š§µ View Thread
š§µ Thread (177 tweets)
Apple isn't killing advertising, Apple is positioning themselves as a middleman between you and the advertisers by using the current zeitgeist as an excuse to lock down their customers data so that they can be the ones making money off of ads being shown to you.
Apple will be selling access to you just as google and facebook are currently doing. The only difference is that your phone will be running the algorithms against your data, compute relevant ad targeting data locally, an then make requests using that.
Edge computing: now all your devices can snoop on you to target you with ads. But because of the magic of machine learning, the people who run these computations have plausible deniability when you suddenly start getting ads from Russians.
Apple makes money off of hardware sales, so you don't have to worry about declinign revenue causing them to shift into profiting off of their captive market?https://t.co/ocwUysOV1t
Apple makes money off of hardware sales, so you don't have to worry about declining revenue causing them to shift into profiting off of their captive market?https://t.co/sNUNzMtcut
Apple makes money off of hardware sales, so you don't have to worry about declining revenue causing them to shift into profiting off of their captive market?https://t.co/WQ5oHg319m
Don't worry, apple isn't in the business of collecting larges swaths of data and acting as a gatekeeper to that data. Oh wait.https://t.co/EWFdTCNgJC
Apple is going to farm you and manipulate you with ads, and you'll like it because their machine learning models are privateā¢.Imagine a company that doesn't follow the herd & thought differently.https://t.co/a89P1QCWKN
Google has been using differential privacy for at least 2 years before apple even thought about using it for their ad targeting business.How might a brand maintain long held sentiment of consumer trust when trying to increase the scale of it's propaganda dissemination platform?
So a company create a digital doppelganger of you by running machine learning algorithms on your phone, and then uploads a model of your habits to their servers and use that to generate predictions about you.Have they stolen your soul?Have they violated your privacy?
Privacy is a red herring. Who cares if your data is public or private?What you ought be worried about is how your data is being used to manipulate you. That's the subtext that all of their brand messaging is dancing around.They want to steal your soul.https://t.co/lJKfa2TDdP
Do you give a shit if they run that machine learning model generation on their server or on your phone? Does that really matter if 3rd party advertisers pay google, facebook, or apple to pollute a target demographic with political ads anyway?
Phones comes preinstalled with a machine learning algorithm that searches your content for incriminating evidence, modeling your behavior by tracking what you do, and then uploads aspects of that model to the cloud?But its OK because your data never leaves your device... right?
An psychological profile goes into your home, makes a bunch of notes about all your belongings, watches you for a few days, then reports back on what they learned. No privacy violation at all. Their notes are anonymized and aggregated.
The firm who hired this profiler then uses this data to target you and 500 others just like you so that you all are given the same message. A message that is inherently designed to manipulate people who share your psychological profile.This is fine because #privacymatters
Here at Apple we care about you, so we're not profiting off of your dataā¢. We're building machine learning models generating privateā¢ metadata. We generate models averaged with 500 similar people whom brands are willing to pay us to access. All the better to target you with. /s
"In fact, selling peopleās information to advertisers would be counter to our business interests, because it would reduce the unique value of our service to advertisers. We have a strong incentive to protect peopleās information from being accessed by anyone else." - Zuckerberg
"Weāre not forming the detailed profile, and then allowing other companies to buy the opportunity to target you. Itās not the business that weāre in." - CookWhat are you doing with all those machine learning systems you're building to better target ads on your ad platform then?
So you're saying you're not creating a detailed profile of people? I wonder how long until pervasive click-fraud swamps advertisers on apple's platform and undermines this privacy preserving feature. https://t.co/MCtH0SK99B
Gatekeepers leveraging brand clout built on multi-billion dollar ad-spend. Trying to convince you they're your friend. Don't mind us as we harvest your data for our profit like everybody else. Your phone does it for us, and you'll pay for the privilege!https://t.co/yzXgMsvcds
Sure they support rĢ¶eĢ¶gĢ¶uĢ¶lĢ¶aĢ¶tĢ¶oĢ¶rĢ¶yĢ¶ Ģ¶cĢ¶aĢ¶pĢ¶tĢ¶uĢ¶rĢ¶eĢ¶ Ģ¶&Ģ¶ Ģ¶bĢ¶uĢ¶iĢ¶lĢ¶dĢ¶iĢ¶nĢ¶gĢ¶ Ģ¶aĢ¶ Ģ¶mĢ¶oĢ¶aĢ¶tĢ¶ Ģ¶/Ģ¶ Ģ¶eĢ¶nĢ¶tĢ¶rĢ¶eĢ¶nĢ¶cĢ¶hĢ¶iĢ¶nĢ¶gĢ¶ Ģ¶iĢ¶nĢ¶cĢ¶uĢ¶mĢ¶bĢ¶eĢ¶nĢ¶tĢ¶.Ģ¶ a data broker clearing house... But this is power play against those who don't use differential privacy. It gives Apple gatekeeper status.
Google's been working on the same model for a while now.https://t.co/k3q6ybyFYS
I wonder how much owkin paid to get their product name dropped in two articles about the topic of federated learning.https://t.co/Es0VYOAqtW
The selling point of 'federated learning' paraphrased: "by offloading the work we do to generate machine learning models of your behavior onto your phone, we don't have to spend all that bandwidth uploading your personal data to us!"Automated spying on your behavior, IMPROVED!
What, you think that machine learning chip was put there to make funny looking emoji faces?https://t.co/klT4tPIVXAhttps://t.co/2nsyHBAqTh
I love how the main selling point is that it helps prevent our own governments from snooping on our personal data. With Apple making it out like Google and Facebook are the ones doing it to you.Meanwhile writing tools to automate harvesting machine models, because ads!
Silver-lining? These new phones are so expensive that not many people are buying them anyway, and it's likely going to get worse as geopolitics drive market collapse.https://t.co/u7P4kKQHRy
Anybody wanna estimate how long it will take until we see widespread phone rootkits maliciously lift machine learning models out of memory and upload 'em to three letter agencies?Reverse engineering machine learning models? There's an app for that.https://t.co/urVM85jnlU
We're going to move the attack surface down to the user's phone instead of in our cloud because we believe in privacy.jpgSeverešandšPersistentšThreathttps://t.co/Kr5I2sLzL1
We're going to hire the "Googleās chief of search and artificial intelligence" to ensure that our machine learning models are really really good at modeling your behavior. It's ok tho because your data stays on your phone.https://t.co/6k5Hl5luaa
We absolutely didn't fight tooth and nail to prevent you from jailbreaking your phone and changing service providers. What do you mean our only service provider was facilitating widespread wire tapping behind the scenes? Oh.https://t.co/bLIK6MXJfE
"I think we're going to very soon start to build models, predictive models of people, that are so accurate. And then nobody will accept the existence of free will anymore. I'm sorry people who thought... this really upsets some people..." https://t.co/GxkkakpWcq
"by selling the tool as a privacy-focused feature, the company is building a new identity system that it owns entirely. [...] personally, I think we need more tools like this, just not from the very platforms further entrenching their own kingdoms."https://t.co/6VTEZ0Sm9Z
"The ad industry needs fixing, yes, and these types of tools are powerful ideasāin the right hands. Iām terrified of running into the open arms of the worldās richest company, leveraging its platform to lock you into its grasp even further."
We must ensure that voters concerns about their personal data staying private is placated. But at the same time still be able to use their personal data to generate psychoanalytic profiles so that we can better target them from afar. trollface.jpghttps://t.co/qG4s8VTYrm
"One of the most interesting things to me here is simply to know that Google are working on this, have a first version in production working with tens of millions of devices, and see significant future expansion for the technology "https://t.co/OMw99ex7B5 https://t.co/Ve0U8gqEiP
"data needed to derive such insights is personal and sensitive, and must be kept private. In addition to privacy [...] deployments of learning systems using this data must also consider resource overhead, computation costs, and communication costs."https://t.co/GI5mJvtCCg
Nothing to see here.https://t.co/nYAklz0iEehttps://t.co/jH2ekCvwxq
How much does apple spend to shove their product down your throat and sway public opinion?https://t.co/oYAhL3eQwm
Phonesšwerešašmistake.https://t.co/tAI65GHw8Fhttps://t.co/AeSMj8coDa
Companies manipulating your photos, embedding tracking IDs as metadata. They're not even using steganography.https://t.co/lfziM6stkp
Worth noting this has been going on since 2012 according to https://t.co/LkSupTRbRx (by way of https://t.co/x8GbhbMIlC from 2015)Even if this is just storing meta-data about an image's permissions in itself (for a DRM like system within site), can be leveraged by 3rd parties. https://t.co/5X6ygBkph2
Did you know that the chip that is responsible for tracking most of the iPhone's Sensors (including the microphone) also runs while it is asleep...https://t.co/da2YRScoehhttps://t.co/2zv7jIMFgJ https://t.co/YxqzAavP5Y
Are we having fun yet?https://t.co/I2EXhMrYvt
"And you don't seem to understandA shame you seemed an honest manAnd all the fears you hold so dearWill turn to whisper in your ear"š¶https://t.co/PqihShfNz7
I was working on using barometric pressure to track occupancy using an old central vacuum system by mirroring some research in HVAC monitoring from 2010.In 2014 I saw apple had barometric pressure sensing Someone made a proof of concept of an idea:https://t.co/cbXN6nUK77
Don't get me started on crowdsourced echolocation."There are already many applications, and we foresee many more. This is only the beginning!"https://t.co/0AbZM3h1j9 https://t.co/rWyGejuicG
What frequencies can you pick up with a google home, siri, or an Alexa?Asking for a friend.https://t.co/lWJ1zKXdyX
I play a game w/ myself:I ask "what is possible",then "who would be doing it",filter it thru my "what incentives would they be under to make that happen",and finally "do they have resources/skill to do it?"Predicting the adjacent possible.https://t.co/ghSPTZFASu
āConstant audio surveillance would produce about 33 times more data daily than Facebook currently consumes, [...] such snooping would be eminently detectable, ringing up noticeable amounts of data on your smartphoneāhttps://t.co/8LC1CCqDmfIs that right?
Good thing these companies aren't exploring "realtime AI" on your phone https://t.co/BoFLabbA9r and embedding all sorts of interesting deep learning chips in there.
It's not like all these tools we've been building the last few years involve voice-to-speech or anything like that.I wonder how far you can compress an audio streamhttps://t.co/8Gw5FojBUoif you focus on words only.https://t.co/cwA7SZoPJRhttps://t.co/5cClfwqEq3
I think it's fun how that Vox article about facebook denying listening in is from December 2018, and this 3 part paper on emulating human hearing and focusing on compression based autoencoders is also from June 2018.https://t.co/8k5jxI43ov
I learn about AI to understand the neuroscience of autism's languge problem."an autoencoder could compress a frequency spectrum with a total of 3500 values into a vector with a length of 500 values. [...] These are only examples"https://t.co/j2iEuqEVN5https://t.co/NEba1KrRZ2 https://t.co/VZ2VBqfzfk
another one from 2018 is this paper on a fast simple solution to the cocktail party problem.https://t.co/frvl7860qYTho they didn't handle lost of noise.https://t.co/kWFbZjvGSV
Theoretically you can minimize error rate of your voice transcription by ensuring your {amazon alexa, google home, apple siri, microsoft cortana, samsung s voice, comcast X1 remote} is close to you, & away from sources of noise like fans or television. Welcome to the cloud baby!
What's he building in there?https://t.co/xYe6KsXsyZhttps://t.co/zN4hNZ2jJj
"But I'm on the outside, I'm looking inI can see through you, see your true colors'Cause inside you're ugly, you're ugly like meI can see through you, see to the real you"š¶https://t.co/SmjglJKZDjhttps://t.co/MDE6XmW4SD
"Brave uses anonymous-but-accountable ad confirmation events to allow us to report campaign performance and delivery for advertisers without tracking users and without personal data leaving the user's device."https://t.co/lJAmTzDtCO
"Brave innovates by being first to market with a privacy-preserving purchase intent mechanism that leverages locally available data that never leaves the device."https://t.co/lmlFcWSIQvSo they're doing the same thing google and apple are intending to do. https://t.co/XcDbJhyKHa
The problem, it seems to me, is you're trying to target propaganda to influence behavior by creating predictive profiles of people, and then using it to sell this as a service, calling it advertising.https://t.co/zde4ChhvOp
Gatekeepers leveraging brand clout built on multi-billion dollar ad-spend. Trying to convince you they're your friend. Don't mind us as we harvest your data for our profit like everybody else. Your phone does it for us, and you'll pay for the privilege!https://t.co/yzXgMsvcds
Who influences the influencers?https://t.co/B8E54IMFGy
Just reading the google AI blog for no reason.https://t.co/XThvMWNMgp
Nothing to see here, move along folks.This isn't intended for surveillance at all... we need to compress lots of audio information so our survielie... ah hmm... I mean voice assistants can better survuel... err... serve you.https://t.co/J6m8OqtZOK
I hate being right.https://t.co/UILNfKiXki
All the better to speak to you with, my dear.https://t.co/aCUbet0Jwj
"When you're rife with devastationThere's a simple explanation:You're a toymaker's creationTrapped inside a crystal ballAnd whichever way he tilts itKnow that we must be resilientWe won't let them break our spirits"š¶https://t.co/ETaANnp2OLhttps://t.co/2BQxnLamMf
Did you know that anxiety induced adrenaline rushes can stop you from stuttering?https://t.co/MvDLXZBTpD
Feeling left out? Don't worry, it's absolutely ethical that we include everybody's voice profile in our tooling. We want to be able to hear everyone!https://t.co/EXSxHY6Hljhttps://t.co/fbYlXm9yMd
Thinking a lot about ethics of recording passively with a real time sensory arrays.https://t.co/u9226I6TKZ"Across N.Y.U.ās microphones, the sound of human voices ā recorded in snippets too brief to follow conversation ā has become less common."https://t.co/Lx7NWEOCrq
We use math to cast a wide net while harvesting data about populations of people. Tricking^H^H^H^H^H^H^H^H Engineers can feeling good about not violating your personal privacy & still show marks^H^H^H^H^H audiences paid propaganda^H^H^H^H^H^H^H^H^H^H ads.https://t.co/HGdbxM43KU
"These changes to how credit scores are calculated could be very helpful if you have had trouble getting approved for credit in the past."https://t.co/WubrnqDFKPhttps://t.co/Fh4WqX85oN https://t.co/VY40lwL4gv
You see.. we gotta do mass collection of people's web browsing history so that disenfranchised people will be able to access loans. Please ignore how this data can and will be abused once collected. https://t.co/Fu2YOkrMxBhttps://t.co/48dnJ9GBuH https://t.co/0CK9AY51Wr
In which @ukzak argues that you should prefer Apple's pervasive tracking system over Google's pervasive tracking system.https://t.co/mkTizS0QYcI'm sure Apple's U1 positional tracking chips are going to be just fine. It's got randomization!https://t.co/niNVrN27Hf
"Regular people, not just impersonal, commercially motivated merchants or advertising networks, can exploit the online advertising ecosystem to extract private information about other people, such as people that they know or that live nearby,"https://t.co/yAVWIcNQ38
Am I a regular person?https://t.co/bqU2GzTMTs
Are you a regular person?https://t.co/yIARxeiT9S
Is this a regular person?https://t.co/x5ZkfMQ78q
The machine is real.https://t.co/FbHD1pLIge
"AT&T has a unique power to extract information from its metadata because it retains so much [ā¦] more than three-quarters of U.S. landline switches, and the second largest share of the nationās wireless infrastructure and cellphone towers, behind Verizon"https://t.co/ujUO9dPEBo
"Introducing this powerful scanning technology on all user devices without fully understanding its vulnerabilities and thinking through the technical and policy consequences would be an extremely dangerous societal experiment."https://t.co/7hBZtO5VLx
"Looking forward to the day that people publicly argue for client-side scanning on NeuralinkOS for thought-crime." šhttps://t.co/KnFN92ZwOr
"An AI cop running in every userās phone and monitoring their private words and actions is the most dangerous thing, both to us as individuals and to our civilization."https://t.co/8Nurup3BR0
"Our results suggest that even heavily sampled anonymized datasets are unlikely to satisfy the modern standards for anonymization set forth by GDPR and seriously challenge the technical and legal adequacy of the de-identification release-and-forget model." https://t.co/VrUDfXokP6
"Privacy advocates and the press noted how this was all giving Apple users a false sense of security without really fixing much. Privacy experts and press outlets also repeatedly informed Apple this was happening, but nothing changed."https://t.co/RL9riRbt4k
"The differential privacy technology used by Apple is rooted in the idea that statistical noise that is slightly biased can mask a userās individual data before it is shared with Apple." https://t.co/JyfstaMXAy
Whoops "On the difficulty of achieving Differential Privacy in practice" https://t.co/2Ww0q7E1Urhttps://t.co/WUjBQrz7TJ
Earlier in thread: https://t.co/JuP6xYB7tL
Google has been using differential privacy for at least 2 years before apple even thought about using it for their ad targeting business.How might a brand maintain long held sentiment of consumer trust when trying to increase the scale of it's propaganda dissemination platform?
Differential privacy only helps if you take into account entropy and information theory concerns about uncertainty. It's not a panacea for privacy.https://t.co/AkYcaS5UNF
@ryanstellar @balajis @AndrewHires @maxrogo There's also "differential privacy" that could tackle aggregation without leaking personal details. Apple uses it to get statistics information https://t.co/0MrpYdS0af and its fairly effective.Hard part w/ genome sequencing is at time of capture & usability of various schemes.
It's not magic."Privacy bugs happen when too little noise is added, resulting in too little achieved privacy, or when too much noise is added, resulting in too much degradation of the result's accuracy." https://t.co/e3ZpSX4ClM
"keep in mind that differential privacy is as much focused on establishing a framework for measuring information leakage"https://t.co/k4dJ3UbzDz
Don't worry, they won't know your personal data. They'll just know aggregate data about the entire bucket and be able to make an educated guess. Suck if they label the entire bucket as bad. You're fault for being too similar to dead beats. https://t.co/sCUkwth6k7
Knock, Knock,Who's there?rot13(Differential Privacy).Qvssreragvny Cevinpl who?https://t.co/dqd00IY6Ik
"In fact, selling peopleās information to advertisers would be counter to our business interests, because it would reduce the unique value of our service to advertisers. We have a strong incentive to protect peopleās information from being accessed by anyone else." - Zuckerberg
"Our goal with Opacus is to preserve the privacy of each training sample while limiting the impact on the accuracy of the final model. Opacus does this by modifying a standard PyTorch optimizer in order to enforce (and measure) DP during training."https://t.co/7fFxLorlnB
"The companies revealed upgrades for their phones that protect data and reduce reliance on the cloud. It also binds users more tightly to their ecosystems."https://t.co/TliPOgeET4https://t.co/S2arPpZmqi
"So when Facebook points out that Apple is using switching costs to take its users hostage, they know what theyāre talking about." š¤£https://t.co/Hr2VL5WQfPhttps://t.co/YwMEVamxrH
Did @snowden's theoretical cellphone where you can physically disconnect the GPS receiver and Cell Connectivity ever get to production?https://t.co/t6tfav4vjz
Why has it taken so long to get analysis of this data visible to the public?All we get is a dashboard with "low, med, high" warnings that have no meaningful breakdown in how they are calculated. And yet the CDC has bought data from multiple brokers?https://t.co/UjIT7mmwZa
This makes me laugh. very fumy. haha.> "We recognize that once models are released, it is impossible to restrict access to only āintendedā uses or to draw reasonable guidelines around what is or is not research."https://t.co/xISztX7MNlhttps://t.co/yqoucbeWJ7
Did you know that you can search Google's YouTube transcripts for when people mention things in the wiretap watch lists?https://t.co/ScHOzj3Wlw
Ai Foom + Funny = fumy.Cuz I like playing with words too.Is that the sound of bootstrapping a pumy engine? yes. Yes it is. https://t.co/1eT2mWjCHS
Louis Rossmann shows off GrapheneOS, which lets you remove Google Play Services, or run it in a sandbox. Looks pretty good. https://t.co/vGRbtLeXAT
"the telecom companies now have the ability to give these networks the exact same set of data they were gathering before. In fact, they could potentially pass improved data due to their ability to track cross-app."https://t.co/2nbUUYTrUjhttps://t.co/FbHD1pLIge
"AT&T has a unique power to extract information from its metadata because it retains so much [ā¦] more than three-quarters of U.S. landline switches, and the second largest share of the nationās wireless infrastructure and cellphone towers, behind Verizon"https://t.co/ujUO9dPEBo
Shothttps://t.co/7WnXjk7P0F https://t.co/9Sj1Beq8uj
Chaserhttps://t.co/eKvhjccBIo https://t.co/flypW02HDN
3 kBps (bytes) = 24 kbps (bits)So we can compress audio waveforms from 24kbps to 3 kbps using your phones' local AI compute chips.This assumes the data isn't compressed further before sending surreptitiously after local threat analysis.
Can we get smaller than 3 kbps if we convert data to text in real time?https://t.co/CvFPZhFoPE
Facebook can do this too to now. It's being done for their VR headset most likely."Today, we are detailing progress that our Fundamental AI Research (FAIR) team has made in the area of AI-powered hypercompression of audio." [OCT 2022] https://t.co/SsAZ1DvPum https://t.co/kNSswrEMan
That 64kb/s to 6kb/s is for a complex waveform such as an MP3.How much further can audio be compressed if you pass it thru a voice isolation pass first and clamp it to a narrower range? https://t.co/zxhQnZbU6V
How many other companies are building compression tools that would allow for trivially recording conversations 24/7?https://t.co/07cSvmRzKX
Sure, Meta wouldn't do that! But its not unheard of to have phones be hacked. https://t.co/xG31XgyhZr
It knows your voice profile. "Apple will let you teach your iPhone to speak in your voice [...] The feature is one of many accessibility tools expected to debut this fall" https://t.co/hcPzgty04HI can do a lot with a voice.https://t.co/hVe2FNcrHs
What's really fun is thinking about what happens when 5g networks allow us to monitor us thru walls. https://t.co/ZaSxaotw9b
We'll be able to take an audio clip generated from a laser bounced off of a house, and recreate the facial profile of the occupants enough that we can make a reasonable guess of who's occupying it. Or maybe from a cellphone / alexa / google voice / siri.https://t.co/5vXT3ZN8Bf
thru walls. For your safety.https://t.co/60e3Dho168
Did you know it's possible to hide signals within another audio signal and use psychoacoustics to make it so you can't hear it? Useful for imperceptible audio watermarking. https://t.co/ZaujT1ETcV
Doing that in Realtime using AI chips would be useful for transmitting audio beacons.https://t.co/xZjSfHvR3o
"nation-state actors located around the world used LLMs to research specific technologies and vulnerabilities, as well as to gain information on regional geopolitics and high-profile individuals" [FEB 2024] https://t.co/HxjPO98S7ahttps://t.co/AnTDOg4U8S
Trillions of dollars."It is possible that the very nature of deep learning models makes such threats inevitable. Thus, we believe that these considerations should be taken into account as we increase usage and reliance on such AI models."https://t.co/mJMITQHg6w
A Daemonsfor you! And a Daemonsfor you! And A Daemonsfor you!Daemons for everybody!https://t.co/uXn7kueYFs
I see nothing about openAI here. I see a front org for funneling AI based military technology with plausible deniability.https://t.co/FyDn7rF48ZSummoning the Demon.