Liver Detoxes !

[Hoover Attack]

Hoover - if you had to choose, would you prefer to hand over to me a log of every conversation you've had over the last year or details of every single thing you've viewed on the internet over the last year?

Which one do you think would be easier for me to get in depth personal information from?

Happy with either.

It's weird that you don't think there is any value in what people say or hear. That data is just as relevant as anything else.

[Mick McQuaid]

Most people's chat is inane nonsense with no value in terms of increasing the click through rate of ad. Sure there would be a few nuggets buried in there, but why sift through sh*t if you have it all nicely wrapped and ready for you?

A computer would have no way of distinguishing whether a conversation is about whose turn it was to put the dishwasher on over something useful without processing it (with voice recognition to decide whose voice to prioritise?), which is billions of times more computationally expensive than processing the text of your searches or other data points and then trying to pull the useful information from it.

My point is, the way our data is collected and used is known about, and the scandal is that we have all just accepted it. How this is going to be used with advances in ai to make automated decisions about our lives should scare everyone. Thinking your phone is listening to you in an incredibly sophisticated and highly secret way (although the results are actually a bit crap) is just silly.

[Max B Gold]

Would wrapping your phone in Bacofoil while talking in Esperanto help in this situation?

[Mick McQuaid]

I've just requested and got a download link to all the data Google has on me. Just under 64GB, the sneaky so and so's have not included the voice transcripts though.

I'm going to try later and stick it through chat gpt with a bit of pre processing and see what information it knows about me. I'm guessing even with my limited coding skills it's going to be a f*** of a lot.

[Mick McQuaid]

From one 16mb file of youtube history with minimal cleaning, chatgpt has deduced that I live in London, support leyton Orient, like a variety of music including punk and classical. I use powerapps and python regularly in my job which is probably related to social care. I am likely to be single and not actively seeking a relationship but have a child or am closely involved with one in a tutoring or mentoring role, the child is likely to be in early years of secondary school. I am not aligned with any political party but am interested in politics with independent and left of centre views. My hobbies include chess, astronomy, maths and problem solving.

Suggested ads I would be interested in are Khan academy (which i get loads), Hotels.com (bit basic) and telescope equipment (Rother valley optics is a regular on my ads).

[Mick McQuaid]

Not only quite accurate but a reminder of what a boring person I am.

[Mick McQuaid]

Added in some data from Google maps and it knows I have a pet, probably a cat. I most likely live in a flat in North West London (obviously Google knows exactly where I live) I am in my 40's and have an income of £40-55k.

My new ad suggestions are Udemy (all the bloody time). Trading212 (doesn't everyone?) Bose headphones and speciality coffee subscriptions.

[Mick McQuaid]

Last update, as while I'm finding it interesting I'm sure no one else is.

I was talking to someone in the office yesterday and they mentioned that they had bought an automatic water fountain for their cat, I said to my kid this morning as the cat was hogging the kitchen tap that maybe we should get one. I have not searched for one ever.

I asked chatgpt to guess some of my next likely purchases from my profile and out of the 20 suggestions that is one of them. Clearly my phone must have recorded me and passed it on to chatgpt rather than it being an example of how easy it is to build a picture of someone from their data and occasionally getting very lucky.

It also suggested Phillips Hue lights which I have neither spoken about or searched but was thinking of getting for my kids room. Obviously it's already advanced to mind reading.

[Hoover Attack]

Your cat uses your kitchen tap? What the f**k?

[Hoover Attack]

That's a bizarre take on events though.

[Mick McQuaid]

I know, which is why I think I should buy an automatic water fountain.

It doesn't actually turn it on, it just sits and stares at me until I do. It is polite enough to drink from mid stream, but still.

[Adz]

Yes, all of that is true.

BUT these tech companies want our 'data' to pass on to third parties to sell us things. Why would they not listen, when they're digesting every other bit of info?

The other one I get is Friend Suggestions for people I have no connection with but happen to have been in the same location as. That's wild.

This shows your complete lack of understanding of technology. That is far easier than listening to everything you say and working things out from that.

[Adz]

Last update, as while I'm finding it interesting I'm sure no one else is.

I was talking to someone in the office yesterday and they mentioned that they had bought an automatic water fountain for their cat, I said to my kid this morning as the cat was hogging the kitchen tap that maybe we should get one. I have not searched for one ever.

I asked chatgpt to guess some of my next likely purchases from my profile and out of the 20 suggestions that is one of them. Clearly my phone must have recorded me and passed it on to chatgpt rather than it being an example of how easy it is to build a picture of someone from their data and occasionally getting very lucky.

It also suggested Phillips Hue lights which I have neither spoken about or searched but was thinking of getting for my kids room. Obviously it's already advanced to mind reading.

I think this is why hoover and ceb are getting all het up. They think they're edgy and independent, and don't want to hear that they're actually easily predicted by an algorithm.

[Adz]

Can you show your workings for your 'not cost effective' claims please?

OK, at a basic 128 bit rate 24 hours of audio would be 1.5gb. Lets conservatively say there are 1billion devices available to google around the world.
Storage cost of s3 is 2 cents per gb per month. So storing the data alone would cost 360 million USD per year for one days worth of data!
You would then need to do speech to text translation, amazon transcribe is 0.0004 cents per second, so roughly 35 bucks a day. So that would by 35 billion for 1 day of translation, you would have to run this everyday on the new data.
You would then have to add the cost of processing the data, which I can't really estimate without knowing the algorithm and compute power required, but probably would run in the billions.

Obviously Google would own use their own infra rather than amazon so it would be cheaper for them, but I know the amazon costs, so easier for me to use that as an example. That aside this should give you an incite into the cost of doing this, not even taking into account the legalities and complexities of doing that.

[Adz]

You could obviously optimize the above by reducing the hours you listen to, so maybe reduce by a 3rd, You're still well into the billions. Whether it's cost effective, you'd have to drill into Googles ad revenue and somehow work out how doing this affects that.

[Adz]

I know, which is why I think I should buy an automatic water fountain.

It doesn't actually turn it on, it just sits and stares at me until I do. It is polite enough to drink from mid stream, but still.

We got one a few months ago, and the cat's are drinking much more than they did from a bowl. I think it's well worth it.

[Long slender neck]

Can you show your workings for your 'not cost effective' claims please?

OK, at a basic 128 bit rate 24 hours of audio would be 1.5gb. Lets conservatively say there are 1billion devices available to google around the world.
Storage cost of s3 is 2 cents per gb per month. So storing the data alone would cost 360 million USD per year for one days worth of data!
You would then need to do speech to text translation, amazon transcribe is 0.0004 cents per second, so roughly 35 bucks a day. So that would by 35 billion for 1 day of translation, you would have to run this everyday on the new data.
You would then have to add the cost of processing the data, which I can't really estimate without knowing the algorithm and compute power required, but probably would run in the billions.

Obviously Google would own use their own infra rather than amazon so it would be cheaper for them, but I know the amazon costs, so easier for me to use that as an example. That aside this should give you an incite into the cost of doing this, not even taking into account the legalities and complexities of doing that.

This could all take place on the device couldn't it? 24/7 recording and processing it elsewhere is unlikely and unnecessary

[Mick McQuaid]

Great idea, but with just a few flaws. It means the battery manufacturers are in on it too, as well as every scientist working on extending battery life. It means the methods for hiding the large language model stored on the phone are known about and considered at every level of the design and that the supposedly open source Android is actually a completely different version to that actually shared with every ios phone manufacturer, and with this massive conspiracy not one person has spilled the beans. Definitely plausible even though I have literally demonstrated that the phenomena of a having a computer give you information you have only spoken about is easily replicated.

[Adz]

Can you show your workings for your 'not cost effective' claims please?

OK, at a basic 128 bit rate 24 hours of audio would be 1.5gb. Lets conservatively say there are 1billion devices available to google around the world.
Storage cost of s3 is 2 cents per gb per month. So storing the data alone would cost 360 million USD per year for one days worth of data!
You would then need to do speech to text translation, amazon transcribe is 0.0004 cents per second, so roughly 35 bucks a day. So that would by 35 billion for 1 day of translation, you would have to run this everyday on the new data.
You would then have to add the cost of processing the data, which I can't really estimate without knowing the algorithm and compute power required, but probably would run in the billions.

Obviously Google would own use their own infra rather than amazon so it would be cheaper for them, but I know the amazon costs, so easier for me to use that as an example. That aside this should give you an incite into the cost of doing this, not even taking into account the legalities and complexities of doing that.

This could all take place on the device couldn't it? 24/7 recording and processing it elsewhere is unlikely and unnecessary

You could do the recording and probably some dodgy transcribing, but it would be easy to find that out. You couldn't do the processing though, so you'd still have to send the transcribed data over the wire and store it. Also on that it would be pretty easy to see your data being sent over the wire to Google be it audio or text. I've never heard this reported, which is another reason it's unlikely.

[Long slender neck]

Offline voice input possible on my phone

[Hoover Attack]

Last update, as while I'm finding it interesting I'm sure no one else is.

I was talking to someone in the office yesterday and they mentioned that they had bought an automatic water fountain for their cat, I said to my kid this morning as the cat was hogging the kitchen tap that maybe we should get one. I have not searched for one ever.

I asked chatgpt to guess some of my next likely purchases from my profile and out of the 20 suggestions that is one of them. Clearly my phone must have recorded me and passed it on to chatgpt rather than it being an example of how easy it is to build a picture of someone from their data and occasionally getting very lucky.

It also suggested Phillips Hue lights which I have neither spoken about or searched but was thinking of getting for my kids room. Obviously it's already advanced to mind reading.

I think this is why hoover and ceb are getting all het up. They think they're edgy and independent, and don't want to hear that they're actually easily predicted by an algorithm.

Not het up mate.

I'm fully aware how predictable everyone is, myself included. I just think it's bizarre that all that data is there and people think it's not being utilised (although I accept the costing reason is a more valid one than the legality of it as initially raised).

[Hoover Attack]

Can you show your workings for your 'not cost effective' claims please?

OK, at a basic 128 bit rate 24 hours of audio would be 1.5gb. Lets conservatively say there are 1billion devices available to google around the world.
Storage cost of s3 is 2 cents per gb per month. So storing the data alone would cost 360 million USD per year for one days worth of data!
You would then need to do speech to text translation, amazon transcribe is 0.0004 cents per second, so roughly 35 bucks a day. So that would by 35 billion for 1 day of translation, you would have to run this everyday on the new data.
You would then have to add the cost of processing the data, which I can't really estimate without knowing the algorithm and compute power required, but probably would run in the billions.

Obviously Google would own use their own infra rather than amazon so it would be cheaper for them, but I know the amazon costs, so easier for me to use that as an example. That aside this should give you an incite into the cost of doing this, not even taking into account the legalities and complexities of doing that.

This could all take place on the device couldn't it? 24/7 recording and processing it elsewhere is unlikely and unnecessary

Doesn't have to be recorded.

[Mick McQuaid]

Yes it does, speech to text can be done almost in real time and the recording can then be deleted, but it's not done by magic and a sprinkling of fairy dust.

[CEB2ElectricBoogaloo]

If the recording can be deleted, then the cost of storing the data that you mention above can be revised downwards significantly, no?

[Mick McQuaid]

CEB, I made the point about it not being cost effective before the point about illegality. The point about the illegality is not that the companies wouldn't wish to do it (although I still contend that the data collected already is far more valuable than peoples conversations) but the sheer amount of people who would have to know about it, including the most junior of software engineers in a whole load of related companies, means that keeping the lid on it would be impossible. All for something that adds next to nothing and is a bit crap?

I didn't refer to the cost of data storage, I guestimated electricity costs for processing. However you do it, the costs are massive.