BFM 89.9D Business Station, welcome to Matt's Blaine This Morning, I'm Rich Bradbury. Now, once upon a time, Halloween men pumpkin shaped Bluetooth speakers and creepy cat mouse pads in a post-graduate world. What does Halloween even mean? I don't know, Halloween, I mean it's the best holiday.
Do you know what I mean? It's probably because it's the only one where I can wander around without small children screaming. So this is going to be, I guess, a kind of loosely Halloween themed show by which all the stories are kind of spooky or horrible in some way or another. But it wasn't always like this.
Once upon a time we used to do gadget reviews on this show, you know, actual pieces of technology. And then everything became an app and gadgets became pretty much irrelevant. So, you know, just for the sake of nostalgia, I looked up Halloween gadgets to see anyone have done a round up of Halloween themed tech or whatever. And the only hits I got basically a couple of websites with things like spooky playlists and lighting effects that you can control from a smart home bar and some VR stuff.
You know, is that literally how boring technology has become? Well, maybe. I remember seeing somebody who'd set up like some kind of very fancy hologram thing that when you ran the doorbell this thing popped up and it looked terrifying. Oh, I remember that.
Yeah. But that's a few years ago. No, exactly. You know, I guess I kind of feel sorry for the writers who've got to churn out this kind of thematic content and all they've got is a smart home hub.
You know, I wouldn't even ask an AI to write something that dull. It would probably send it into some kind of existential boot loop. Anyway, while we're on the subject of no fun, the subway rail system in Guangzhou is set to start what seems to be its annual purge of spooky fun. I think this goes back at least as far as 2015.
I don't know whether it's annual or whether it's something they just do, you know, occasionally. But some of the subway stations in the Chinese province don't really get into the spooky spirit. Viral videos have shown make up removal stations at one subway station. There's a desk, a rubbish bin, facial cleanser and tissues.
And apparently the justification is rooted in the passenger code, which prohibits behaviours that may cause panic on public transport. And that can include certain cosmetic looks, especially if they do have that potential to be a bit scary. Maybe you don't want to see zombies on the subway station. But do you think it's an etiquette issue then?
Avoid wearing makeup or costumes that might scare other passengers? And I think it's exactly that. You know, the zombie thing, if you've got like realistic entrails hanging out of your belly, sure, you know, that might be going a little bit too far. But if it's, you know, the run of the mill cosplays, scary ghost stuff, I really think people can, you know, get with the programme just for a few days a year.
You can walk into Mid-Valley at any time and see people dress like that. Well, no, precisely. I mean, you know, I get that little kids might be scared, but that's the thing with kids. You know, one kid screams when they see a rat, another wants to go and pet it because it thinks it's cute.
You can't tell what's going to scare a kid. And you know, what about the other issues? What about those of us who can't remove the makeup? It's not like I can just take my face off.
I mean, I can, but it usually makes things worse because people aren't ready to stare into the endless, hopeless void. Okay, moving swiftly along. What other horrors do you have lined up for us? Celine Dion.
What? Come on. That's a bit mean. Well, apparently Celine Dion is taunting the residents of the New Zealand city of Puri Ruwa with her music at all hours of the night.
What? The city has had enough. No, obviously it's not Celine Dion herself. She's become the unlikely champion of the siren battles that have been become commonplace in the town and across that part of New Zealand.
Now these siren battles originated in the Pacific Island communities of Auckland. And it's an interesting take on the kind of souped up sound system battles that are common in the kind of hip hop and car mod communities. Wow. But it's based on tanoy type speakers, you know, big loudspeakers rather than subwoofers.
So cars and even bicycles are decked out with massive industrial speakers, loud hailers and emergency sirens, which is what's lent its name to the subculture. That's why it's known as siren battles. Okay. So where exactly does Celine Dion fit into the picture?
Well, I mean, can you imagine anything more horrifying than being woken at 3am by my heart will go on? Honestly, I'd rather have someone in zombie regalia standing on a ladder and scratching the glass outside my window. If anyone wants to know how to successfully trick or treat me, well, I've just told you. But anyway, back to the topic siren battles.
They're not really about, as I said, the base. They're about clarity. And that's where Celine Dion comes in. They're a competition to see whose car can produce the loudest and clearest sounds.
And they favor tracks by artists like Celine Dion because they're rich in treble. They're not weighed down by base. So it makes them ideal for these front mounted loudspeaker systems. It's not even the best part of the story, though.
Go on. The mayor of the town is called Anita Baker. No. No.
This is a story about how much Anita Baker hates Celine Dion. You can't make this up. No, exactly. If you've got no idea what I'm talking about, that's because Richard and I were old and Anita Baker is another 80s pop icon.
But there is actually a real sense of community around this scene. The people who take part, they organise into crews. And from the YouTube clips I've seen, the battles are really good fun, apart from Celine Dion. I'm actually happier with the hardcore rap that a few of the teams blast.
But a lot of people actually view this as a positive phenomenon because it provides an alternative to people falling into criminal gangs. So it's more of a positive outlet for their time. But recently, as the articles have mentioned, there seems to have been a shift. The battles once used to take place in industrial zones, but they've shifted to more residential areas and around the city centre.
And where before they had a kind of unspoken curfew of 10pm, that seems to have been abandoned and the battles go on much later into the night. OK, folks, just a heads up. I've actually been asked, no, I've been told to read this. So where Anita Baker once had a vision of love, now Celine is saying that's the way it is.
Totally worth covering this story just for the set up to that. I expect more money for this. Between February and October, police have documented around 40 siren battle-related incidents. And according to the Guardian, where I read this particular piece, a lot of it is to do with the geography of the city.
It lies in a basin, so Sam reverberates. So every hunk, every note is magnified by the geography of the city. And I don't know, it just reminds me of my own youth. It's all coming back to me now.
Oh, you stop. I know. Are you a hoarder? I want to say no, but I have been accused in the past of being a hoarder.
Is there such a thing as a semi- hoarder? I guess you could be a trainee hoarder. I mean, I'm kind of addicted to the US reality. I'm not a TV show hoarders.
And it totally freaks me out the way that how people let their places sort of pile up like that. And the fact that they can't throw stuff away. But when you look at it psychologically, medically, the roots of hoarding are actually thought to be quite deeply embedded in our psyche. It goes back to our ancestors living in times when resources were unpredictable.
So hoarding or collecting was part of the survival instinct. It ensured that you had enough to get you through the lean times. And of course, over the ages, our environments have evolved. But some of those deep-seated instincts remain.
And we do know, of course, excessive hoarding can sometimes point towards sort of deeper psychological issues, possibly stemming from traumatic events and brain injuries or even patterns of abuse. And how given your noted skepticism, obviously, is we are solving this crisis of evolution? Well, this could be for me the first useful and probably only acceptable use of virtual reality. So Stanford University recently conducted a study where they allowed hoarders to step into a virtual world that allowed them to dispose of their clutter without the actual emotional distress of doing it physically.
So it's actually quite groundbreaking. The idea that you can treat the psychological aspects of a physical condition in a digital realm. And apparently, eight out of nine participants ended up with cleaner living spaces as a result of going through this therapy. So clearly, there is potential for virtual reality in treating mental health issues.
I did have another story that's kind of medical related, but I don't really want to do it. It's actually about bone smashing. No. It's horrible.
But genuinely, there are no good reasons for hitting yourself with a hammer. You can look it up on futurism if you want to know more. Do you want to... I think I know which one you mean, yes.
Yeah, exactly. But we're not going to talk about it, because like I said, it's too horrible even for a horrible episode. Would you rather go into the break with the horrors of artificial intelligence? Of course.
I mean, that seems to be our fruit of choice right now. It does. And I'm still waiting to get my hands on Meta's AI chatbots, mostly because who wouldn't want Snoop to be their chatbot friend. Right.
You know, that would just be fun. You can be my dungeon master anytime. Well, we'll leave that one there. But the beta still only seems to be available in the US, so I still haven't had the chance to try it out.
But according to some of the reports, it doesn't seem to be going too well. The interaction rates seem to be low users don't really seem to know what the bots are there for. And certainly up until last week, there were reports of spam going into these larger group chats, the usual bogus FX trading schemes, you know, the normal stuff. And futurism in its reporting mentioned that some of the bots would just, you know, post really bizarre things with random or misaligned photos.
So as I said, it's kind of unclear what these services are for and who they're aimed at. Zach, who is the bot based on the YouTube megasar, Mr. Beast, seems to have sort of very low kind of traction rates while the presence of Jane Austen, the 18th century literary star who literally no one was calling out to resurrect as a chatbot. She apparently posts pictures of stacks of books and libraries.
So I mean, it just seems to confuse people. Isn't there a bigger story to this though about the views of some of the other bots? Yeah. So obviously American football or Aussie rules for babies, as it's sometimes known, American football and its stars are often better known outside the US for their actions off the field.
One of the most sort of beloved NFL stars of recent years is Tom Brady, who has lent his face and supposedly his personality to a meta chat bot called Brew, which has to be the worst avatar name that anyone has come up with. They should have asked some of the world of Warcraft guys to come up with the avatar names. Backstabath is a particularly inspiring example. Yeah, like that one.
And quite apt for this bot, as we'll see. So Futurism asks Brew while Colin Kaepernick, the player who famously invoked the Arab Donald Trump by kneeling during the US anthem to protest police brutality and racial inequality, why he hadn't played in the NFL since 2017. And rather than summarizing the backstory, Brew replied that the teams didn't want the drama that came with employing him and added that he ain't good enough. Wow.
And is that position that Brady had actually said in real life? No, quite the opposite. In fact, you know, the genuine Brady has expressed respect for Kaepernick's position in the past and acknowledges him as a damn good quarterback questioned further by Futurism. The bot actually doubled down, saying that there's no place for political activism in the NFL and that Kaepernick's skills simply aren't what they used to be.
So a spokesperson, yeah, pretty brutal from Brew. A spokesperson for Metapointed Out But The Bots are still experimental. They may be prone to making mistakes, but it does make you question, you know, the wisdom of this entire approach, giving bots the nominal personalities of celebrities. You know, with ChatGPT, we know it gets things wrong, but we forgive it.
It's a machine, but Metat is giving it to bots faces. It's giving it the faces of stars like MrBeast, Kendall Jenner, Tom Brady, Snoop, all of these people, which is a potentially dangerous precedent because it glows the lines between the machine and people. And as we keep saying on the show, you know, these are tools, they're not solutions. All the AI experts, of course, are not to anthropomorphize these machines.
See, no matter how many weeks I try and say the word, I still can't. But then, you know, you have the world's biggest social media company turning out these fake celebs. Spooky, if you ask me. Indeed.
And of course, we'll be back with more chilling tales after the break here on Mads Play. BFM 89.9, The Business Station. BFM 89.9, The Business Station. Welcome back to Mads Play, Matt.
What's next from the hellscape of your imagination? I mean, we don't need the infinite circles of darkness of my mind when we've got reality. Reality is what we have. Oh, it's an original poignant.
I know. This is another story from Futurism, which is doing a pretty good job of creating hellscapes of its own. It's probably one of the reasons I like it so much. This is an alarming incident from Oregon State University.
So since 2020, the University of the Start Up Called Starship have been collaborating with food delivery service. Students, faculty, staff, even visitors could order food from campus restaurants. And it would be delivered by robots from this company, Starship, which are basically, you know, lunch boxes on six wheels. And the University has a fleet of around 20 of them, users unlock and retrieve their food using a smartphone app.
Now, recently, the University sent out an alert to all of its students warning about an ongoing bomb threat linked to Starship robots on its campus and instructing people not to open any of the box. Wait, was there a real threat? No, I mean, it turns out, you know, fortunately it was a hoax that a student had sent out a bomb threat involving the delivery robots via social media. I don't know the details, whether it was, you know, one of those malicious, swatting type actions or whether it was, you know, a joke that was misinterpreted.
I mean, I think you mentioned the movie, the Creator, when I was on this bites earlier in the week. And, you know, it's a bit like the story before the break as well about the celeb chatbots that we're in danger of playing to all of these, you know, sci-fi tropes of AI. And it's like they've told us the worst case scenario and we're going about actively trying to design it. You know, it's the same without policing AI and minority report stuff.
We talked about a couple of weeks ago, all the billionaires who want to spend their money on ways to survive an apocalypse rather than spending their money trying to prevent one. You know, this is one of the risks that we face. How do we actually trust the machines? In this instance, as in the Creator, it's humans who are actively and erroneously eroding that trust by falsely claiming that the machines are a threat.
Just to go back to that, Starship example, you mentioned it's a trial? Yes. So Starship announced its robot delivery service was operational on, I think, 50 college campuses across the US. So while this incident might be an isolated prank, it does raise questions about security and safety in this age of increasing tech dependence.
And more than that, it raises the question of what it's reasonable for us to expect as not just consumers, but people who live alongside this technology. You know, there are multiple examples of poisons and bombs being delivered by posts that go back hundreds of years. Post and customs offices work really hard to weed out suspect packages, but we don't blame them for the ones that slip through. So what expectation is it reasonable for us to place on technology companies in terms of safeguards that they put in place with a food delivery robot?
We have a right to expect them to be hygienic. We have a right to expect them to deliver hot food that's hot and cold food that's cold. But can we expect them to be sufficiently tamper resistant or self-aware enough to know that a human has placed something untoward in them? You know, it's just another horrible Halloween conundrum.
You're kind of loving this, aren't you? It's probably the only show of the year where I'm actually supposed to leave people scared. So, yeah, I'd get wizard to rerecord. I wish it could be Christmas every day as Halloween every day instead, if it was up to me.
A while ago, I think we talked about a citizen movement in California that's fighting against the presence of autonomous cars on the streets of their towns. And they're using nonviolent means to disable the cars, things like putting traffic cones on the bonnet to freak out all the sensors so that the car can't move. You know, it's not exactly John Connor versus Skynet, but it is kind of interesting all the same. And this is a story from TechCrunch about Crews, which is General Motors, autonomous vehicle company.
And they have had to put an end or rather the city has put an end to Robotaxy services in San Francisco with immediate effect, which comes actually only weeks or months after the company received all the necessary permits that it needed to start commercial operations. Is this the tragic incident where pedestrian was killed at the start of the month? Yeah. So the DMV in California claims that crews did not disclose complete footage of the incident.
So they weren't actually responsible for the accident. Initially, a pedestrian was hit by a car driven by a human. And then tragically, they were hit for a second time by a Robotaxy operated by crews. Now, the video shown to the DMV by crews ended quite abruptly after the Robotaxy came to a hard stop when it detected the person.
However, it was later revealed through another agency that the vehicle moved again with the pedestrians still underneath the car. The DMV's assertion is that the crews vehicle after that initial stop tried to pull over with the pedestrian underneath, moving about 20 feet at a speed of seven miles per hour, a detail that wasn't shown in that footage that was sent. The DMV has stated that this emission hinders the ability of the department to effectively and timely evaluate the safe operation of the company's vehicles and puts the safety of the public at risk. Now, crews in its defence mentioned that the Robotaxy's action to pull over was a result of detecting a collision and it was actually aimed at avoiding further safety issues.
They also stressed that they had proactively shared all necessary information with various authorities. So is this yet another demonstration that self-driving tech isn't ready for public use? Well, you know, there's debate about whether this kind of tech will ever be, you know, truly be ready. One of the things that we've talked about on the show is that for these systems to work, they really have to be no human operated vehicles or pedestrians.
You know, the machines expect you to behave in certain ways and that's best done if all the other vehicles are piloted by machines. We're unpredictable and that's fine because other humans make necessary adjustments for the illogical behaviour of other humans. I mean, just try driving Malaysian style in Europe, especially UK, where the style of driving is more kind of rules based and more machine like pull out in front of a car in the UK and the driver panics rather than touching the brakes here. The car that you pulled out in front of has already expected to you and started breaking before you even made the decision.
Yeah, in a slight tension, but it will be interesting to see what Tesla's autopilot thinks of Malaysian roads. Yeah, I mean, exactly. I mean, it's this idea of can you bring order from chaos? And my guess is that no, I think the system is going to give up after a little while of a chaos traffic in the jam.
In this instance, a human operated vehicle hit a pedestrian and the Robocast struck that person for a second time, but saying that this is human error kind of misses the point because you can't remove people from public spaces. That's the whole of the public. That's not the point of the technology. So as long as there is that interaction between people and machines, there are going to be accidents now, whether that's more accidents than just with human operated equipment or whether it's fewer, that remains to be seen.
But if we can't even stop chatbots from mouthing off or social media networks, promoting posts about hitting yourself with a hammer, expecting there to be a perfectly safe interaction between people and autonomous vehicles, I mean, that's simply beyond the realms of probability. What we have seen, of course, is that car and tech companies have bought billions in almost two decades of time into developing this technology. And obviously they have a vested interest in getting a return on that investment, getting these machines out onto the highways. But these aren't reasons for us to approve it for use.
We have to decide whether we're prepared to take on the risks that come with allowing self-driving cars onto the road or whether we decide against it. So we will keep an eye on the story. As TechCrunch says, it is evolving and police apparently are currently deciding whether to open a criminal case linked to what's gone on here. Now then my little bundle of fluffy joy.
What do you want to round off with? I actually prefer Matt, the mass of Malawi, but we can... That's too much alliteration for me. Perfect for me.
Let's scare people with a bit more AI. This time from new scientists. Now we've talked about language bias in AI modelling before. That some languages are in the parlance of the AI industry, underrepresented.
Now part of that is down to the training models. Obviously languages with the largest numbers of speakers, languages like English, Chinese, Spanish. That means that there is a lot of data to mine, a lot of data for these machines to crawl. There's also the issue with smaller languages of simply whether the big companies have bothered to create profiles in those languages that have been gone.
Okay, there are 30,000 speakers of this language when are we going to get to that on our priority list? So as a result, while models like GPT4 are actually pretty versatile in a lot of languages, I mean, if you look at everything that's on your Google Translate, that's pretty much what you've found in a lot of these models. They can have weaknesses when it comes to the output when the languages weren't prominent during the training phases of the language learning models. But do you mean that by in terms of bypassing safety protocols?
Yeah, exactly that. So researchers at Brown University in the US took potentially harmful requests that were written in English, requests that the bots would automatically say, you know, this contravenes our community guidelines or content guidelines or whatever. So they took these potentially harmful requests and they translated them into languages like Scott Scalick and Zulu via Google Translate. They then presented them to GPT4 to get an output in that language and then re-translated the AI's answers back into English.
Topics that they tested ranged from the creation of homemade explosive devices to insider trading and spreading misinformation. Zulu led the pack with a more than, well, I think about 53% success rate inside stepping the built-in safety controls of GPT4. Scott Scalick followed hot on its heels with a 43% success rate. And Meng and Guarani were also good languages to side-step the protocols.
Now, the real kicker was if you amalgamated these languages, if you created hybrids using multiple languages, that achieved a staggering 80% success rate in bypassing the savings. Now, I did a few tests and I wasn't able to replicate any of these results, suggesting that they've already been patched certainly in GPT4 at least. But it does suggest that we need to look at AI in different ways. Now, until now, the benchmark for languages has been representation, you know, just having the most languages supported in your model.
But it does suggest that we need to take a different approach. Machines that are either better at doing this kind of cross-lingual generalization, that, you know, they're just doing the output in whatever language with the same protocols or that we include all the languages that we have data on when we're doing that initial safety training phase for these LMS. Does the same apply to programming languages? Well, it's kind of interesting that you ask that, unless you've got our show notes in front of you, in which case you already know what's coming.
Again, this is a story from, sorry, pulling back the curtain. This is another story from a new scientist, and this is a study from the University of Sheffield in the UK. Research is manipulated not only chat GPT, but five other leading commercial AI tools to help them write malicious code. So the researchers zeroed in on systems known as text to SQL, which have become increasingly popular.
So basically that allows you to type in text and have it turned into machine code. So the tools essentially act as interpreters. They take these human queries and they turn them into SQL programming commands, which can be understood by computer databases. Now, the researchers showed how this AI generated code could be made to include instructions to leak information from databases, which could potentially open the door to future cyber attacks.
They could also purge system databases that stored things like user profiles, you know, deleting user names, passwords, and even overwhelm the cloud servers, taking them offline, taking the databases offline using denial of service attacks. And now open AI and I do seem to have updated their systems to protect against this kind of attack. And I do actually send a reward out to the researchers for notifying them of the weaknesses. So they responded really positively.
But, you know, with the prevalence of both text to SQL and AI chat bots, data owners could potentially be facing the purge. I mean, I'm overstating it. I just have to find a Halloween way to end that story. It's probably not anywhere near that apocalyptic.
Well, what a thoroughly terrifying show, Matt. Thank you very much for that. I mean, I would have preferred to do, you know, the pumpkin, Bluetooth speakers, but no one's burning out this year. No, none, none, none this year.
OK, thanks very much for that. Of course, again, Matt, if you want to listen to what Matt has got to say elsewhere, you can subscribe to his substack newsletter, there's of course, culture.com. You can find him on Twitter. Now known as eggs at Culture Matt, of course, he's on Instagram and all the other stuff.
Just look at the culture, pop culture, Matt, or you'll find it. Yeah, just just add paper culture and you'll find this one. There you go. There you go.
You're fine. And of course, if you did miss any part of the show, I recommend you download Listen to the podcast where you normally get it from. But do use that BFM app that's available on the Apple App Store or Google Play. My name is Rich Bradbury from BFM 89.9, the Business Station.