
It would be a very crude way of doing it and would almost certainly backfire.
00:18:56.980 --> 00:19:13.782
And so I think part of the letter is this acknowledgment that, look, if you wanna do this as industrial policy and try and, favor American frontier model companies, probably take a different approach than banning open weight models because that will backfire.
00:19:14.083 --> 00:19:33.573
The reality is that, you know, once the open weight models are out there and people have downloaded them and have access to them and can run them on their own hardware, claiming that you ban them is useless, and if anything, would only probably drive more people, especially outside the US, into using those models and, making them sort of core infrastructure.
00:19:33.573 --> 00:19:46.528
If you want to build something that was cheaper and almost as good as the frontier models, then, banning them, you get a sort of Streisand effect situation that would actually only, only promote them even further.
00:19:46.718 --> 00:19:53.375
I mean, we see this, you know, every few months when, when a new version of a top Chinese open weight model comes out.
00:19:53.375 --> 00:20:16.731
You know, it started whatever, a year and a half ago with the release of DeepSeek, where, like, it shook the stock market for, you know, a few days, where suddenly there was this realization that, a Chinese lab that people hadn't really heard of was, in very short order and very inexpensively, relatively inexpensively, producing a model that isn't as good as the frontier models, but was pretty close.
00:20:17.051 --> 00:20:18.561
And we're seeing that all the time.
00:20:18.612 --> 00:20:29.346
And there was, you know, just the Kimi K3 release of, once again, a model that is very good, and matches on some of the benchmarks, matches the frontier models.
00:20:29.537 --> 00:20:31.366
On others, it doesn't come as close.
00:20:31.366 --> 00:20:39.551
I don't think a- anyone really thinks it's quite as good as, you know, the top models from Anthropic or OpenAI or Google.
00:20:39.744 --> 00:20:50.957
but it, like, kind of demonstrates how close this race is and, some people had considered the open weight models were always, you know, maybe a year behind in quality of the frontier models.
00:20:51.267 --> 00:20:59.375
And I think what we're seeing is that it's much closer than that, and that has some of these companies worried about, what the impact is.
00:20:59.486 --> 00:21:14.589
And so the US government may try and do something, and the statement here is like,"Don- but don't ban the open models," because that would be a sort of very crude and probably very ineffective method of trying to regulate them and to protect the American companies
00:21:14.886 --> 00:21:23.692
And, and that, brings me to something that I've been thinking about whilst reading a lot of the commentary this week about the different models of regulation, and I definitely recommend Kat's piece.
00:21:23.712 --> 00:21:36.297
I'll also include in the show notes a piece, from the Atlantic Council which looks at a kind of slightly alternative approach to regulating, through a more federated approach that, that takes into account many more voices, which I think is something that
00:21:36.443 --> 00:21:39.894
from, from a former Ctrl-Alt-Speech guest host, so yes.
00:21:40.593 --> 00:21:42.752
We hadn't talked about it, so I wasn't sure if you'd read it, but yeah.
00:21:43.028 --> 00:21:50.425
Osse writes a really good piece about, there's an alternative way that isn't, being talked about by the CEOs of e- of these frontier labs for obvious reasons.
00:21:50.766 --> 00:22:00.175
So, so the thing I've been kind of wrangling with it whilst, doing this reading is the fact that AI is kind of a mixture of product, industry, and infrastructure.
00:22:00.339 --> 00:22:05.182
And, and the kind of, the different suggestions that the CEOs have put forward kind of speak to that.
00:22:05.509 --> 00:22:12.769
So if you look back and, and kind of assess the social media platforms, they were quite clearly intermediaries, and they could kind of be regulated as such.
00:22:13.022 --> 00:22:18.326
Whereas AI is, both being compared to electricity, or, you know, nuclear.
00:22:18.445 --> 00:22:25.209
It's being compared to, you know, the kind of defense industry, for what it, what it can lead to or what it can cause.
00:22:25.568 --> 00:22:35.623
And it's also, as you mentioned, kind of being seen in a kind of very product-focused, quite kind of jurisdictional, way where certain models can only, be used in certain contexts.
00:22:35.792 --> 00:22:39.732
Is that why it's, it's, this is such a tricky conversation to have?
00:22:40.073 --> 00:22:42.923
Because it, it's essentially all three things at once.
00:22:43.371 --> 00:22:45.320
Yeah, I think, I think that's definitely part of it.
00:22:45.320 --> 00:23:00.060
I think also people are still thinking about how to regulate or not regulate or, how they perhaps misregulated the social media, part of the internet, that that is like just sort of weighing on everybody's minds.
00:23:00.060 --> 00:23:12.230
And so the AI space is coming at it very differently And because it's a very different type of product, and where the potential harms are different kinds of harms, right?
00:23:12.240 --> 00:23:27.631
The harms that people talk about regarding social media, there's like constant debate about it because they're sort of amorphous you know, The stories that you can hear regarding AI tools, you can tell very cinematic stories about potential harms.
00:23:28.010 --> 00:23:41.030
You know, I mean, there's like the really extreme sort of science fiction versions, but there are other ones where you can see where the tools can be directly harmful, that you don't have to tell a detailed story to, to explain the harms.
00:23:41.060 --> 00:23:43.708
And that also just leads to a, different element.
00:23:44.137 --> 00:23:54.453
And then on top of that, you have the kind of different approaches each of the leading companies are taking to the market and where they sort of are, trying to find product market fit.
00:23:54.588 --> 00:24:03.560
this is an incredible space in, in terms of how quickly it has grown, how much money has gone into it, how much people are actually using these tools very, very quickly.
00:24:03.895 --> 00:24:08.064
There's lots of complaints that the tools are terrible, and they're, overhyped, and they don't live up to it.
00:24:08.324 --> 00:24:16.236
But that hasn't changed the fact that more people and more companies have adopted these tools way faster than almost every other new technology.
00:24:16.407 --> 00:24:20.127
And because of that, all the different companies are sort of trying to figure out where they slot in.
00:24:20.147 --> 00:24:31.778
And if you look at the different regulatory proposals, often you can sort of see why the story makes sense based on the sort of market niche that each of the companies thinks they're best at.
00:24:32.221 --> 00:24:42.800
and so it's, it's, you know, it's really fascinating to see, but it is a very different story than, the sort of social media questions and, and the questions about how to regulate that.
00:24:43.057 --> 00:24:53.433
and you also definitely have-- there is one, clear thing, which is that the AI space, unlike social media, doesn't have as clear a moat.
00:24:53.483 --> 00:24:55.394
It doesn't have as clear lock-in.
00:24:55.663 --> 00:25:09.971
Almost all of these tools, it's very easy to move between them, especially in the coding space, and even more so in the, in the chat space now, you can export your chats in some ways, and it's relatively straightforward to sort of move from one to the other.
00:25:09.971 --> 00:25:11.780
But there are, you know, some complications.
00:25:12.240 --> 00:25:15.523
But in the coding space, like everything is separate.
00:25:15.523 --> 00:25:26.520
Like you can drop in, all of your context and information and code are in separate standardized files stored on GitHub, and you can drop in any model that you want.
00:25:26.530 --> 00:25:28.161
There's no lock-in here.
00:25:28.461 --> 00:25:37.076
And that also just creates a very different scenario where the people within the companies recognize that they don't really have a moat.
00:25:37.426 --> 00:25:47.710
You know, they can sort of keep trying to, like, build the better model, but then you just get these like, every other week a company releases a model that's like, X percent better than the, than the other models.
00:25:47.839 --> 00:25:53.440
It's hard to build a lasting, sustainable business when you're sort of constantly fighting that.
00:25:53.440 --> 00:26:09.809
So some of this too, I think, is self-serving in the sense of like, if we can regulate in the right way, we can create some level of moat, and this is somewhat what Zuckerberg is pushing back against by saying like,"We don't want there to be just three or four giant companies who can do this.
00:26:09.839 --> 00:26:14.710
We want there to be more." But that's because he's not in the, the three or four giant companies right now.
00:26:15.557 --> 00:26:15.936
Yeah.
00:26:16.136 --> 00:26:18.646
Yeah, it's kind of interoperability by accident in a way, isn't it?
00:26:18.916 --> 00:26:37.376
Um, the other thing to note, and I think what must be keeping Zuckerberg a little bit awake at night is, is the views of some of his team and the views of the kind of Meta team as well, because they, they were part of a letter that came out this week, which people can read at pacingthefrontier.com.
00:26:37.376 --> 00:26:44.810
And th- this is a letter that's been signed by almost 1,300 employees of frontier AI companies.
00:26:45.240 --> 00:26:59.151
And it, it's a fairly remarkable letter, because it calls for the US government to kind of intervene in creating what it calls technical and governance tools to slow down the pace of development, of AI.
00:26:59.151 --> 00:27:06.233
And so y- in this letter, you've got, the, chief scientist of Meta AI, you've got the VP of AI research of, of Meta.
00:27:06.834 --> 00:27:17.881
You've got a lot of Meta folks who, are calling publicly for something that is completely, different to what Zuckerberg and his Wall Street Journal op-ed has, outlined.
00:27:18.240 --> 00:27:24.471
Which again, must make for a really interesting internal conversation, I'll say that.
00:27:24.941 --> 00:27:38.263
and we can debate and, and talk about maybe on a different episode'cause we're, we're running long here, you know, whether this in itself is a, a signaling mechanism to regulators or, to different markets about the way that AI should be regulated.
00:27:38.693 --> 00:27:52.931
But the fact that this letter was put together and signed by all of these folks working for Anthropic, OpenAI, Meta, this week when he made such a big play, his big Stone Cold Steve Austin play, is really fascinating to me
00:27:53.119 --> 00:28:04.279
Yeah, it, it is really interesting and, and it feels like this is the kind of letter that I think the other labs themselves support, but Zuckerberg's statement is definitely against this.
00:28:04.279 --> 00:28:20.779
So it is kind of interesting, you know, the-- my sort of cynical read on this letter, and we've seen other letters in the past that have sort of said,"Slow down, slow down, slow down." But, you know, uh, a lot of those have come from, from academia or from think tanks or, NGO type organizations.
00:28:20.779 --> 00:28:22.619
This is from the employees at the companies.
00:28:22.819 --> 00:28:35.487
So my, my cynical read on this is that, like, the people within these companies, especially the chief scientist type folks, are, like, working nonstop because the pace of this industry is so fast that they would like to get some sleep.
00:28:35.527 --> 00:28:46.247
And so they're sort of saying, like,"Can the government come in and slow it down so, like, I could take a vacation maybe?" But, but you know, it's, it's interesting.
00:28:46.277 --> 00:28:51.054
I always find these kinds of letters, fascinating because of what they're saying and what they're signaling.
00:28:51.247 --> 00:28:59.344
But, we live in a capitalistic society that, frankly doesn't allow companies to slow down in the way that they're hoping.
00:28:59.344 --> 00:29:08.759
So really what they're trying to do is, like, get the government to create a forcing function, because individually, none of these companies is going to agree to slow down when their competitors are not.
00:29:09.170 --> 00:29:17.874
and we're like, we're already seeing that with like, OpenAI's latest model, and because of the way Anthropic was pricing, a whole bunch of people started to switch from Anthropic to OpenAI.
00:29:17.884 --> 00:29:23.624
Like as soon as, as soon as you fall behind, people will switch because of that interoperability and how easy it is.
00:29:23.953 --> 00:29:33.586
So this is sort of a- an attempt to say like,"Hey, can we sort of get the government to come in and, give us a chance to catch our breath a little bit?" I don't know if it'll work, but it is an interesting thing to see.
00:29:33.964 --> 00:29:34.295
Yeah.
00:29:34.509 --> 00:29:44.894
yeah, for all the, kind of academics and researchers who in 2023 called for a slowing down, I sympathize with you because it, i- in hindsight, you were right.
00:29:45.085 --> 00:29:48.255
And, uh, this is, everyone catching up with that thinking.
00:29:48.464 --> 00:29:49.454
I've got more to say on that.
00:29:49.454 --> 00:29:55.875
I'll probably write a piece for Everything in Moderation about it next week, but we should move on, Mike, and, and talk about the old regulatory regime, the way, you know...
00:29:56.164 --> 00:30:01.612
we've only just started to kind of, regulate the, old, the, the Web 2.0 social media companies, you know.
00:30:01.612 --> 00:30:06.719
When-- It feels like we're, we're, now juggling t-two massive regulatory regimes at once.
00:30:06.769 --> 00:30:08.409
I'm not sure how, how we plan to do that.
00:30:08.409 --> 00:30:21.368
But I wanted to talk, talk a bit about, just some of the, music coming out of the EU recently in relation to its regulation of, social media platforms and other intermediaries under the DSA and the DMA.
00:30:21.769 --> 00:30:29.019
I wrote about this briefly in Everything in Moderation last week, but because we did our live recording on Wednesday, we didn't get to talk about it.
00:30:29.328 --> 00:30:49.335
the big news, the news that many of our listeners will have heard about is the fact that Google has been, found in breach of the Digital Markets Act and is being fined 890 million euros for self-preferencing on its search product and steering consumers on its Google Play Store.
00:30:49.689 --> 00:31:01.045
so this is a, long-term investigation that's finally kinda come to fruition, and it kind of amounts to, the biggest, investigation outcome under the DMA so far.
00:31:01.315 --> 00:31:16.358
Um, it's been much awaited and, essentially amounts to the EU calling for Google to treat consumers in a kind of fair and non-discriminatory manner Google has, has said this since the investigation has been taking place, it has started to actually kind of change its product.
00:31:16.615 --> 00:31:24.618
and it put out a short statement last week, which I interpreted as, it not necessarily intending to fight the fine, although we will see about that.
00:31:24.628 --> 00:31:29.068
A lot of other platforms are obviously appealing against European Commission judgments at the moment.
00:31:29.515 --> 00:31:54.674
but it comes after a recent flurry of, big, cases against platforms, not least the, uh, Meta investigation, which, just a couple of weeks ago came out under the DSA, which found that its products fell foul of its addictive design rules and were causing consumers to a remarkable quote,"Shift their brain to autopilot mode," um, and, you know, lead to compulsive use.
00:31:54.714 --> 00:31:58.162
Not sure how you, a regulator necessarily measures that.
00:31:58.551 --> 00:32:00.261
but that was, that was a big announcement.
00:32:00.501 --> 00:32:05.969
And in February, we saw a similar kind of, outcome of the TikTok case under the DSA.
00:32:06.439 --> 00:32:20.755
Again, similar words, all about the addictive elements of, features like autoplay and infinite scroll, and calling for TikTok to really change its product, to avoid those more kind of, attention, grabbing features that it's known for.
00:32:21.102 --> 00:32:30.558
you might at this point might say,"Hey, you know, this is right for Donald Trump to wade in." Uh, This is fodder for, for Trump and his kind of anti-EU approach.
00:32:31.075 --> 00:32:47.412
the commission has actually done a smart thing by also, fining AliExpress, a really large fine,€550 million, for not mitigating risks on its marketplace, so allowing illegal products, unsafe products, and counterfeit goods to be sold through AliExpress.
00:32:47.821 --> 00:32:54.765
And that also comes, not long after it, it did a similarly large fine, of Temu for similar practices.
00:32:55.095 --> 00:33:09.451
So the EU essentially has, I would say, lined up a bunch of its kind of DSA investigations to look like it is not going heavy on US tech companies, but is treating everyone fairly similarly.
00:33:09.759 --> 00:33:25.144
Whether or not that's going to appeal to Trump, as we kind of roll into the second half of 2026, I would doubt it, but at least we're seeing, the commission kind of, bare its teeth, I would say, or roll up its sleeves in terms of regulating companies in a way that kind of it set out to do three or four years ago.
00:33:25.556 --> 00:33:25.945
Yeah.
00:33:25.965 --> 00:33:28.605
I mean, so a few things to respond to on that.
00:33:28.605 --> 00:33:41.096
I mean, yeah, Trump has already come out and said that he wants to institute 50% tariffs on, on the EU to punish them for the Google fine in particular, which is, you know, again, like that's not, that's not how tariffs work.
00:33:41.096 --> 00:33:42.826
It's not how any of this is supposed to work.
00:33:43.165 --> 00:33:57.019
And, the particular issues or features or setups that Google was fined for are the things that they've been sued over elsewhere as well, including in the US and, they've changed a bunch of their practices in part because they were actually losing those lawsuits.
00:33:57.019 --> 00:34:07.796
So it's not necessarily an EU-specific thing, and if they choose not to fight the EU fine, it'll basically be because they've already kind of moved past those particular things, the self-preferencing.
00:34:07.905 --> 00:34:15.242
They still do some of the self-preferencing, but they've sort of been trying to, do, less of that basically because they keep getting dinged on it.
00:34:15.592 --> 00:34:22.568
Honestly, the thing I don't understand, and I've argued this for years, I, I think Google should have stopped doing the self-preferencing.
00:34:22.949 --> 00:34:34.121
They never should have done it in the first place because it was such an easy vector of attack on them, and if they made their products as good as they claim they were, they wouldn't need the self-preferencing.
00:34:34.132 --> 00:34:37.561
It would work out that way anyways because people would just use the Google anyways.
00:34:37.831 --> 00:34:42.532
The anti-steering stuff also, like, that was just stupid.
00:34:42.561 --> 00:34:43.681
I don't understand why they did it.
00:34:43.831 --> 00:34:45.442
Uh, like it makes no sense.
00:34:45.815 --> 00:34:50.856
and so they've been moving away from that and, you know, so we'll see what happens.
00:34:50.976 --> 00:34:54.686
But yeah, so like Trump reacting strongly specifically to that is funny.
00:34:55.065 --> 00:34:58.025
Now, the other point though is like, yeah, they're going after Chinese companies.
00:34:58.025 --> 00:34:59.331
They're going after American companies.
00:34:59.416 --> 00:35:05.786
Notice they're not going after European companies because there aren't any big successful European internet companies, right?
00:35:05.786 --> 00:35:06.286
So like
00:35:06.340 --> 00:35:07.141
Oh, here we go.
00:35:07.150 --> 00:35:07.550
Here we go.
00:35:10.045 --> 00:35:13.146
so I mean, there is that argument as well.
00:35:13.155 --> 00:35:18.085
It's like, you know, you haven't been able to create your own, uh, company.
00:35:18.085 --> 00:35:22.925
So that is, you know, slightly cynical take, but I, I, I'll just mention that in passing.
00:35:24.447 --> 00:35:26.067
Well, it, just tell Donald that.
00:35:26.197 --> 00:35:27.487
Just tell him it's not personal.
00:35:28.117 --> 00:35:32.067
You know, you, you, you build it, people are gonna come and try and, try and take you down.
00:35:32.206 --> 00:35:32.586
That's how it
00:35:32.831 --> 00:35:33.862
Yeah, I guess so.
00:35:33.891 --> 00:35:34.552
I, I,
00:35:34.626 --> 00:35:39.516
and I think we see that actually in the, in the story we've talked about, uh, already with, with AI regulation.
00:35:40.217 --> 00:35:43.478
and where, where's the big EU AI companies as well, right?
00:35:43.507 --> 00:35:47.987
You have Mistral, but notice how they're not, they're not a big part of this conversation either, so.
00:35:48.128 --> 00:35:48.409
No.
00:35:48.858 --> 00:35:50.148
Don't worry, the AI Act is
00:35:50.347 --> 00:35:52.137
Yeah, we'll see how that works.
00:35:52.539 --> 00:35:53.079
he says.
00:35:54.115 --> 00:36:12.498
but yeah, I think a bunch of interesting announcements by the commission recently that I th- yeah, I would say deliver on the promise of early this we- this year that it's going into full enforcement mode, and we're really starting to see that, and we, we, as you say, we're seeing the consequences from across the Atlantic as well, which is, I think, important to note.
00:36:12.782 --> 00:36:16.083
that replaces our eye-catching round-up, Mike.
00:36:16.103 --> 00:36:21.463
We'll go back next week to kind of rounding up a few of the other smaller stories in a way that is less focused on regulation.
00:36:21.710 --> 00:36:28.275
we're gonna go now to, our Patreon segment, and we talked a bit about regulating, uh, online harms and how that's difficult.
00:36:28.275 --> 00:36:31.259
We talked about regulating AI and how that's difficult.
00:36:31.438 --> 00:36:43.498
weirdly, the state of Minnesota had a bit of a roadmap for at least regulating AI, and we're gonna talk about how it let that opportunity slip in the Patreon segment.
00:36:44.003 --> 00:36:48.293
obviously this is a great chance for listeners to use their special Trust Con discount code.
00:36:48.652 --> 00:36:51.512
But give, give them a kind of flavor of what, what we're gonna be talking about.
00:36:51.750 --> 00:36:52.059
Yeah.
00:36:52.059 --> 00:37:05.056
So there, there is a big story this week that you may have seen regarding Minnesota's law against nudify apps and Elon uh, and xAI, which is now SpaceX, suing over it.
00:37:05.065 --> 00:37:18.932
And I think most of the reporting has been fairly, inaccurate and missing the point because I think there is a larger story that Minnesota had a very clear legal roadmap for how they could regulate these types of apps, and they chose to ignore it.
00:37:19.253 --> 00:37:24.282
and I think that's gonna cost the state of Minnesota in court badly.
00:37:24.628 --> 00:37:32.230
but there is an interesting discussion to be had about what would have happened if they had actually followed the accurate roadmap that a court had given them a few years ago.
00:37:32.480 --> 00:37:44.036
And so that's what we're gonna go into in the, uh, Patreon episode and hopefully, as we said, you can now use that discount code and, uh, check it out and listen to what we have to say.
00:37:44.213 --> 00:37:44.552
Super.
00:37:44.603 --> 00:37:45.032
Let's go
00:37:47.021 --> 00:37:50.302
That's the end of this week's free edition of Ctrl Alt Speech.
00:37:50.682 --> 00:37:57.561
To hear the extended episode with additional coverage, become a Patreon supporter at patreon.com/ctrlaltspeech.
00:37:57.822 --> 00:38:00.362
That's C-T-R-L Alt Speech.
00:38:00.943 --> 00:38:07.362
If your organization is interested in sponsoring the podcast, contact us via our website at ctrlaltspeech.com.
00:38:07.802 --> 00:38:08.521
Thanks for listening
00:00:00.029 --> 00:00:02.942
Mike, you're back in your familiar setting of your home office.
00:00:02.962 --> 00:00:04.232
I'm back in the UK.
00:00:04.312 --> 00:00:13.682
We're no longer at Trust Con, but that doesn't mean we haven't been perusing the internet for fun, fun things that we, you know, might take our mind off of the current moment.
00:00:14.300 --> 00:00:14.710
Yes
00:00:15.108 --> 00:00:16.562
what have you been kind of perusing this week?
00:00:16.815 --> 00:00:25.315
so mine, mine relates to a sort of pop culture reference, for people who grew up in the 1980s, which is, uh, in the US for me.
00:00:25.315 --> 00:00:30.138
So I don't know, did the, cartoon show Scooby-Doo make it to the UK?
00:00:30.138 --> 00:00:30.518
Okay.
00:00:30.847 --> 00:00:31.457
Yeah.
00:00:32.027 --> 00:00:32.747
What do you think we are?
00:00:32.747 --> 00:00:33.167
Kind of
00:00:33.258 --> 00:00:34.527
I, I don't know.
00:00:34.618 --> 00:00:42.927
The, every once in a while you hear random things and they're little cultural things that, you know, did or did not leave the US, but okay
00:00:43.317 --> 00:00:48.356
this is a bit like when Elon Musk says that London is full of pedophiles and rapists, isn't it?
00:00:49.097 --> 00:00:56.746
It-- I don't know who's, I don't know who's told you that we didn't get Scooby-Doo, but they need to, they need to course correct what they've said.
00:00:56.796 --> 00:00:57.046
That's
00:00:57.249 --> 00:00:57.749
Okay.
00:00:58.088 --> 00:00:58.558
All right.
00:00:58.558 --> 00:00:59.029
Fair enough.
00:00:59.069 --> 00:01:01.209
All right, so then you'll, you'll get this reference.
00:01:01.209 --> 00:01:01.389
The...
00:01:01.668 --> 00:01:07.784
So, in Scooby-Doo, they drove around in the Mystery Machine, the famous brightly colored van.
00:01:07.899 --> 00:01:11.671
So I found, there have been a few pe- I know this has been done.
00:01:11.671 --> 00:01:17.698
This is probably not the only one, but people have made their own exact replicas of the Mystery Van.
00:01:18.069 --> 00:01:22.349
and there is one that is currently for sale based in Oregon.
00:01:22.599 --> 00:01:27.371
It is, currently for sale for, uh, has a bid of$7,000 as we speak,
00:01:28.602 --> 00:01:29.322
Let's buy it.
00:01:29.811 --> 00:01:31.882
looks amazing.
00:01:32.266 --> 00:01:33.025
Really?
00:01:33.091 --> 00:01:34.191
Like you, you...
00:01:34.191 --> 00:01:41.064
I, I will put the link in the show notes, and you need to see this, not just for the outside, because the outside looks exactly as you expect it to.
00:01:41.194 --> 00:01:42.594
It looks like the Mystery Machine.
00:01:43.105 --> 00:01:45.305
The inside, though, that is the story.
00:01:46.078 --> 00:01:46.849
Really?
00:01:47.004 --> 00:01:48.495
have kitted out this thing.
00:01:48.745 --> 00:01:50.875
It has Scooby-Doo headrests.
00:01:51.295 --> 00:02:04.561
The entire interior of the van is in brightly colored plush stuff, and the back of the van looks like, you know, m- my childhood dream of what the back of a van would look like.
00:02:04.831 --> 00:02:07.662
It's got a giant TV, video games.
00:02:07.852 --> 00:02:10.692
There's Scooby-Doo memorabilia everywhere.
00:02:11.051 --> 00:02:11.891
Oh my God.
00:02:12.032 --> 00:02:13.072
It is astounding.
00:02:13.622 --> 00:02:14.092
Yeah.
00:02:14.391 --> 00:02:15.951
This, the Ctrl-Alt-Speech mobile.
00:02:16.876 --> 00:02:18.836
I, I think, I think it may need to be.
00:02:19.635 --> 00:02:25.175
It even, it even, it even has customized license plates that say,"Zoinks."
00:02:25.944 --> 00:02:27.974
Oh, nice touch.
00:02:28.365 --> 00:02:29.754
Really nice touch.
00:02:30.465 --> 00:02:32.305
don't know, I don't know why we're still recording.
00:02:32.395 --> 00:02:34.764
You, you should be, you should be heading straight to Oregon.
00:02:37.995 --> 00:02:38.455
Okay.
00:02:38.528 --> 00:02:38.688
you?
00:02:38.688 --> 00:02:40.149
What, what, what fun thing did you find?
00:02:40.812 --> 00:02:46.272
Well, I'm gonna ask you, about something that I'm pretty sure hasn't made its way to the US, at least not yet.
00:02:46.665 --> 00:02:47.639
what do you know about cricket?
00:02:50.135 --> 00:02:52.756
I do, I do actually know a little bit about cricket.
00:02:52.756 --> 00:02:57.742
I'm a huge baseball fan, and cricket is obviously sort of a, related sport.
00:02:57.752 --> 00:03:07.733
And at times I have spent time to figure out how cricket works, and then I remember for a few months, and then I tend to forget some of the rules as compared to baseball.
00:03:07.763 --> 00:03:10.272
But I, I'm, you know, g- generally aware of it.
00:03:10.527 --> 00:03:11.168
Okay.
00:03:11.168 --> 00:03:11.228
Okay.
00:03:11.228 --> 00:03:15.485
Well, there has been a, cricket story this week that has broken out of the UK.
00:03:15.485 --> 00:03:16.534
It's gone global.
00:03:16.544 --> 00:03:23.770
There's been reporting about this story all over the world, and that's because it concerns a very local cricket team called Saltburn.
00:03:25.288 --> 00:03:40.170
an amateur club, which has got under a bit of, gotten into hot, some hot water because it has been accused of cheating by, wait for it, clicking the fingers of one of the fielders to imitate the batter hitting the ball.
00:03:41.161 --> 00:03:52.038
And, and so what's happened is that they, this, team has done incredibly well by, having lots of catches taken by the wicket keeper, the guy who stands behind the stumps,
00:03:52.461 --> 00:03:52.872
Uh-huh
00:03:53.064 --> 00:03:55.427
which are actually not legitimate catches.
00:03:55.467 --> 00:04:02.508
They are somebody standing close by pretending clicking their fingers that the batsman has hit it.
00:04:02.737 --> 00:04:11.634
And, you know, we, we like a rule in the UK, But rules, cricket rules are even more, important to us as a nation, and this has, this has gone big.
00:04:11.694 --> 00:04:12.844
This has gone incredibly big.
00:04:12.974 --> 00:04:25.845
And it's, and it's all because, it's all because people have been looking very closely at live streams of this amateur cricket team and spotting where this one particular fielder has been clicking their fingers.
00:04:26.471 --> 00:04:26.480
Uh?
00:04:26.774 --> 00:04:35.404
we have a, a kind of entire, group of internet sleuths who've been going back through old matches and spotting who this offending
00:04:35.555 --> 00:04:37.475
this is what the internet is for
00:04:37.845 --> 00:04:38.574
Exactly.
00:04:38.605 --> 00:04:41.954
Bringing cricketers, offending cricketers to justice.
00:04:42.259 --> 00:04:42.928
There we go
00:04:44.375 --> 00:04:45.144
Let's get cracking.
00:04:53.642 --> 00:05:02.012
Hello and welcome to Ctrl Alt Speech, the podcast where we make sense of the major debates shaping online speech, platform power, content moderation, and the future of the internet.
00:05:02.452 --> 00:05:05.262
It's July the 30th, 2026, and this week.
00:05:05.322 --> 00:05:16.629
We're talking about the race to regulate AI, Europe's latest attempts to bring big tech platforms to heel, and in our special segment for Patreon listeners, how xAI is challenging lawmakers in Minnesota.
00:05:16.910 --> 00:05:25.079
Ctrl Alt Speech is the weekly coping mechanism of Mike Masnick, founder and editor of Techdirt, and me, Ben Whitelaw, founder and editor of Everything in Moderation.
00:05:25.560 --> 00:05:27.139
Mike, we're back from Trust Con.
00:05:27.449 --> 00:05:30.930
You seem like you've recovered from three intense days.
00:05:32.648 --> 00:05:34.028
Let's say, let's say that.
00:05:34.088 --> 00:05:36.411
Uh, it's, you know, it's been a lot.
00:05:36.528 --> 00:05:39.728
I can hear in the background that, that your son is happy that you're home.
00:05:41.165 --> 00:05:46.966
Yep, that is, that is, uh, that is my son who is doing well and making his presence felt.
00:05:47.894 --> 00:05:48.694
Yeah, it's good.
00:05:48.704 --> 00:05:49.223
It's good.
00:05:49.223 --> 00:05:52.153
We always, we always enjoy when, when he says hello.
00:05:53.579 --> 00:05:59.319
This is the glamour of recording a podcast on a Thursday evening in the UK right at bedtime.
00:05:59.519 --> 00:06:00.720
There's, there's nothing better.
00:06:01.139 --> 00:06:04.959
Uh, you know, you could argue that I'm getting out of it, but, yeah.
00:06:05.216 --> 00:06:08.326
let's cr- It's better to be here than it is there, I'll say that.
00:06:10.586 --> 00:06:11.665
Um, yeah.
00:06:11.665 --> 00:06:13.766
Wh- what did you come away from Trust Con thinking, Mike?
00:06:14.060 --> 00:06:14.569
What was your takeaways?
00:06:15.665 --> 00:06:16.156
It was great.
00:06:16.165 --> 00:06:16.836
It's always great.
00:06:16.855 --> 00:06:18.086
There's always lots of people to talk to.
00:06:18.086 --> 00:06:23.232
It's always good to get a, feel for what's happening and what people are thinking about and the different perspectives.
00:06:23.262 --> 00:06:29.851
Obviously, a lot of AI talk, I think, was sort of the theme in sort of thinking through sort of...
00:06:30.101 --> 00:06:34.041
this was a conference that historically focused more on other areas of the internet.
00:06:34.041 --> 00:06:42.059
I mean, you know, when it started, the AI boom wasn't really happening, and now it is, and so everyone's sort of figuring out how those things play into each other.
00:06:42.418 --> 00:06:43.048
I thought it was great.
00:06:43.319 --> 00:06:49.869
lots of excitement in thinking about how, how the world should work and how internet speech works in, in the age of AI.
00:06:50.500 --> 00:06:50.971
Yeah.
00:06:51.040 --> 00:07:00.550
And, and for those who, who listened to the live episode that went out on the feed last week, we had a brilliant attendance for our live recording of Ctrl Alt Speech.
00:07:00.550 --> 00:07:04.478
We had You know, if, if I was Donald Trump, I'd say maybe there was millions of people.
00:07:04.548 --> 00:07:05.298
Uh,
00:07:05.370 --> 00:07:06.079
something like that.
00:07:06.079 --> 00:07:10.279
It's the largest, the largest crowd ever, I think, you know, undeniably.
00:07:10.536 --> 00:07:12.168
couldn't see the end of the crowds.
00:07:12.423 --> 00:07:16.374
but yeah, the people who came were fantastic, and we got some brilliant questions.
00:07:16.663 --> 00:07:25.767
we got some great response, and thanks to everyone who, who found us, who, came and talked to us, and actually who, who also got one of our special cards, that we were giving out at Trust Con.
00:07:26.151 --> 00:07:27.021
As promised
00:07:27.398 --> 00:07:30.038
this was We, we do deliver on Ctrl Alt Speech.
00:07:30.295 --> 00:07:37.480
the people who did come up to us, they, got a special 90% discount on the first month of our Patreon, supportership.
00:07:37.851 --> 00:07:43.630
And we're actually gonna be extending that to everyone who listens to the podcast now in celebration of the live recording going very well.
00:07:43.901 --> 00:07:56.240
So if you're listening to this and you want to you've been thinking about becoming a Patreon supporter, you want to kind of upgrade, you want to get the longer episodes and the earlier, launch date of the podcast, then now is the time to try it.
00:07:56.250 --> 00:08:08.398
You can use the discount code TRUSTCON, T-R-U-S-T-C-O-N, all caps, no space, and that will get you a basically free first month of our Patreon.
00:08:08.807 --> 00:08:09.367
And
00:08:09.656 --> 00:08:10.286
Close to free.
00:08:10.326 --> 00:08:11.526
Let's not overpromise.
00:08:11.555 --> 00:08:14.286
It is a 90% discount.
00:08:14.315 --> 00:08:21.435
I don't want, I don't want the FTC coming down on us for misrepresenting and promising free when it is not free.
00:08:21.435 --> 00:08:24.615
It is a heavily discounted month of our Patreon
00:08:24.648 --> 00:08:29.187
I, I'm sure the FTC are listening and are very concerned about our practices.
00:08:30.432 --> 00:08:31.622
The, you do not know.
00:08:33.744 --> 00:08:35.888
and that, that's, a great opportunity.
00:08:36.217 --> 00:08:46.033
We're actually g- we've got a great segment, for the episode today, really diving into how you regulate AI and maybe some missed opportunities that one particular US state has missed.
00:08:46.371 --> 00:08:49.464
Before we do that, Mike, well, let's jump into this week's stories.
00:08:49.464 --> 00:08:51.104
We've got a lot to, to go through.
00:08:51.504 --> 00:08:59.063
a lot of the conversation, as you say, at Trust Corn and elsewhere has been about AI regulation, and it seems increasingly that there is a demand for it.
00:08:59.583 --> 00:09:06.124
What this week has told me more than anything is that not necessarily everyone agrees what that looks like.
00:09:06.323 --> 00:09:09.124
And there's a few stories that kind of have brought us to that conclusion
00:09:09.595 --> 00:09:10.735
Th- there are a whole bunch of stories.
00:09:10.735 --> 00:09:16.434
I mean, uh, we're sort of entering the battlefield of what is, AI regulation going to look like.
00:09:16.434 --> 00:09:25.626
And, and, you know, Cat talked about this on the podcast a little bit last week and mentioned that an article th- that she had co-authored was, was coming out.
00:09:25.658 --> 00:09:44.725
and that is now out, and is absolutely worth reading, sort of looking at how can you, design an AI regulator properly, and it walks through the various challenges and the suggestions that some of the industry folks are making and where they could go wrong and how to correct them.
00:09:44.765 --> 00:10:00.279
And it is, you know, this is a very thoughtful piece, which is somewhat rare in this space, seeing actual thoughtful pieces on, on how to actually do this, and sort of recognizes that it is extraordinarily complicated to do well and do right, and it would be very easy to get it wrong.
00:10:00.279 --> 00:10:03.729
So I hope that more people will read this particular piece.
00:10:03.863 --> 00:10:20.436
Now, at the same time, you also had, Mark Zuckerberg sort of YOLO-ing into, to his version of, of what the regulation space should be because you know, we've had, Sam Altman give his suggestion for how AI should be regulated.
00:10:20.436 --> 00:10:23.342
We've had Dario, Amodei give his version.
00:10:23.408 --> 00:10:27.041
We had, Demis Hablis, from DeepMind give his version of it.
00:10:27.376 --> 00:10:29.229
but we hadn't really heard from Mark Zuckerberg.
00:10:29.239 --> 00:10:47.282
So, unlike the others who mostly did it on their own corporate blogs, Zuckerberg gets the op-ed pages of The Wall Street Journal, which shows a little bit of how these companies are, different, which ones are a little more established and able to commandeer the op-ed pages of The Wall Street Journal.
00:10:47.662 --> 00:11:41.589
And he gives his position of basically complaining about the other AI companies talking about doom and the end of the world and everybody losing their jobs and saying, and points out how ridiculous that is, and basically makes the argument that the best path forward is to allow AI companies to innovate as quickly as possible, not put in place all these other things that will slow, slow them down, s-sort of, you know, look sideways at the various proposals which may very well come to pass that AI models will have to get pre-vetted over a period of time by either the government or some sort of industry consortium that reviews them for safety issues, points out that, even that delay can be problematic, and then finally comes out in favor directly of allowing open weight models to be available.
00:11:41.961 --> 00:12:01.152
And this was also sort of piggybacking on a separate story, which was that late last week, there was a, an open letter that came out, I think officially put together by NVIDIA, but signed on by Most of the various AI companies.
00:12:01.591 --> 00:12:32.525
Notably absent when the letter came out was it did not include OpenAI, Anthropic, or Google, which are sort of the three biggest frontier model producers, which was notable, talking about the need to allow for open weight models and the importance of having those open competitive models to the frontier models in order to allow for more exploration, especially on the safety front, allowing people to test things, and for there to be more competition to drive innovation.
00:12:32.785 --> 00:12:42.755
It is notable that in the days immediately after this, both Google and OpenAI did join the letter and, put their names on it, but they were a few days late.
00:12:42.975 --> 00:12:50.975
And then on top of all that, and I will take a pause after this to-- because I know I'm just running through a whole bunch of these different things, but they all sort of play into this.
00:12:51.241 --> 00:13:10.869
Dario Amodei at, Anthropic put out a thing basically explaining why they refused to sign this letter, but claiming that they are supportive of open weight models, even though they are also the only company really that hasn't released one, and that, they don't think that they should be banned.
00:13:11.328 --> 00:13:23.591
Even as there have been rumors that the US government was considering banning open weight models, most of which are the, the most successful ones are coming from China, and are seen as very competitive to the, American companies.
00:13:23.922 --> 00:13:49.730
But rather, he suggests that there are other things that should be banned, and that includes distillation, of the frontier models, which is a method by which many of the open weight models are being made, which is basically, building a model, testing it against a frontier model, and adjusting it in the training process to make it make the results closer to that frontier model, and sort of a shortcut to making a model much better.
00:13:49.980 --> 00:14:02.053
And also, he suggests banning sending chips to, you know, the, the highest end chips to China because he's, you know, as he presents it, at least he's very worried about how, Chinese leadership in AI will, will develop.
00:14:02.394 --> 00:14:13.634
So there, there's a whole bunch of different pieces here of all of these companies sort of trying to jockey for position in, in how AI regulation should come about.
00:14:13.775 --> 00:14:14.206
Yeah.
00:14:14.296 --> 00:14:16.332
And, and there's a lot to get into there.
00:14:16.352 --> 00:14:35.174
I wanna start with Zuckerberg, who from the kind of way you described it and from, you know, his Wall Street Journal op-ed and his-- he's also given an interview to, the Financial Times talking about Chinese open weight models, resembles that kind of meme in which"Stone Cold" Steve Austin kind of like walks into the ring.
00:14:35.184 --> 00:14:41.804
You know, that's that kind of famous one where he's like kind of bowling it in as if, as if he kind of owns the place.
00:14:41.995 --> 00:14:44.835
That's, that's how I imagine this to, to have played out.
00:14:45.445 --> 00:14:54.620
Um, what do you make of him making this media push now, how do we think about his intervention relative to these other CEOs?
00:14:54.650 --> 00:15:02.765
Because as you mentioned, Meta is, behind in terms of development of AI models relative to those other, those companies.
00:15:03.035 --> 00:15:14.251
And he, whilst being the CEO of the world's largest social media platform, doesn't have the same kind of clout, I would say, as Amodei or Altman or Habis in, in these kind of conversations.
00:15:14.261 --> 00:15:15.792
So how do we, how do we think about that?
00:15:16.253 --> 00:15:16.643
Yeah.
00:15:16.663 --> 00:15:29.020
I mean, it is interesting because Meta is definitely far behind these other companies in terms of their AI, aspirations and, product and, and the company is now under pressure.
00:15:29.159 --> 00:15:33.169
In fact, there's been a lot of concern from investors that they're spending a ton of money.
00:15:33.429 --> 00:15:37.200
They spent a ton of money on this, uh, the whole, uh, you know,
00:15:37.384 --> 00:15:37.393
VR
00:15:37.770 --> 00:15:49.263
uh, virtual VR world stuff, which appears to have, you know, flopped in, in all sorts of ways, and now they're spending a ridiculous amount of money to not be a major player in the AI space.
00:15:49.613 --> 00:15:58.091
Now, I think, if you just take it at face value, I actually think that Zuckerberg's interview and the op-ed are not wrong.
00:15:58.110 --> 00:16:12.994
I mean, I think he actually makes a lot of reasonable and accurate points about, how open innovation works and how it should work, and the problems of concentrating power in the hands of a few, which is something, you know, I've talked about for years, and he addresses that.
00:16:13.283 --> 00:16:24.384
It's kind of funny to hear that coming from Mark Zuckerberg, because in the social media space, that's not how he acts at all, and he's very much about concentrating power and, and taking control of things.
00:16:24.638 --> 00:16:27.168
But the context here is that they're behind.
00:16:27.177 --> 00:16:34.245
You know, they're, they may be, in fifth place, arguably, maybe sixth or seventh place in the, in the AI race, and they're way behind.
00:16:34.245 --> 00:16:54.486
And so for him to say,"Yes, we need, like more faster innovation, and we don't want it to become too concentrated," is, very much self-serving for him and his position in this race, which, is certainly an important economic race, especially given how much they've spent on it to, you know, not have very much impact.
00:16:54.756 --> 00:17:01.620
So, I think he's saying things that are true, but you should definitely read it through the lens of, you know, because they're losing.
00:17:01.904 --> 00:17:02.345
Yeah.
00:17:02.524 --> 00:17:03.085
Yeah, yeah.
00:17:03.134 --> 00:17:03.784
Makes sense.
00:17:03.835 --> 00:17:12.856
The open weights conversation as well has come about because of the Hugging Face OpenAI story that we touched on in the live recording last week, right?
00:17:12.856 --> 00:17:29.893
Where, in the triage of, the kind of now famous hack that the OpenAI agents made of Hugging Face, it was not possible to kind of triage that situation without having an open weight model that didn't have guardrails and that could kind of then, be used to, decipher what had gone on.
00:17:30.280 --> 00:17:36.371
That seems to be the reason why this kind of letter from NVIDIA has come about and why there is this kind of discussion about that.
00:17:36.837 --> 00:17:48.025
how do we kind of, again, parse out whether companies are advocating for or against open weight models for their own self-interest versus for a kind of broader societal benefit?
00:17:50.171 --> 00:17:50.371
Yeah.
00:17:50.401 --> 00:17:52.861
I mean, you know, these are companies, right?
00:17:52.861 --> 00:17:56.520
They're always advocating for their own self-interest in, in some form or another.
00:17:56.877 --> 00:18:08.622
but it is an interesting discussion, and I think that the fact that almost everybody other than Anthropic eventually signed on to this letter, there is this sort of recognition that open-weight models are not gonna go away.
00:18:08.762 --> 00:18:22.693
and, I think the discussion, you- you're right that, that the discussion sort of ramped up because of the Hugging Face situation And Hugging Face's use of open models and, and Hugging Face hosts a bunch of the open weight models.
00:18:22.693 --> 00:18:24.334
That's where you can download many of them.
00:18:24.593 --> 00:18:40.920
but I think the bigger catalyst for why this was a concern is because, Anthropic in particular, but also the other leading companies have gotten worried about the quality of the Chinese open weight models in particular.
00:18:41.170 --> 00:18:52.950
And so there was this suggestion of should the US government take on this industrial policy of effectively banning Chinese open weight models in some form or another?
00:18:53.230 --> 00:18:56.980
It would be a very crude way of doing it and would almost certainly backfire.
00:18:56.980 --> 00:19:13.782
And so I think part of the letter is this acknowledgment that, look, if you wanna do this as industrial policy and try and, favor American frontier model companies, probably take a different approach than banning open weight models because that will backfire.
00:19:14.083 --> 00:19:33.573
The reality is that, you know, once the open weight models are out there and people have downloaded them and have access to them and can run them on their own hardware, claiming that you ban them is useless, and if anything, would only probably drive more people, especially outside the US, into using those models and, making them sort of core infrastructure.
00:19:33.573 --> 00:19:46.528
If you want to build something that was cheaper and almost as good as the frontier models, then, banning them, you get a sort of Streisand effect situation that would actually only, only promote them even further.
00:19:46.718 --> 00:19:53.375
I mean, we see this, you know, every few months when, when a new version of a top Chinese open weight model comes out.
00:19:53.375 --> 00:20:16.731
You know, it started whatever, a year and a half ago with the release of DeepSeek, where, like, it shook the stock market for, you know, a few days, where suddenly there was this realization that, a Chinese lab that people hadn't really heard of was, in very short order and very inexpensively, relatively inexpensively, producing a model that isn't as good as the frontier models, but was pretty close.
00:20:17.051 --> 00:20:18.561
And we're seeing that all the time.
00:20:18.612 --> 00:20:29.346
And there was, you know, just the Kimi K3 release of, once again, a model that is very good, and matches on some of the benchmarks, matches the frontier models.
00:20:29.537 --> 00:20:31.366
On others, it doesn't come as close.
00:20:31.366 --> 00:20:39.551
I don't think a- anyone really thinks it's quite as good as, you know, the top models from Anthropic or OpenAI or Google.
00:20:39.744 --> 00:20:50.957
but it, like, kind of demonstrates how close this race is and, some people had considered the open weight models were always, you know, maybe a year behind in quality of the frontier models.
00:20:51.267 --> 00:20:59.375
And I think what we're seeing is that it's much closer than that, and that has some of these companies worried about, what the impact is.
00:20:59.486 --> 00:21:14.589
And so the US government may try and do something, and the statement here is like,"Don- but don't ban the open models," because that would be a sort of very crude and probably very ineffective method of trying to regulate them and to protect the American companies
00:21:14.886 --> 00:21:23.692
And, and that, brings me to something that I've been thinking about whilst reading a lot of the commentary this week about the different models of regulation, and I definitely recommend Kat's piece.
00:21:23.712 --> 00:21:36.297
I'll also include in the show notes a piece, from the Atlantic Council which looks at a kind of slightly alternative approach to regulating, through a more federated approach that, that takes into account many more voices, which I think is something that
00:21:36.443 --> 00:21:39.894
from, from a former Ctrl-Alt-Speech guest host, so yes.
00:21:40.593 --> 00:21:42.752
We hadn't talked about it, so I wasn't sure if you'd read it, but yeah.
00:21:43.028 --> 00:21:50.425
Osse writes a really good piece about, there's an alternative way that isn't, being talked about by the CEOs of e- of these frontier labs for obvious reasons.
00:21:50.766 --> 00:22:00.175
So, so the thing I've been kind of wrangling with it whilst, doing this reading is the fact that AI is kind of a mixture of product, industry, and infrastructure.
00:22:00.339 --> 00:22:05.182
And, and the kind of, the different suggestions that the CEOs have put forward kind of speak to that.
00:22:05.509 --> 00:22:12.769
So if you look back and, and kind of assess the social media platforms, they were quite clearly intermediaries, and they could kind of be regulated as such.
00:22:13.022 --> 00:22:18.326
Whereas AI is, both being compared to electricity, or, you know, nuclear.
00:22:18.445 --> 00:22:25.209
It's being compared to, you know, the kind of defense industry, for what it, what it can lead to or what it can cause.
00:22:25.568 --> 00:22:35.623
And it's also, as you mentioned, kind of being seen in a kind of very product-focused, quite kind of jurisdictional, way where certain models can only, be used in certain contexts.
00:22:35.792 --> 00:22:39.732
Is that why it's, it's, this is such a tricky conversation to have?
00:22:40.073 --> 00:22:42.923
Because it, it's essentially all three things at once.
00:22:43.371 --> 00:22:45.320
Yeah, I think, I think that's definitely part of it.
00:22:45.320 --> 00:23:00.060
I think also people are still thinking about how to regulate or not regulate or, how they perhaps misregulated the social media, part of the internet, that that is like just sort of weighing on everybody's minds.
00:23:00.060 --> 00:23:12.230
And so the AI space is coming at it very differently And because it's a very different type of product, and where the potential harms are different kinds of harms, right?
00:23:12.240 --> 00:23:27.631
The harms that people talk about regarding social media, there's like constant debate about it because they're sort of amorphous you know, The stories that you can hear regarding AI tools, you can tell very cinematic stories about potential harms.
00:23:28.010 --> 00:23:41.030
You know, I mean, there's like the really extreme sort of science fiction versions, but there are other ones where you can see where the tools can be directly harmful, that you don't have to tell a detailed story to, to explain the harms.
00:23:41.060 --> 00:23:43.708
And that also just leads to a, different element.
00:23:44.137 --> 00:23:54.453
And then on top of that, you have the kind of different approaches each of the leading companies are taking to the market and where they sort of are, trying to find product market fit.
00:23:54.588 --> 00:24:03.560
this is an incredible space in, in terms of how quickly it has grown, how much money has gone into it, how much people are actually using these tools very, very quickly.
00:24:03.895 --> 00:24:08.064
There's lots of complaints that the tools are terrible, and they're, overhyped, and they don't live up to it.
00:24:08.324 --> 00:24:16.236
But that hasn't changed the fact that more people and more companies have adopted these tools way faster than almost every other new technology.
00:24:16.407 --> 00:24:20.127
And because of that, all the different companies are sort of trying to figure out where they slot in.
00:24:20.147 --> 00:24:31.778
And if you look at the different regulatory proposals, often you can sort of see why the story makes sense based on the sort of market niche that each of the companies thinks they're best at.
00:24:32.221 --> 00:24:42.800
and so it's, it's, you know, it's really fascinating to see, but it is a very different story than, the sort of social media questions and, and the questions about how to regulate that.
00:24:43.057 --> 00:24:53.433
and you also definitely have-- there is one, clear thing, which is that the AI space, unlike social media, doesn't have as clear a moat.
00:24:53.483 --> 00:24:55.394
It doesn't have as clear lock-in.
00:24:55.663 --> 00:25:09.971
Almost all of these tools, it's very easy to move between them, especially in the coding space, and even more so in the, in the chat space now, you can export your chats in some ways, and it's relatively straightforward to sort of move from one to the other.
00:25:09.971 --> 00:25:11.780
But there are, you know, some complications.
00:25:12.240 --> 00:25:15.523
But in the coding space, like everything is separate.
00:25:15.523 --> 00:25:26.520
Like you can drop in, all of your context and information and code are in separate standardized files stored on GitHub, and you can drop in any model that you want.
00:25:26.530 --> 00:25:28.161
There's no lock-in here.
00:25:28.461 --> 00:25:37.076
And that also just creates a very different scenario where the people within the companies recognize that they don't really have a moat.
00:25:37.426 --> 00:25:47.710
You know, they can sort of keep trying to, like, build the better model, but then you just get these like, every other week a company releases a model that's like, X percent better than the, than the other models.
00:25:47.839 --> 00:25:53.440
It's hard to build a lasting, sustainable business when you're sort of constantly fighting that.
00:25:53.440 --> 00:26:09.809
So some of this too, I think, is self-serving in the sense of like, if we can regulate in the right way, we can create some level of moat, and this is somewhat what Zuckerberg is pushing back against by saying like,"We don't want there to be just three or four giant companies who can do this.
00:26:09.839 --> 00:26:14.710
We want there to be more." But that's because he's not in the, the three or four giant companies right now.
00:26:15.557 --> 00:26:15.936
Yeah.
00:26:16.136 --> 00:26:18.646
Yeah, it's kind of interoperability by accident in a way, isn't it?
00:26:18.916 --> 00:26:37.376
Um, the other thing to note, and I think what must be keeping Zuckerberg a little bit awake at night is, is the views of some of his team and the views of the kind of Meta team as well, because they, they were part of a letter that came out this week, which people can read at pacingthefrontier.com.
00:26:37.376 --> 00:26:44.810
And th- this is a letter that's been signed by almost 1,300 employees of frontier AI companies.
00:26:45.240 --> 00:26:59.151
And it, it's a fairly remarkable letter, because it calls for the US government to kind of intervene in creating what it calls technical and governance tools to slow down the pace of development, of AI.
00:26:59.151 --> 00:27:06.233
And so y- in this letter, you've got, the, chief scientist of Meta AI, you've got the VP of AI research of, of Meta.
00:27:06.834 --> 00:27:17.881
You've got a lot of Meta folks who, are calling publicly for something that is completely, different to what Zuckerberg and his Wall Street Journal op-ed has, outlined.
00:27:18.240 --> 00:27:24.471
Which again, must make for a really interesting internal conversation, I'll say that.
00:27:24.941 --> 00:27:38.263
and we can debate and, and talk about maybe on a different episode'cause we're, we're running long here, you know, whether this in itself is a, a signaling mechanism to regulators or, to different markets about the way that AI should be regulated.
00:27:38.693 --> 00:27:52.931
But the fact that this letter was put together and signed by all of these folks working for Anthropic, OpenAI, Meta, this week when he made such a big play, his big Stone Cold Steve Austin play, is really fascinating to me
00:27:53.119 --> 00:28:04.279
Yeah, it, it is really interesting and, and it feels like this is the kind of letter that I think the other labs themselves support, but Zuckerberg's statement is definitely against this.
00:28:04.279 --> 00:28:20.779
So it is kind of interesting, you know, the-- my sort of cynical read on this letter, and we've seen other letters in the past that have sort of said,"Slow down, slow down, slow down." But, you know, uh, a lot of those have come from, from academia or from think tanks or, NGO type organizations.
00:28:20.779 --> 00:28:22.619
This is from the employees at the companies.
00:28:22.819 --> 00:28:35.487
So my, my cynical read on this is that, like, the people within these companies, especially the chief scientist type folks, are, like, working nonstop because the pace of this industry is so fast that they would like to get some sleep.
00:28:35.527 --> 00:28:46.247
And so they're sort of saying, like,"Can the government come in and slow it down so, like, I could take a vacation maybe?" But, but you know, it's, it's interesting.
00:28:46.277 --> 00:28:51.054
I always find these kinds of letters, fascinating because of what they're saying and what they're signaling.
00:28:51.247 --> 00:28:59.344
But, we live in a capitalistic society that, frankly doesn't allow companies to slow down in the way that they're hoping.
00:28:59.344 --> 00:29:08.759
So really what they're trying to do is, like, get the government to create a forcing function, because individually, none of these companies is going to agree to slow down when their competitors are not.
00:29:09.170 --> 00:29:17.874
and we're like, we're already seeing that with like, OpenAI's latest model, and because of the way Anthropic was pricing, a whole bunch of people started to switch from Anthropic to OpenAI.
00:29:17.884 --> 00:29:23.624
Like as soon as, as soon as you fall behind, people will switch because of that interoperability and how easy it is.
00:29:23.953 --> 00:29:33.586
So this is sort of a- an attempt to say like,"Hey, can we sort of get the government to come in and, give us a chance to catch our breath a little bit?" I don't know if it'll work, but it is an interesting thing to see.
00:29:33.964 --> 00:29:34.295
Yeah.
00:29:34.509 --> 00:29:44.894
yeah, for all the, kind of academics and researchers who in 2023 called for a slowing down, I sympathize with you because it, i- in hindsight, you were right.
00:29:45.085 --> 00:29:48.255
And, uh, this is, everyone catching up with that thinking.
00:29:48.464 --> 00:29:49.454
I've got more to say on that.
00:29:49.454 --> 00:29:55.875
I'll probably write a piece for Everything in Moderation about it next week, but we should move on, Mike, and, and talk about the old regulatory regime, the way, you know...
00:29:56.164 --> 00:30:01.612
we've only just started to kind of, regulate the, old, the, the Web 2.0 social media companies, you know.
00:30:01.612 --> 00:30:06.719
When-- It feels like we're, we're, now juggling t-two massive regulatory regimes at once.
00:30:06.769 --> 00:30:08.409
I'm not sure how, how we plan to do that.
00:30:08.409 --> 00:30:21.368
But I wanted to talk, talk a bit about, just some of the, music coming out of the EU recently in relation to its regulation of, social media platforms and other intermediaries under the DSA and the DMA.
00:30:21.769 --> 00:30:29.019
I wrote about this briefly in Everything in Moderation last week, but because we did our live recording on Wednesday, we didn't get to talk about it.
00:30:29.328 --> 00:30:49.335
the big news, the news that many of our listeners will have heard about is the fact that Google has been, found in breach of the Digital Markets Act and is being fined 890 million euros for self-preferencing on its search product and steering consumers on its Google Play Store.
00:30:49.689 --> 00:31:01.045
so this is a, long-term investigation that's finally kinda come to fruition, and it kind of amounts to, the biggest, investigation outcome under the DMA so far.
00:31:01.315 --> 00:31:16.358
Um, it's been much awaited and, essentially amounts to the EU calling for Google to treat consumers in a kind of fair and non-discriminatory manner Google has, has said this since the investigation has been taking place, it has started to actually kind of change its product.
00:31:16.615 --> 00:31:24.618
and it put out a short statement last week, which I interpreted as, it not necessarily intending to fight the fine, although we will see about that.
00:31:24.628 --> 00:31:29.068
A lot of other platforms are obviously appealing against European Commission judgments at the moment.
00:31:29.515 --> 00:31:54.674
but it comes after a recent flurry of, big, cases against platforms, not least the, uh, Meta investigation, which, just a couple of weeks ago came out under the DSA, which found that its products fell foul of its addictive design rules and were causing consumers to a remarkable quote,"Shift their brain to autopilot mode," um, and, you know, lead to compulsive use.
00:31:54.714 --> 00:31:58.162
Not sure how you, a regulator necessarily measures that.
00:31:58.551 --> 00:32:00.261
but that was, that was a big announcement.
00:32:00.501 --> 00:32:05.969
And in February, we saw a similar kind of, outcome of the TikTok case under the DSA.
00:32:06.439 --> 00:32:20.755
Again, similar words, all about the addictive elements of, features like autoplay and infinite scroll, and calling for TikTok to really change its product, to avoid those more kind of, attention, grabbing features that it's known for.
00:32:21.102 --> 00:32:30.558
you might at this point might say,"Hey, you know, this is right for Donald Trump to wade in." Uh, This is fodder for, for Trump and his kind of anti-EU approach.
00:32:31.075 --> 00:32:47.412
the commission has actually done a smart thing by also, fining AliExpress, a really large fine,€550 million, for not mitigating risks on its marketplace, so allowing illegal products, unsafe products, and counterfeit goods to be sold through AliExpress.
00:32:47.821 --> 00:32:54.765
And that also comes, not long after it, it did a similarly large fine, of Temu for similar practices.
00:32:55.095 --> 00:33:09.451
So the EU essentially has, I would say, lined up a bunch of its kind of DSA investigations to look like it is not going heavy on US tech companies, but is treating everyone fairly similarly.
00:33:09.759 --> 00:33:25.144
Whether or not that's going to appeal to Trump, as we kind of roll into the second half of 2026, I would doubt it, but at least we're seeing, the commission kind of, bare its teeth, I would say, or roll up its sleeves in terms of regulating companies in a way that kind of it set out to do three or four years ago.
00:33:25.556 --> 00:33:25.945
Yeah.
00:33:25.965 --> 00:33:28.605
I mean, so a few things to respond to on that.
00:33:28.605 --> 00:33:41.096
I mean, yeah, Trump has already come out and said that he wants to institute 50% tariffs on, on the EU to punish them for the Google fine in particular, which is, you know, again, like that's not, that's not how tariffs work.
00:33:41.096 --> 00:33:42.826
It's not how any of this is supposed to work.
00:33:43.165 --> 00:33:57.019
And, the particular issues or features or setups that Google was fined for are the things that they've been sued over elsewhere as well, including in the US and, they've changed a bunch of their practices in part because they were actually losing those lawsuits.
00:33:57.019 --> 00:34:07.796
So it's not necessarily an EU-specific thing, and if they choose not to fight the EU fine, it'll basically be because they've already kind of moved past those particular things, the self-preferencing.
00:34:07.905 --> 00:34:15.242
They still do some of the self-preferencing, but they've sort of been trying to, do, less of that basically because they keep getting dinged on it.
00:34:15.592 --> 00:34:22.568
Honestly, the thing I don't understand, and I've argued this for years, I, I think Google should have stopped doing the self-preferencing.
00:34:22.949 --> 00:34:34.121
They never should have done it in the first place because it was such an easy vector of attack on them, and if they made their products as good as they claim they were, they wouldn't need the self-preferencing.
00:34:34.132 --> 00:34:37.561
It would work out that way anyways because people would just use the Google anyways.
00:34:37.831 --> 00:34:42.532
The anti-steering stuff also, like, that was just stupid.
00:34:42.561 --> 00:34:43.681
I don't understand why they did it.
00:34:43.831 --> 00:34:45.442
Uh, like it makes no sense.
00:34:45.815 --> 00:34:50.856
and so they've been moving away from that and, you know, so we'll see what happens.
00:34:50.976 --> 00:34:54.686
But yeah, so like Trump reacting strongly specifically to that is funny.
00:34:55.065 --> 00:34:58.025
Now, the other point though is like, yeah, they're going after Chinese companies.
00:34:58.025 --> 00:34:59.331
They're going after American companies.
00:34:59.416 --> 00:35:05.786
Notice they're not going after European companies because there aren't any big successful European internet companies, right?
00:35:05.786 --> 00:35:06.286
So like
00:35:06.340 --> 00:35:07.141
Oh, here we go.
00:35:07.150 --> 00:35:07.550
Here we go.
00:35:10.045 --> 00:35:13.146
so I mean, there is that argument as well.
00:35:13.155 --> 00:35:18.085
It's like, you know, you haven't been able to create your own, uh, company.
00:35:18.085 --> 00:35:22.925
So that is, you know, slightly cynical take, but I, I, I'll just mention that in passing.
00:35:24.447 --> 00:35:26.067
Well, it, just tell Donald that.
00:35:26.197 --> 00:35:27.487
Just tell him it's not personal.
00:35:28.117 --> 00:35:32.067
You know, you, you, you build it, people are gonna come and try and, try and take you down.
00:35:32.206 --> 00:35:32.586
That's how it
00:35:32.831 --> 00:35:33.862
Yeah, I guess so.
00:35:33.891 --> 00:35:34.552
I, I,
00:35:34.626 --> 00:35:39.516
and I think we see that actually in the, in the story we've talked about, uh, already with, with AI regulation.
00:35:40.217 --> 00:35:43.478
and where, where's the big EU AI companies as well, right?
00:35:43.507 --> 00:35:47.987
You have Mistral, but notice how they're not, they're not a big part of this conversation either, so.
00:35:48.128 --> 00:35:48.409
No.
00:35:48.858 --> 00:35:50.148
Don't worry, the AI Act is
00:35:50.347 --> 00:35:52.137
Yeah, we'll see how that works.
00:35:52.539 --> 00:35:53.079
he says.
00:35:54.115 --> 00:36:12.498
but yeah, I think a bunch of interesting announcements by the commission recently that I th- yeah, I would say deliver on the promise of early this we- this year that it's going into full enforcement mode, and we're really starting to see that, and we, we, as you say, we're seeing the consequences from across the Atlantic as well, which is, I think, important to note.
00:36:12.782 --> 00:36:16.083
that replaces our eye-catching round-up, Mike.
00:36:16.103 --> 00:36:21.463
We'll go back next week to kind of rounding up a few of the other smaller stories in a way that is less focused on regulation.
00:36:21.710 --> 00:36:28.275
we're gonna go now to, our Patreon segment, and we talked a bit about regulating, uh, online harms and how that's difficult.
00:36:28.275 --> 00:36:31.259
We talked about regulating AI and how that's difficult.
00:36:31.438 --> 00:36:43.498
weirdly, the state of Minnesota had a bit of a roadmap for at least regulating AI, and we're gonna talk about how it let that opportunity slip in the Patreon segment.
00:36:44.003 --> 00:36:48.293
obviously this is a great chance for listeners to use their special Trust Con discount code.
00:36:48.652 --> 00:36:51.512
But give, give them a kind of flavor of what, what we're gonna be talking about.
00:36:51.750 --> 00:36:52.059
Yeah.
00:36:52.059 --> 00:37:05.056
So there, there is a big story this week that you may have seen regarding Minnesota's law against nudify apps and Elon uh, and xAI, which is now SpaceX, suing over it.
00:37:05.065 --> 00:37:18.932
And I think most of the reporting has been fairly, inaccurate and missing the point because I think there is a larger story that Minnesota had a very clear legal roadmap for how they could regulate these types of apps, and they chose to ignore it.
00:37:19.253 --> 00:37:24.282
and I think that's gonna cost the state of Minnesota in court badly.
00:37:24.628 --> 00:37:32.230
but there is an interesting discussion to be had about what would have happened if they had actually followed the accurate roadmap that a court had given them a few years ago.
00:37:32.480 --> 00:37:44.036
And so that's what we're gonna go into in the, uh, Patreon episode and hopefully, as we said, you can now use that discount code and, uh, check it out and listen to what we have to say.
00:37:44.213 --> 00:37:44.552
Super.
00:37:44.603 --> 00:37:45.032
Let's go
00:37:47.021 --> 00:37:50.302
That's the end of this week's free edition of Ctrl Alt Speech.
00:37:50.682 --> 00:37:57.561
To hear the extended episode with additional coverage, become a Patreon supporter at patreon.com/ctrlaltspeech.
00:37:57.822 --> 00:38:00.362
That's C-T-R-L Alt Speech.
00:38:00.943 --> 00:38:07.362
If your organization is interested in sponsoring the podcast, contact us via our website at ctrlaltspeech.com.
00:38:07.802 --> 00:38:08.521
Thanks for listening
