Tech News Weekly 396 Transcript

Please be advised this transcript is AI-generated and may not be word for word. Time codes refer to the approximate times in the ad-supported version of the show.

0:00:00 - Mikah Sargent
Coming up on Tech News Weekly. Emily Forlini is here. We talk about Neuralink's biggest competitor, then Beeper's new messaging app under the auspices of Automatic, before I talk about OpenAI's breaking new chat GPT agent and, round things out with 404 Media's, Emanuel Maiberg, who discusses Hugging Face, hosting 5,000 non-consensual AI models. All of that coming up on Tech News Weekly. Stay tuned.

0:00:39 - Mikah Sargent
This is Tech News Weekly episode 396. With Emily Forlini and me, Micah Sargent, Recorded Thursday, July 17th 2025. OpenAI's new chat GPT agent is here. Hello and welcome to Tech News Weekly, the show where every week, we talk to and about the people making and breaking the tech news. I am your host, Micah Sargent, and I am joined this week by the Emily Forlini. Hello, Emily.

0:01:11 - Emily Forlini
Hello, how's it going?

0:01:13 - Mikah Sargent
It's going well. Right now you're giving this very ethereal sort of I've just walked into Rivendell kind of vibe.

0:01:26 - Emily Forlini
Yes, and I think it's pretty great. Wow, that's exactly. I've been working my whole life to get to that point. Definitely had an Orlando Bloom like Legolas poster in my childhood bedroom, like life size. So, yeah, best compliment you could give me, maybe Legolas is here.

0:01:41 - Mikah Sargent
Maybe he is Well. For anyone who is tuning in for the first time, this is the part of the show where we share our stories of the week. These are stories that, in some cases, we have written and in other cases as it is in my case stories that I've read, that I think are interesting and that we want to talk about with all of you. Emily, would you like to kick us off this week?

0:02:06 - Emily Forlini
Sure, we're going to start with mine. So this week I want to talk about a big feature that I came out with a couple of days ago, and I was basically one of the only reporters to go inside the New York headquarters of a big Neuralink competitor. So Neuralink is Elon Musk's brain implant company, and so there are actually a lot of other companies and this one's called Synchron and it is, I believe, a little further ahead of Neuralink, so its implant is in 10 people and Neuralink is in seven. So, yeah, just really fascinating look at how implants work and the fact that maybe we'll all get one someday, and it was just very mind-blowing.

0:02:48 - Mikah Sargent
Mind-blowing. Very good, very good. So, yeah, I'm curious to hear, I think, sometimes something that gets missed, because it's been my experience that journalists who are very focused on the journalism which is a good thing but because of that they end up being a little bit sort of not coy per se, but they don't want to talk about themselves and the experience of it, and so this is where I get to go. You went into this place and you were the only journalist to get to do that. Like what? What was that like going in? What? How did they? Did they give you a tour? I mean, were you able to look at different models of of? Did you get to talk with any? Those are the kinds of things I think are really fascinating that we sometimes miss in these stories.

0:03:42 - Emily Forlini
Sure, yeah. So I didn't even go a step back. So they contacted me after I wrote an article about how Apple is enabling people, with their implant, to control iPads and iPhones. So it's basically a Bluetooth connection that Apple has built for all called brain computer interfaces, bcis so for all BCIs, and this company, synchron, is the first one to do it. So people who have the implant can, you know, bluetooth connect to Apple devices and poke around on them with their thoughts.

So I wrote an article about that. They contacted me and I was like, oh, I'm located in New York area, like I'd love to come meet you guys. And they were just like sure, and we just turns out. I don't know if many journalists have even asked that question, but it was really interesting because they don't post their address publicly and they kind of like had to get clearance internally to get somebody to send me an email with the address. And then they gave me the address. But then I got there in Brooklyn and I was wandering around for like 15, 20 minutes in this big area of all kinds of like wacky startups or like smokestacks and there was a steel door with 10 locks all around the door and also 10 handles Wow.

So I was just like what is this place? And I'm like looking around for this like brain implant startup. And then I just happened to walk into one of the buildings that kind of looked like it might have a security guard, and then my colleague, our photographer, was there already, so I guess he picked the same building. And then, also serendipitously, a Synchron employee was in the lobby going up to work and she's like, oh yeah, I'll take you guys up. And on the way up she's like, oh yeah, yeah, they don't really give out the address. And oh yeah, did you notice? The visitor's pass email actually has two other addresses that are inaccurate at the bottom for a different part of Brooklyn. And I was like what?

0:05:33 - Mikah Sargent
Goodness, you chose the right one.

0:05:35 - Emily Forlini
Right, I think I ignored those, but it was just a very mysterious experience. And then we got to the top of the elevator. We went inside. It looked kind of startup-y, like trendy plants, an open kitchen. We could hear people cleaning dishes and stuff. There was open desks and then like a big Florida ceiling windows overlooking Manhattan and we just kind of like sat on this leather couch right next to a giant brain model and just waited for them to come and grab us. And then we basically went into the CEO's office. We talked to him, then we had a demo of the tech from the COO and then we just kind of like talked to some employees. We pet their adorable golden retriever so innocent compared to the topic of brain implants, the poor, poor little guy.

0:06:23 - Mikah Sargent
You mean the retriever didn't have a brain implant.

0:06:26 - Emily Forlini
I know I was looking at him like oh did they implant you, can, you work?

0:06:30 - Emanuel Maiberg
with an iPad, yeah.

0:06:31 - Emily Forlini
Wow. So, that was kind of like the experiential aspect.

0:06:37 - Mikah Sargent
And I mean this is still just I mean you said kind of mind-blowing, mind-boggling it is all of these things in terms of forming this connection In your time, kind of working with them and talking to them. Did they discuss the realistic state of the technology? I think that, yeah, especially whenever you're working with a company that, on the other hand, is a Musk company, you get these sort of grandiose proclamations and you have to take those things with a grain of salt. Did you find that there was more realism here, or is it still kind of in that vein of? Did you find that there was more realism here, or is it still kind of in that vein of tech?

0:07:32 - Emily Forlini
optimism that borders on, perhaps, idealism. What was that experience like? Yes, you're asking such great questions. So the CEO was more realistic and he brought up real issues with if everybody had a brain implant which, by the way, he expects to make kids available to anyone by the 2040s and so if anyone could have a brain implant, what would that mean? Would that give the company who makes them too much control over your literal thoughts and actions and what you're doing on these tech devices? And so he was open to discussing things like bias, discrimination, even subversive control like evil companies Like. He was willing to talk about these ethical issues, which is what I'm interested in. So I like that Also comes from a medical background, so he was very interested in just hype, I guess hype. Yeah, he hyped their implantation method quite a bit because it doesn't actually drill into the skull like Neuralink. It inserts a little snake up your body through a vein up to top of your brain, so it sits the top of your brain in a large vein that is, I guess, close enough to get the right electrical signal. So we talked a lot about that.

And then there was I thought the COO was a little more of like a sales guy and he his confidence was dripping from the walls. He was like there's no way it's not going to work. We think Neuralink is over-engineering. I could not get him to crack. I tried to poke so many holes. But he really is selling the thing. So it the confidence levels were very high. But, um, I did interview a patient after and I also interviewed a Neuralink patient and I could compare them and the Neuralink one is a little more powerful, but it does have that trade-off of it's extremely invasive and it's drilling into your skull and putting it on the brain. So it depends who the patient is like, how much they need. And I do agree with Synchron that probably the least invasive thing typically wins out in medicine.

0:09:29 - Mikah Sargent
Yeah, I mean with laparoscopic surgeries and things like that, overtaking in as many cases as possible the more complicated ones. I mean it's often a cost-saving metric, if nothing else, and we know that if there's one thing the healthcare industry cares about, it's saving money. So I'm not surprised to hear that in that way, you kind of are going oh, we may have found a gold mine here, at least a vein of gold here, a gold vein? Oh, a gold vein. One of the other aspects of this that I was curious about you said Neuralink is more powerful. When you say that is that according to the two patients that you interviewed and from your own understanding of what each could do, you've determined that Neuralink is more powerful. Is that even what the companies say, that Neuralink is more powerful because it has a more direct connection? Where did that more powerful come from? And then also in that, what does it mean in terms of being more powerful?

0:10:30 - Emily Forlini
Yeah, so that does come from the company itself. So their main technical challenge is achieving the same performance as Neuralink, because they're just one layer removed physically like a tissue layer above.

And then I formed my own opinion just by talking to the two patients. So the first one this is guy Rodney in Australia. He has a, a neural or a synchronic implant. He can't talk or move, so he uses his implant to control um devices. But he can't control the mouse, so he has to use eye tracking and then once his eyes fall on what he wants to press, he uses his implant to select it. And the way he does that, which is really crazy, is he kind of they program the how do I even put this? Like they almost make it like a hotkey. So when he wants to press select, he thinks move my foot. Oh, got it my foot, so that's his that's his signal.

It's like they can. It could be like, uh, move my pinky. It could be like scratch my back, I don't know. It could be just like another. It's related to a body part and so their brain sends signals. So it's almost like a hotkey, like a shortcut that you program.

So, his is move my foot for select, so he can, like you know, browse news articles. He can send texts, but he has to, you know, like move his eye to the letter and select it. So the Neuralink patient, who is a younger guy and does not have a degenerative disease he was paralyzed after a diving accident, which is really unfortunate, so he's 31. He can't move, but he can speak, so he can dictate, which is really helpful, and then with his Neuralink he can also move the cursor, which is a huge benefit. You don't have to use eye tracking, which is exhausting, and then he too has aspects of his body that are programmed to take actions. Does that make sense?

0:12:20 - Mikah Sargent
Yeah, absolutely no, that makes perfect sense.

0:12:22 - Emily Forlini
Yeah, it's different things different inputs like eye tracking, dictation but then both of them had the same thing when they want to use their implant, they have to think about a body part.

0:12:30 - Mikah Sargent
Yeah, I'm sure that when you're looking at signals that your brain sends, there are probably ones that are easier to discover and separate from the pack.

0:12:41 - Emily Forlini
Exactly Everyone has good and bad ones and, depending on if you, have a disease that's affected a certain part of your brain, or you don't have a disease or just your own biology. So it's super hands-on with. The companies like Neuralink and Synchron have worked with these patients to program their pathways and they're always optimizing them and yeah, just really tiny like single digits. People have these implants and none of them are available on the market. They're not FDA approved. So this is really kind of like a renegade behind the scenes thing. That's a very, very active space, which is what I learned.

0:13:15 - Mikah Sargent
Now there's one part in your article of a photograph of a silver briefcase with foam on the inside very spy moment, which is really kind of neat. Obviously they're operating in New York, right, but are they implanting in the US? Is there a reason why the guy came from Australia as opposed to like what's, what's that part of it? Do you know?

0:13:50 - Emily Forlini
The Australian connection, I'm guessing, is that the CEO is Australian and there are like universities there and just activities. I mean I don't know exactly how he was selected, but there there is a connection there. So he's Australian. He came to the U? S at a pretty young age and found that the US had more research funding for this area and he thought it was very commendable and he was able to connect with someone actually in the army and kind of discuss his idea and it's kind of grown from there. So I'm not sure if that Australian patient came to the US or what, but that's the Australia connection. It's not like just any country anywhere.

0:14:27 - Mikah Sargent
Understood Anything else that stuck out to you or that you think is something of interest in this that you want to mention before we head into a break.

0:14:41 - Emily Forlini
Yeah, I think.

Just one last thing is we think of this space as kind of for people with disabilities, and it very much is at this point.

But, to be clear, like the ambition for all these companies is to make it for everybody, and you can see that the writing is on the wall a little bit. Maybe in like 50, 100 years, you know, think about, we've gone from cell phones, which are kind of annoying and cumbersome to use, right, we're like hunched over, we're getting carpal tunnel, like that's not the best experience possible. So now people are going to glasses like oh, you can sit up, you can talk to an AI. You know, we're experimenting with these different form factors. So the ultimate seamless form factor that'll give you the best posture is probably a brain implant. So it's like you could see how the trend is going there. And then, with Apple exploring them, nvidia is also partnering with Synchron to train AIs on brainwaves, which is crazy. So it's very long-term, but I left with an impression that there would be appetite for this if it worked properly, and that very powerful people and companies are wanting to go in that direction.

0:15:55 - Mikah Sargent
Understood. Yeah, we look at the accessibility features in our tech already, and many of those features that first existed as accessibility-only features end up working their way into everybody using them.

And in some cases, there are features in that portion that people use regularly, and so I'm not surprised that this is another place where you know it could be open to everybody eventually, so very cool. Everyone should, of course, head over to read the article. We'll link it in the show notes. All right, we are going to take a quick break. Before we come back with my story of the week.

I want to tell you about Smarty, who is bringing you this episode of Tech News Weekly. Discover what's possible when address data works for you.

Smarty is revolutionizing how you handle address information, bringing automation, speed and accuracy to processes that used to be manual, error-prone and well frustrating. With Smarty's cloud-based address validation APIs, you can instantly check and correct addresses in real time. No more bad data compliance risks, undeliverable mail or those costly delays. Add autocomplete to your web forms so your customers select valid, verified addresses as they type. This will improve their user experience and yield much better data for you. Companies like Fabletics have drastically increased conversion rates for new customers, especially internationally, with Smarty. Want more than just clean addresses. Well, Smarty's Property Data API unlocks 350 plus insights on every address, from square footage to tax history, automatically enriching your database. Also, it's incredibly fast 25,000 plus addresses per second and very easy to integrate. The Red Cross needed accurate address data to allocate resources. A project manager says, quote the Smarty tool has been fundamental. I've never experienced any issues with the tools and they seem to be getting better all the time. The address verification really does make an impact. We're able to reach the communities we serve because we have good addresses. Smarty is a 2025 award winner across many G2 categories, like best results, best usability, users most likely to recommend and high performer for small business. Smarty is also USPS CASS and SOC 2 certified and HIPAA compliant.

Whether you're building your first form or modernizing an entire platform, Smarty gives you the tools to do it smarter. Try it yourself. Get 1,000 free lookups when you sign up for a 42-day free trial. Visit smarty.com/twit to learn more. That's smarty.com/twit and I love that 42-day free trial. I did ask the folks at Smarty where'd you get that number from and they said, yeah, it absolutely is because of Douglas Adams, and I said I knew I liked you people, so check out that 42-day free trial. Thank you, Smarty.

Let's head back to the show. All right, we're talking about a saga at this point the story of Beeper, the ambitious all-in-one messaging app. It's a compelling case study in the power dynamics of the tech industry, because what started as a promising solution to messaging fragmentation quickly turned into a high-stakes cat-and-mouse game with Apple over its coveted iMessage platform. Now, after a contentious battle that caught the attention of lawmakers and left some users locked out of their own Macs, beeper is moving forward with a new privacy-focused model that may change the game for good. So you may remember because I know we talked about it on this show, dear listeners the kind of conflict that happened. It was Pebble smartwatch creator, Eric Majikowski, who I actually had on the show when this was all going on and talked to him about how they had made Beeper Mini, and Beeper Mini was a form of Beeper, which is this multi-platform messaging app that would give iMessage access to people who didn't have Apple devices. They essentially figured out the way that Apple was kind of signing up new phone numbers and new accounts so that you could message with other iMessage users, and while it was there for a little bit and cost money and I think that was kind of part of the problem is that Beeper was charging for this feature. Charging for this feature, apple figured out how Beeper was pulling this off and quickly patched the issue. Beeper then issued a fix, and then Apple once again figured out a way to stop the company from doing so.

Now, when this happened, I remember me myself kind of going oh, this is a little bit cringe, because Apple and other big tech companies have been at the sort of center target for a lot of lawmakers, and in particular, when it comes to sort of anti-competitive practices practices, and you know, apple says that the real reason for this is that, for for stopping, it has to do with the fact that, a, they were breaking terms and conditions by doing so, but B, it was a security concern, et cetera, et cetera, et cetera, and that it was, uh, end up.

It ended up kind of bogging down the iMessage platform, because many, many, many, many many more users, in theory, were joining it, and the big concern there is that iMessage would then be used for sending spam.

When that happened, though, the US Department of Justice investigated Apple's treatment of Beeper Mini and we saw the EU also looking into it and all of that kind of came to a head and resulted in Beeper Mini just stopped.

They just said, okay, we're not going to do this anymore. Well, beeper was acquired by Automatic, that's, with two Ts, automatic and that company, of course, is responsible for WordPress and now, at this point, many other online services and platforms, and in that acquisition, the company has worked to recreate Beeper in a way that is still providing for communication between iMessage users, but only if you have an account and an actual Apple device. So it's sort of following the rules, but it works with many different services all at once, and the idea is that you, as someone who is using this service, you as someone who is using this service, have direct messages from Instagram, and you have direct messages on X, and you have messages on LinkedIn and you have all of these different things, and so what this does is it brings those all into one place Slack as well and lets you communicate in one place with all these different platforms. I like the idea. I wanted to hear your thoughts on on a you know, one messaging app to rule them all situation.

0:23:22 - Emily Forlini
Ooh, another Lord of the Rings reference for our episode. Ooh, micah talk dirty to me. Okay. Messaging apps though yeah, it's good. It's, I think, very user-friendly. You know, you don't have to go check messages everywhere, like sometimes, you know, I forget someone messaged me on Instagram, but then I also have my like WhatsApp groups and I have my text messages. I mean, you know, maybe someone from Facebook marketplace responded back about something I want. I mean, there's messages.

0:23:52 - Mikah Sargent
Finally buying you that stupid table. I have that I don't need anymore.

0:23:56 - Emily Forlini
Stupid table from some weird person who's just reselling it from a junkyard. Ah, they need me to respond. So it's like there's just messages everywhere and I do feel like streamlining them is great, but of course, it does take away the branding of the platform and you want to be as engaged with the platform I branding of the platform and you want to be as engaged with the platform.

0:24:14 - Mikah Sargent
I don't have to scroll Instagram, check an Instagram message Like that's pretty cool for me, not for Instagram. Yeah, so that's the big thing, right? That has been the issue in the past with Beeper. If you are no longer a daily active user because you're coming from somewhere else, that's going to be a problem. Now, according to Beeper's current team, they say that they have good relationships with some of the companies that they're pulling data from. They have no relationship with others and they are working toward having relationships with some. So it's very much kind of all over the place. But one thing that they've done, at least, or attempted to do, is to honor the different platforms kind of UI. So it's kind of hard to describe, but it's much easier to just give an example, which is that they say if Telegram is displaying an ad in your messages, then Beeper will also display that ad in your messages.

0:25:23 - Emily Forlini
That's respectful. I don't know that we have that much sympathy for these companies. You know that they're pulling messages from. Mostly, we're all just kind of stressed out and overloaded with all of the communication so it's hard to side with Instagram or Facebook or Apple. In this case, Like I just want to respond to my people and not worry about it.

0:25:43 - Mikah Sargent
Absolutely, and I think that's the difference there right Between the user experience and the experience that I want versus being able to provide an experience by having to play ball with what the companies want, and it's that balancing act that's difficult had a pretty active subset of users who were using like most power user people of Twitter were using third-party clients that offered more tools, granular control, et cetera, et cetera, and because they were coming from there and their most engaged users were coming from those platforms, they weren't able to serve up the ads and make the money off of them, and we saw changes to the API over time and that ended up kind of for me, I ended up moving back to the Twitter app, because when Twitter would launch new features like polls and other stuff, it would take a long time, if ever, to come to those third-party services. So they figured out a way to kind of funnel people back in and I think that, as much as something like this is really cool, it is still, ultimately, as much as I don't care about what the company wants. If the company is not getting what it wants, then I don't see this being a long lasting service, and that's my concern, because what if I start to really like it and rely on it. And then you know, instagram decides well, meta decides no, we're not doing that. And Slack goes well yeah, no, we're not doing that either.

There's also originally the concern and this is where this new version kind of comes into play when you did this in the before time, before it was acquired and rebuilt, you would have to connect to Beeper's kind of cloud service. So what would happen is you would log into your Instagram account and then one of their Mac minis or something that they had in a server cloud elsewhere in the world would create an instance and log into your Instagram account and then it would kind of mirror it back to you. But that meant that Beeper, the company, had access to your Instagram account and could see everything, and in that way you were sort of going I have to trust that Beeper is going to do right by my information Plus, I also ran into issues where when I would log in, it would say a computer in some other country has just logged into your account. Are you sure you're okay with that and I'm going? Well, I think that was beeper. But what if I hit yes and then it wasn't beep, and so there was kind of that fear as well. That's where they've improved it now in its new rendition, where it basically, instead of going out to some server and some Mac mini cloud somewhere, it just uses your computer locally. So your computer is the mirror itself and that means that it's going to maintain that end-to-end encryption that you would not otherwise have. So I think that that answers the question of, or answers the problem that some privacy advocates had with it, answers the problem that some privacy advocates had with it.

But ultimately I do wonder what the size of this market actually is and how many people are even aware that this is a possibility. In the first place I wasn't. And then, yeah, and then want to go into the work of of building all of this out, for the free account gives you five logins to five different accounts, uh, the. There's like a plus account that gives you access to like reminders, but also, I think, up to 10, uh accounts, and then there's like a plus plus version that gives you unlimited access, and they are pushing they're pushing that toward kind of companies that like a social media manager or something like that. So, yeah, it's a lot to do to get all of your messages in one place and the times when I've used this. There's this service, beeper, and then there was a service called textscom that was actually also bought by Automatic and the new version, beeper. And then there was a service called Textscom that was actually also bought by Automatic and the new version of Beeper is the Frankenstein project of the two brought together and I used both of them and they both were just clunky enough that I didn't stick to it.

Plus, it pairs with what you mentioned about feeling a little overwhelmed suddenly now, where I did not know I had 15 messages on LinkedIn because I never look at LinkedIn. Now I know I've got 15 messages there and all of the Instagram messages that I get that are both real and are spam and are everything in between. It's not sorted the same way, so I don't have it sorted by, like these are probably fake. So I now have where I thought before I had those two messages from my friends that I keep marking as unread because I'll definitely get back to it a little bit and respond to them. Now I, instead of two, it's like 15, 30, 60 different messages that I've got to respond to because I decided to log into everything and look, I'm doing little anxiety movements right now just thinking about it. So yeah, it's kind of a lot to burden oneself with in the first place.

0:31:25 - Emily Forlini
Interesting. I didn't think about that. And there's LinkedIn in particular. Just so many bogus messages. Oh my goodness, like they give people premium subscriptions. They could just message anyone they want. It's like get out of my inbox, like I don't need an MBA, like I do get those advertisements.

0:31:43 - Mikah Sargent
Oh yeah, absolutely, what the?

0:31:45 - Emily Forlini
heck is that there's just so many bogus messages. And you're right, like I don't.

0:31:48 - Mikah Sargent
I don't want that next to my text to my friends but yeah, it kind of takes you out of maybe the good conversations that you're having. Ultimately, is it just that we have adapted to having our messages spread out across multiple places? Believe that I'm that sort of optimized tech individual who could, who would want to have inbox zero across all of my platforms. But then the second I start to actually try it, I go, oh no, I do not want this at all. This is horrible. Uh, keep those. I am an inbox zero person.

0:32:28 - Emily Forlini
So maybe that's why it appeals to me, because I'm kind of doing that anyway and, yeah, if I could just have it all in one place. I don't, I don't know I feel like you should try it. I should try it, I think. I think it's a good idea, I think it's almost like a power to the people moment, like you don't need to get sucked in and distracted by all these tech platforms all the time. You can kind of prioritize.

0:32:51 - Mikah Sargent
But but you do, you raise an interesting point that actually could just make your overall tech experience just worse so yeah, well, uh, I as is my way, as much as I'm saying, you know, this is horrible, or this, this would give me anxiety, or whatever. I probably will, because I want to try these different services. I'll probably try it out just to see how they've integrated the two different services, and then I figure I will probably also go okay, no, this is still not for me, but hey, they've pulled it off, or they haven't pulled it off. I think that's the big question, right Of. Will they get it right this time in a way that appeals to more users, such that it is going to stick around? That's the big question.

0:33:41 - Emily Forlini
Well, they have a good name Beeper.

0:33:43 - Mikah Sargent
Beeper is a great name, that's a retro name. Yeah, so it gets the people Call me beat me if you want to reach me If you want to reach me.

0:33:51 - Emily Forlini
There's a lot of good stuff associated with the word beeper. So I think their mission is cool, but it sounds like they've got to clean it up. They got to get some VC money and pour some gas on it.

0:34:02 - Mikah Sargent
Make it a little nicer, make it a unicorn or all those terms, all right. Well, emily, I want to thank you so much for taking the time to join me today. If people would like to keep up with what you're doing, where are the places they should go to do that?

0:34:16 - Emily Forlini
Yeah, so you can find me. I mean, my work is on PCMag. I'm active, I think, mostly on Blue Sky these days, so my social is just at Emily Forlini and anything, even LinkedIn. You can send me a LinkedIn, just don't offer me an MBA. So I'm pretty available. I also have a new podcast going with another tech journalist. It's called the super intelligent podcast and it's kind of like this but maybe a little more meandering, philosophical, not as new space, but if you just want to chat about tech, come hang out with me and my cohost, mike Elgin, and it's called super intelligent podcast.

0:34:53 - Mikah Sargent
Beautiful. Thank you so much for joining us today and we will see you again soon.

0:34:57 - Emily Forlini
Okay, thank you, micah.

0:34:59 - Mikah Sargent
All righty folks, it is time for another little break before we come back with the rest of the show.

I want to tell you, this time about ThreatLocker, who's bringing you this episode of Tech News Weekly. We know ransomware is harming businesses worldwide, but ThreatLocker can prevent you from becoming the next victim. ThreatLocker's zero-trust platform takes a proactive deny-by-default approach that blocks every unauthorized action, protecting you from both known and unknown threats. It's trusted by global enterprises like JetBlue and Port of Vancouver. ThreatLocker will shield you from zero-day exploits and supply chain attacks, while providing complete audit trails for compliance.

As more cyber criminals turn to malvertising, you need more than just traditional security tools. Attackers are creating convincing fake websites impersonating popular brands like AI tools and software applications, distributed through social media ads and hijacked accounts. Then they use legitimate ad networks to deliver malware, affecting anyone who browses on work systems. Traditional security tools often miss these attacks because they use file-less payloads that run in memory and exploit trusted services that bypass typical filters. Well, ThreatLocker's innovative ring-fencing technology strengthens endpoint defense by controlling what applications and scripts can access or execute containing potential threats, even if malicious ads successfully reach the device. ThreatLocker works across all industries, supports Mac environments, provides 24-7 US-based support and enables comprehensive visibility and control. Jack Senesap, director of IT Infrastructure and Security at Redner's Market, says when it comes to ThreatLocker, the team stands by their product. ThreatLocker's onboarding phase was a very good experience and they were very hands-on. ThreatLocker was able to help me and guide me to where I am in our environment today. Get unprecedented protection quickly, easily and cost-effectively with ThreatLocker.

Visit Visit threatlocker.com/twit to get a free 30-day trial and learn more about how ThreatLocker can help mitigate unknown threats and ensure compliance. That's Visit threatlocker.com/twit, and we thank ThreatLocker for sponsoring this week's episode of Tech News Weekly.

All righty, we are back from the break and we've got a quick little story for you, because it is a breaking news story. As I was gearing up for this episode, imagine telling your computer to plan a dinner party or summarize your unread emails, or build a presentation from a spreadsheet, and it just does it. That's the vision OpenAI is inching toward with the launch of its new ChatGPT agent, a general purpose AI assistant that can operate your computer on your behalf. It's built on a new model and bolstered with safety features. This tool combines web browsing, file access and even terminal commands to take autonomous action, not just generate text. It's OpenAI's most ambitious step yet into the age of agentic AI. It's a very familiar term at this point. Agentic AI is the future and this is ChatGPT's take on it. It's going to do more Click, scroll, download, analyze, generate and even automate these tasks, like sending emails or building slide decks.

This morning, as I was gearing up for the show, openai was having its live stream about its ChatGPT agent. There were a few publications who gained access to and got like a presentation of the new agent. The Verge describes it as a tool that can complete work on your behalf using its own virtual computer, and it uses OpeningEye's previous tools, including Operator, which is that browser automation tool, and Deep Research as well, which is the multi-site synthesis tool and is now available in agent mode for ChatGPT Pro Plus and Team users. So some of the things that OpenAI says it can do review your calendar and prep you for meetings, where, instead of having to copy and paste all of the meetings into OpenAI's chat GPT and say, can you give me a rundown? No, it will just go and do that for you. Create editable presentations from scratch. Summarize documents shared from Gmail or Google Drive using the connectors. Feature plan and book dinners. Compare product reviews, automate weekly tasks. One example was an OpenAI employee is using it to automatically process parking requests.

It works through ChatGPT's natural language interface so you can type commands like slash agent or just activate the agent mode in the chat. And one thing to understand is that it is a little bit slower than your other chat GPT prompts, even if this is from product lead Isa Fulford, who says even if it takes 15 minutes, half an hour, it's quite a big speed up. So 15 minutes to 30 minutes, that's a lot of time, but in terms of what they saw when they first were working on this to now, a lot faster. They, of course, benchmarked it 41.6% on Humanity's last exam, which is double the 03 and 04 mini scores. 27.4% on Frontier Math with tool access. That's 6.3% for 04 mini. So quite a jump there. And it is built with reinforcement learning and trained on workflows, specifically where you need to chain tools like browsers, terminals and data importers together.

The one thing that they talked about I thought was interesting during the live stream are the risks of kind of stop not stopgap measures, but sort of measures in place to mitigate risk because of the fact that it has high capability. If it can do all of these things on your behalf, if it can run terminal commands, that's a concern to have. So the agent pauses before doing anything irreversible, like purchases or emails, and asks for user confirmation. If the user leaves the tab during a sensitive task, like finance, well then the agent will stop. It's limited to specific get requests. There's no open access and terminal. There's no long-term data retention. As of now, it could be that that changes over time and some of the things that they really have worked on in its training, of course, prompt injection, defense, sensitive task confirmation, privacy aware behaviors. So it does not exfiltrate or infer private information. And in doing so, the company says we are taking a precautionary approach and implementing safeguards consistent with high capability models, which is an interesting term that they're using, or phrase is an interesting term that they're using, or phrase I should say that they're using for these models.

Of course, this isn't the only group that's working on this. We've seen Google, we've seen Perplexity, we've seen Anthropic. Even Klarna have been building agentic systems. That's kind of the new buzzword. Klarna's AI agent handled the work of 700 human agents in a single month, which led to enterprise agent development, so having these agents work on the behalf of many enterprises. But the difference here with OpenAI's agent, where agentic models up to this point have kind of been task-focused that was part of the idea. This is a specific task that this agent goes and does and this agent goes and does that, this agent goes and does that and then it all comes together and kind of gives you the answer. It's this combined access to apps, to terminals, and this novel model that's been trained specifically for this, so that it's all working together to provide that experience that maybe people have been thinking about when they think about AI working for this purpose.

Of course, the question becomes can AI agents ever fully handle complex human workflows with the security, privacy implications and the need for some human interaction? It's those handoff moments, right, the times where the agent steps up to the plate but you have to swing the bat at the plate to actually hit the ball. Will it ever get there? Can it safely get there? Can we do it in a way that we trust? How do we trust these agents with partial autonomy? Because the first time one goes and buys something that I did not want it to buy, or books, I don't know a plane ticket instead of a ride share and costs me a bunch of money. It's going to be a long time before I ever want to use that thing again and ever trust that again. Plus, no, even in a sandboxed environment, giving AI tools full computer access has its own set of concerns and trust issues that I know I would have and that I imagine many people would have. I can imagine the Security Now crowd being like. This is absolutely nothing we would ever do. So that is kind of something that we're bearing in mind and, because this was just announced today, openai doesn't have the full suite of all the information about it, so we'll be keeping a look on that. So, yes, we'll include a link in the show notes to the Verge's coverage and, I believe, techcrunch's coverage of this new tool, and I'm sure we'll be keeping an even closer eye on it as well. All right, with that, it is time to move on to our final story and this an interview. All right, I'm very excited for this next conversation Joining us from 404 Media, Emanuel Maiberg.

Welcome to the show, Emmanuel. Hey, thanks for having me, absolutely so. I saw your piece over on 404 Media titled Hugging Face is Hosting 5,000 Non-Consensual AI Models of Real People. I was hoping to start. Could you please explain what these AI models are and, I think, interestingly, what first led you to investigate them there on Hugging Face.

0:45:45 - Emanuel Maiberg
Yeah, so it's a bit of a story. I'll try to get through it quick. But there's this website called Civitai and it's a website where people share custom AI image generation models. They use different base models, mostly stable diffusion, but basically you take the basic stable diffusion model and then you're able to train it to produce a certain kind of image very well, better than the general model could. So you could feed it a couple of image of, let's say, a certain type of anime you like and then it gets really proficient at generating images that look a lot like that anime. And you can do the same thing for people. So I'm not a celebrity. There isn't a bunch of images of me on the internet, Like there is a Taylor Swift. So if you type Emanuel Myberg into stable diffusion, it doesn't really generate anything that looks like me, but you can train a model to do that. And Civitai is a place where people share models that are trained to generate specific styles and then also a lot of celebrities. As you might imagine, the internet being what it is, a lot of people use those models to create non-consensual porn of those celebrities.

I've reported on this platform for almost two years now and there's been a lot of developments. They've been caught doing a lot of bad stuff, but ultimately they received some pressure from payment processors about the type of content that was hosted on their website and because of that they made the decision to remove all models that are designed to recreate the likeness of real people. And I was talking to a researcher who was tracking all the data that's been uploaded to Civitai. So this is both AI-generated images and also these models, and this researcher saw that overnight, once Civitai made the decision to remove these models, she tracked more than 50,000 models that are designed to recreate the likeness of celebrities, YouTubers, Instagram influencers, like big celebrities, but also like pretty minor internet personalities, and also some like regular people. I found a few cases of that happening. All of those were removed overnight and, for reasons we can get into, I was able to see that some of those models were then re-uploaded to Hugging Faith.

0:48:39 - Mikah Sargent
Okay, yeah, so you report that the models that were previously banned from Civit AI were kind of migrated over, and the way that this was done, I think, is it tells us a little bit about sort of the mindset behind it, because you say that they were kind of uploaded or that they were uploaded with generic names to hide the true purpose. So can you talk about a finding them in that way and then how the people behind the effort are kind of organizing and then the people that are attempting to find these models that were removed from the one platform? If they're uploaded with generic names, how is anyone finding what they're looking for?

0:49:28 - Emanuel Maiberg
Right, yeah, good question. So Civitai has a really huge community. It's a very large platform with many users and there was some lead up to Civitize decision to remove these models and the community felt that there were some minor policy changes that weren't as strict that preceded this decision and there was sort of a feeling in the air that something was coming. And once people felt that there was an effort by the community to start migrating the models to other places on the internet so they could still use them and some people in the community started a Discord channel and they started downloading models and started finding other places to host them. Once Civitai announced that the decision has been made to remove all models of real people, that effort really kicked into high gear and people started grabbing as much stuff as they can before the models were completely removed from Civitai.

And from what I've seen now, the primary way of sharing these models is they take the model, which previously was named after a celebrity, very clearly in the URL, in the file name, in the model card, which is like the little bit of text that describes how the model works and what it's designed to do kind of strip all that identifying and, you know, arguably incriminating information and upload it very generically to Hugging Face, just calling it test model or LoRa, which is a type of way that people modify the models. So that's how the models appear on Hugging Face Elsewhere. The community created a bespoke website that is easily searchable. You can search for celebrity names. You can search for the Civitai models that used to be hosted on Civitai and that you used previously but is no longer there. You can use a certain type of hash that Civitai used to identify all its models.

And that will take you to a page that has all that information, the name of the celebrity sample images and then at the bottom links to where it is hosted on Hugging Face. So on Hugging Face there's nothing really incriminating or identifying about the models, but then there's a third-party website that is kind of used as a map or a search tool in order to find all the models via celebrity name or whatever it is that you want to make.

0:52:16 - Mikah Sargent
Understood, okay, so I would argue a very clever method for doing that. We should talk about Hugging Face itself, because this is a major multi-billion dollar platform in the AI space which, I'll be honest in researching a little bit more about this, did not realize the valuation for Hugging Face was so high. What do its content policies actually say about this kind of activity, and what was the company's response when you reached out for comment?

0:52:52 - Emanuel Maiberg
Yeah, so Hugging Face is a huge company. It has emerged as I don't know if I would say the default, but the most visible place or the most prominent place where people share AI models, tools, resources. It's just a huge platform for this entire generative AI machine learning revolution that we're in, and they have this huge multi-million dollar contract with Amazon and they've taken on millions in investment and it's a very useful platform used across the industry by big companies, by individuals, open source community. It's just like a very valuable resource and platform and, as any other platform that is open in this way, they have, like, the moderation challenge in front of them and I would say that, unlike some of the other big AI companies, hugging Faith definitely puts itself forward as a company that really cares about ethics and putting people in front of profit or in front of AI for AI's sake, and they have a pretty robust ethics team and ethics statement and they have all these different efforts to make the AI industry better and I think that's all really admirable In terms of the policy and I understand why they would do this. It is not super specific and it is not it doesn't have any language against these type of models, right?

So, going back to Civitai, the policy there, for example, was back in the day or like just before they made this policy change is people could upload media and AI models that recreate the likeness of real people. Separately, people could upload adult content and models that generated all content. What they were not allowed to do is kind of combine those two things together. You weren't allowed to create non-consensual adult content of real people. The issue, as my reporting shows, is that people did that anyway, sometimes on platform and most definitely off platform. They used the Civitai platform in order to make that content. No policy that I could find that doesn't allow people to take images of me from social media and train a model that generates images of Emanuel Myberg. What they have is kind of like vague language about like AI should always have consent and people are now allowed to do things that are criminal, which would include, you know, adult images of sexual images of underage people.

But there's and they also have a strict policy about the same thing that Civitai had, which is like no non-consensual sexual content, but the models themselves that are uploaded, most of the time they're not explicitly sexual. It's just that they're able to recreate the likeness of a real person and it's very easy to use that in order to create non-consensual sexual content, and that's what people use it for. I've seen that over and over again in my reporting. Like that's exactly what people use the models for. But if you just look at the model, it doesn't explicitly say that this is what it does. So, technically, there's no policy that I can find that restricts this kind of content. It's just that it's clear that it's being used to cause harm. And it's really clear because these are models that were specifically removed from, like, an AI porn site at the behest of payment processors that told Civitai we're not gonna do business with you if you continue to host this stuff. And now this stuff lives on a platform that has, you know, way more visibility, way more investment, way higher valuation.

0:57:25 - Mikah Sargent
Yeah, I mean that gray area that exists there gives that freedom to kind of appear as though that is an important consideration and meaning that you still are following your rules, but that this stuff can be used in this way. I was curious to hear, beyond the platforms themselves, what did your investigation kind of reveal about the community that is organizing the effort to preserve and share these models, and it sounds like you've touched on what seems to be the motivation for sharing and preserving these models. But do they suggest there's anything other than just creating non-consensual porn?

0:58:09 - Emanuel Maiberg
essentially, face say in response to this article. They didn't say anything. They did not respond. I approached them in multiple ways, multiple times. I've talked to them before about other stories that I've written and they responded to me and they just chose not to respond here. We still haven't heard from them on this issue. Just wanted to close the loop on that.

To your other question about the community.

You know there's sort of like a tongue-in-cheek argument that people in this community sometimes make, which is, you know, freedom of speech and satire and creative freedom, and sometimes there's like a minority of people that sometimes make that argument.

I think at this point most of them are pretty clear about what they do with these models. And again, I think my reporting shows that if you go to communities on Telegram where a lot of this content originates Telegram, where a lot of this content originates Telegram, 4chan, some Discord channels I would say Discord does a better job of policing this stuff. But if you look at the actual online spaces where this non-consensual content originates and they talk about, they share information, right, like, oh, here's how you can do this, here's how you can do that Civitai and these models, right that now live on Hugging your Face, they come up over and over and over again. People create like non-consensual content of a YouTuber who's not that big Maybe she has a few thousand followers and somebody asks like how did you make that? Inevitably, it always leads back to like a Civitai model.

1:00:02 - Mikah Sargent
Yeah, Wow, I'm really glad that you are not just covering this topic, but that you are paying close enough attention to see sort of the transition from one platform to the other and the way that people are doing this, so that, going forward, people are paying more attention to that being done. Of course, people can head over to 404 Media to check out this article and many others. Is that the best place for people to go if they want to keep up to date with what you are writing?

1:00:38 - Emanuel Maiberg
Yeah, please. Much like Twit we're read if supported 44mediaco. It's our subscribers that allow us to write these investigations. Would really, really love people to check us out and consider supporting us.

1:00:52 - Mikah Sargent
Awesome. Thank you so much for your time today. We appreciate it. All right, thank you so much, all righty folks. That brings us to the end of this episode of Tech News Weekly.

The show publishes every Thursday at twit.tv/tnw. That's where you go to subscribe to the show in audio and video formats. This is my time to remind you all about our awesome Club Twit at twit.tv/clubtwit. When you join the club and you are, you know, one of the supporters, then you gain access to some awesome benefits. All of our content ad-free. It's just the content, none of the ads. You also gain access to our Twit+ feeds that include our Club Twit events, our Club Twit shows, news events as well, and clips and bits from our different shows. All of that is available to you. So the first day that you sign up, you're going to have this huge back catalog of great stuff to check out, which we love. You will also gain access to our Discord, a fun place to go to chat with your fellow Club Twit members, and those of us here at Twit Love to see you there. We're always seeing new people pop up as they join the club, and I look forward to seeing you join us there as well.

So again, twit.tv/clubwit. We've got monthly and yearly plans and we appreciate your support. If you would like to follow me online, I'm at Micah Sargent on many, a social media network where you can do chihuahuacoffee that's C-H-I-H-U-A-H-U-Acoffee, where I've got links to the places I'm most active online. Be sure to check out later today iOS Today and Hands on Apple, and you can always check out on Sundays, my show Hands on Tech, where I answer your tech questions in depth. Thanks so much for tuning in and I'll catch you again next week for another episode of Tech News Weekly. Bye-bye.

1:02:42 - Leo Laporte
From Silicon Valley boardrooms to tomorrow's AI breakthroughs. If you need to keep up to date with tech, you need twit.tv. At Twit, we're tech experts who understand what's happening and can keep you in the loop, and we do it in a thoughtful, informed and fun way. Start your Sundays with this Week in Tech, a roundtable of tech journalists and people in the know with a rundown of the week's most important tech news. But that's not all. All week long, you can stay ahead of security threats with Steve Gibson and Security.

Now Keep up on all things Apple with MacBreak Weekly. Listen to the most informed Microsoft experts in the world with Paul Theriot and Richard Campbell on Windows Weekly. Our flagship shows feature tech's most respected voices, giving you the insight you need to understand and benefit from the changing world of technology. We give you analysis you won't find anywhere else, so you can make smarter business decisions and take advantage of the technologies transforming your world. Whether you're a CEO, it professional or simply passionate about technology, twit's network of shows gives you the edge you need in today's digital landscape. Don't miss a minute. Subscribe to twit.tv today.