HN Reader

This is cool when some independent hacker / artist does it as "Lemmy".

When a big telecom does it, the second thing they do with it is to fuck up the spam detection so bad that every third phone call I make gets answered by "Daisy".

And just think about it - why would a telecom need this tech? They can already drop the spam calls and stop routing calls from the bad actor telecoms who enable the spammers. They don't do that because they prefer to collect a few cents a call from them rather than serve their customers better. It's everyone else who needs this.

The scam and spam call problem is really bad in Germany to this day. And has been for 10 years.

A couple years ago I would sit at my desk thinking about a really hard problem in silence. The phone rings. Spam call. Every 30-180 minutes another one. If you now think turn the phone off, well not that easy as CEO of a business when people expect you to be reachable.

It creamed my corn so much that I recorded my own voice samples as a senile "Opa Denny" (german grandpa Denny), modelled after Lenny. Complete with background ducks hanging out on the couch to Opas dismay, later in the call. It works on autopilot without interaction because on Asterisk, and with the largest German SIP provider at least, you can extract the calling peer identity from the SIP header. So I wrote a scoring system based on indicated number, black and whitelist regexs for number and for calling peer, greylist for the geographically surrounding number prefixes, etc. A legit mobile call would show up as number@t-mobile.de for example, while a spam call would say fakenumber@01012.com.

Asterisk would record the call in wideband stereo, normalize the audio, and mail it to me as MP3 attachment. Funny for a while, but these days I just throw all such calls onto the mailbox. Since they need a real person to scam or create a sale, the call is finished right away.

It works great to this day, because I never published it.

Having dealt with O2's support line, I wouldn't be surprised to discover they'd deployed a time-wasting AI there too.

Very nice to see, my grandmother was recently scammed out of a large amount of money. Luckily the bank reimbursed her.

Scammers are a stain on the reputation of India. You could argue it's unfair to tar an entire country with the same brush, but quite clearly rule of law isn't properly functioning over there and there's complicity in letting them do this. Same goes for Nigeria.

I was a little surprise to read that they're using speech-to-text and text-to-speech rather than an end-to-end speech model. Won't that horrible latency? (I guess the old-person persona disguises it a little...)

Jolly Roger has been doing this for a while, the example phone calls are hilarious. https://youtube.com/@jollyrogertelephone?si=sB3Tiql0wsUlXldo

Good on them!

> and one in five (22%) experiencing a fraud attempt every single week

That seems low. I guess they mean phone calls only.

I get -no exaggeration- several hundred spam mails a day (and these are the ones that made it past the first line defense). I also get 10 or more scam texts.

Some of the phising emails that I get, are frighteningly realistic (but invariably seem to have at least one speling eror in them).

If folks live in a normal suburban development, they are highly unlikely to get several hundred crooks, knocking on their door, every day, but being on the Internet, means exactly that.

The crooks only need to land you once, you have to have a perfect record, avoiding them.

If this is just a PR stunt, imagine what they're cooking up behind the scenes: AI reps to answer actual customer calls. No more paying a horde of $10/hr meat slaves in some call center! You know some execs creamed themselves over this already.

Reminded me of Lenny and he didn't even need any AI to waste a lot of scammers' time.

I default to never answering now. Every time I feel like it might be an actual person on the other end and risk answering, it has turned out to be a scammer or spam call. Not sure why governments don't do more against them - scam and spam callers have destroyed phone communication.

It's the year 2035. 80% of all energy is consumed by AI scammers talking on the phone with AI Grammas.

Another solution - drop numeric phone numbers all together and switch to alphanumeric or verified contact only. With numeric, sequential phone numbers you can just robocall all the numbers until you find a victim. Making the search space significantly larger should solve that attack vector. Of course, this will effectively be the same as transitioning from IPv4 to v6 - with all the same associated pain

It'll help with so many things: - in contact syncing systems you can't rainbow table your way to decrypting numbers - numbers can be permanently burned once they're released or deemed as spam. This means every service could ban spammers safely without fear of burning a real user. - people could more easily have alt numbers, non-voip numbers for untrusted services.

Lenny was the OG https://www.youtube.com/watch?v=vWrkDOt_IfM

I'm imagining this is just a publicity stunt, and I'll say it's a very good one. However I can't see it being very practical. There are lots of scam calls to keep up with and LLMs and text-to-speech models are expensive to run. If they do run this in production, the costs of running hundreds of 'Daisies' will inevitably be passed onto the consumer, and worse still, if the scammers are calling in through PSTN lines or cellular this will use up our already scarce bandwidth. I've frequently had difficulty connecting through trunk lines from Belgium and Germany to numbers in Britain, and that's without a legion of AI grannies sitting on the phone!

Whenever I go home to visit my elderly parents, I'm amazed at how many scam calls they get. Like at least 4 or 5 an hour. And they pick up each one because they're from a time when you simply pick up your phone when it rings, and they won't change their habits.

It's long past time that phone calls should stop being synchronous and immediate. They are only this way because historic landlines were synchronous and immediate, and nobody changed it. This functionality should be culturally unacceptable in 2024.

Imagine if the concept of a land line never existed. Then, a startup came out with an app that: 1. allowed random people to cause your handheld device to stop what it was doing and immediately ring and vibrate and pop up a dialog over whatever app was active, 2. when you pushed a button, an instant two way audio channel was initiated, and 3. they could do all this without any authentication whatsoever--just needing to know a non-secret 10 digit number. This kind of app would fail every App Store guideline, and it would be laughed out of the ecosystem as unacceptable to end users. But in reality, it's "The Phone App" that comes pre-installed on your damn phone!! Insane that this kind of application is acceptable.

Instating an AI model as "Head of Scammer Relations" is hilarious. I think the point here is to collect the phone numbers of scam call centers and have Daisy call them so that scammers waste time talking to it rather than a victim.

I applied for a mortgage today through my bank. Before the automated “next steps” email could hit my inbox my phone was already ringing and ringing and ringing. About 10 minutes straight of automated spam callers that are watching who knows what public ledger on now.

I guess it pretty much means retiring this number next year. I’ve inly had the number a little over a year and the spam has slowly crept up despite best efforts to give fake numbers where possible.

Seems stupid to publicly announce this product so that scammers are now aware of it? Also what happens when scammers start using AIs of their own to do the scamming?

KitBoga, scam baiter par excellence, did this in 2021[0].

[0] https://www.bbc.co.uk/news/av/technology-56458267

Scammers can jam this by reporting legitimate phone numbers as scam numbers, no?

Isn't text-to-speech very expensive? Last time I checked, Eleven Labs, the most natural-sounding TTS provider, were charging insane amounts per token. Wasting so much on scammers sounds unsustainable and just silly.

I can already see AI scammers phoning AI anti-scam agents.

Or we could just make these calls illegal?

I really enjoy the Google Pixel exclusive call screener.

https://youtu.be/V2IyttWHJfs?si=mreFXk6h-yDAJ3D7&t=50

I've had a few friends get new numbers, and seeing "Hey it's your Aunt, I got a new phone" live transcribed on the screen has come in handy.

If spammers like this are in another country, their country should be cut off from the global phone and internet system or we should simply drop bombs on their call centers.

This reminds me of Dead Internet Theory. We have a phone network of AI scammers endlessly calling AI counter scammers. What a time to be alive.

What's the best API for real-time conversations?

Strangely, I looked at OpenAI's offering.

Despite it being crazy expensive, they also didn't offer a reference implementation anywhere close to the functionality that is demo'ed in the regular ChatGPT app.

Nobody seems to get the interruptions correct - people are doing all sorts of weak workarounds like push-to-talk, etc.

This reminds me of “shadow banning” - instead of letting the misbehaving user know they’ve been banned, and let them find a way around it, you make them believe they’re not banned and let them waste their time interacting with the system (without actually interacting with others), this makes them spend less time on actual misuse and it makes the penalty for it more expensive. Good strategy. Cruel too.

So I don’t think that this is just entertaining PR, I can see why it’s better than simply banning the scammers. Still a question of cost though.

This scratches an itch, sure, but the people manning the phones for a scammer are likely to be in a pretty desperate situation themselves. They may or may not be free to leave, etc. I'd like to find a way to help them roll over on their bosses and find their way to something more productive, rather than just making their bosses madder at them.

I see a large number of spam emails with the "fake invoice" scam, and I've often thought an AI powered voice bot could really waste the time of the scammers call center and save some senior citizens money and grief. Has anyone developed an open source equivalent?

My father setup his own solution for robocalls. If the phone rings more than three times it kicks the call to the fax machine. Apparently fax lines are supposed to be kept clear because of legal reasons? All I know is half the time I call I get a dialup sound as my call gets routed to the fax machine.

I wish they would have a number I can cal to speak to her. That would be fun

The "Daisy" AI sounds exactly like a female version of Bill Nighy.

So, scammers start using these things too (obviously), and so AI will be trying to scam AI, then they'll be like "we need to detect when our bots are talking to a bot, so they can exit the call sooner." GANWAR

"I got all the time in the world" I never felt that so literal and scary

This may be the best use case for AI yet.

As long this doesn't put Kitboga out of business.

What I find amusing is that scammers have my phone number but my wife's name - so I get random calls where they say "Can I speak to Ms Arethuza" and I say "No" and end the call.

I love that they are weaponizing the perceived kindness of senior citizens in this way. Many of the victims of scams are some of the nicest people ever that were taken advantage of in some capacity - usually while trying to help someone. It's the digital age equivalent of staging a broken down cart and then robbing some old folks. I think most of us hate the idea of the "solution" being to not go near anyone with a broken down cart out of fear.

I'm not saying this fixes everything, but I would rather a world where scammers odds at making a living at this are so poor they won't bother versus a world where everyone has to block every number by default and live in metaphorical bunkers to never interact because you might be a scammer.

So it's just an AI re-implementation of the Telecrapper 2000 [1]? The original site is down but there are plenty of YouTube videos [2] of it still available.

[1] https://hackaday.com/2005/09/08/telecrapper-2000/

[2] https://m.youtube.com/watch?v=QlK_zHisT_A

If you use a VoIP service like Twilio and voip.ms, you can set up a very simple IVR menu that just asks unknown callers to press 1 to be connected to you. No AI involved.

For me, this has been surprisingly effective against robocalls. Obviously this isn’t going to work against scammers who call directly, but most of the spam calls I receive start with some pre-recorded message which isn’t going to pass the menu.

Edit: s/auto/pre

Great PR move, but surely better to invest in more basic security measures supported by their app (chain of trust, verified callers/messaging etc). Instead their app is primarily a react native sales tool. Part of the reason that o2 was so affected by scammers calling from south Asian call centres with “the latest offers” was because they used to do exactly this with their customers.

Seems like the logical endpoint of a lot of this is people getting paid directly for their attention. Want to call me? I've set a price of $5/call that I answer, and an additional $10/minute of listen time after the first 10 seconds. Want to send me an email? $1/email and $5/100 words. Anyone I have emailed is automatically on my allow-list, which I can also adjust manually.

Of course, the scammers have created their own AI to call people and scam them so it's just playing catchup.

It's like corporate avatars for hiring or cold call bots for sales.

I'd say stick to white listed numbers, but pro phones can't do that, and they are the most prone to spam.

Wait until the scammers become AIs too. Thinking of a world where phone conversations between humans is the exception ...

Imagine a day when a scam bot will talk to AI granny.

Actually the google phone as some spam blocking additional to the one in my carrier which is crap, I was using the samsung app and I was getting spam calls always, I set up google one , now I get none

Surprised there aren't any comments whining about this because the scammers are bigger victims. Usually there's at least one.

So we should all be relieved now because scammers can't train and deploy AI models but old people will surely adopt them.

Fight AI scams with AI assistants and just let them have the conversation themselves

Follow-up idea, probably silly:

Robogranny Olympics. Allow people to submit their own bots, Micromouse/koth.org/Battlecode-style. Bots are run on random phone numbers for several weeks, and are ranked on average call length (total time/num calls). The robot that can keep spammers on the line the longest wins the prize.

A condition of submission is that the bot code and weights are published (not open sourced, just published). Competition is repeated yearly.

Sadly, many scam callers are actually slaves in Myanmar. Those unable to detect this may face physical punishments, or worse. Not that I'm supporting the practice of scamming, but this kind of scam prevention is also morally dubious.

cant even stand to watch (or listen to) more than 5 seconds of the obviously generated granny. i doubt even the audio will be convincing enough.

Other controversies aside, this could end up being a bit of a organic Turing test, if the AI ever becomes convincing to enough scammers.

Or Virgin/O2 could, you know, just actually block the scammers.

That's great, but I can't tell who are fraudsters until after hearing what they say.

This goes to show how tech can be a double edged sword. Sure, AI is going to be abused by scammers, but in the never ending arms race its also going to be used against them.

This feels more like misery porn disguised as a crusade to 'protect' the vulnerable. I’m not suggesting we make it easier for people to get scammed, but intentionally engaging seems like a recipe for a harsher clap back. Education > Escalation.

Who Timewastes the Timewaster?

If Jim Browning is involved, it might not be terrible. Let’s see.

Weird how she doesn’t sound like an old lady. Just hearing the voice, she sounds about 45.

honeAI pot

Or make every phone call cost a cent. .But then again therecis large growth potential in human hostile software abd services and defenses against that

Wait till the AI scammers start calling

They would be better sorting themselves out first rather than working on this nonsense

I wonder when will the scammer also create an AI scammer.

Joke aside, is this going to be cost effective? What would it cost to keep a scammer on the phone for an hour? Who will pay this bill?