last item here seems to indicate that the OP tweet is a (bad faith?) overly broad reading. The "in accordance with our privacy policy" link part is showing right there in that original tweet, and doesn't seem to have been checked.
I know a lot of people think Musk is simply a mustachio-twirling comic book villain so there's not a lot of critical thinking ever applied to this sort of story, but even still, he'd have to be an order of magnitude dumber than buying Twitter, to think they could just train on all browsing data and that people would be chill about it.
The UK's BT Broadband did this around 2007 via Phorm. Actually they did worse - used the data to inject custom advertising. Not only were their customers chill with it, so was the Information Commissioner's Office, the government arm that ostensibly protects our privacy.
> , to think they could just train on all browsing data and that people would be chill about it.
I mean, he does think he can do what he wants with impunity and track record is that yes, he can. He thinks that asking his bot to not produce child porn is an imposition, for example.
How is ai-produced "porn" different than hentai? We're getting very divorced from the reason CSAM is illegal in the first place.
The previous status quo was that skilled digital artists could easily produce fake porn with real people's faces pasted on using Photohop. Now, lots of people can cheaply produce the same. So? Eventually, the dumber people will come around to where most of us are, which is understanding that obviously the existence of an image of a thing is not proof of the existence of the thing itself, and we can stop pearl-clutching about what weirdos make fake AI pictures of, just like we don't have to care what some weirdo doodles in the privacy of their own home.
Because it was a massive money pit that was already failing. If he wanted to have a microblogging site at x.com he could have built one at x.com and paid everyone interesting to use it for a year, for far less than he paid for Twitter.
They sold it for a lot more than it was actually worth, got actual cash from it and got a return on their investment, why wouldn't they have sold? Twitter was barely profitable when he bought it, not selling would've been crazy.
We heard this story a million times, but who's to say what's Twitter really worth? Musk bought much more than a "barely profitable website" that day. And the liberal leaders should have seen this coming. Yet they were busy laughing out loud, and mocking, and making predictions on how he'll run the site into the ground, and humiliate himself, etc.
So, as most of the web is HTTPS now they have DNS requests (if users haven't used a third party DNS like 1.1.1.1), and IP addresses. Maybe the SNI domain name if they are doing packet inspection.
Not really sure how useful this would be on model training?
Maybe ranking which sites it should give as answers based on popularity?
This x thread may not be the best source of clarity on what is actually being default opted-into. Sorry. I looked into it and it seems that Starlink denies browsing history would be shared [0]. Seems I can't edit the title any more.
> Do you share my personal information for AI training?
We are committed to protecting your privacy. In some instances, we may share personal information with trusted third-party partners who, among other activities, help us develop AI-enabled tools that improve your customer experience, although you can always opt out. Rest assured that we take reasonable safeguards to protect and secure your information whenever it is used or shared.
> Will these AI models see my Internet history?
No, your internet history will never be shared with AI models, including individual browsing habits or geolocation tracking, and we comply with laws prohibiting unauthorized surveillance.
> What personal information does Starlink collect from me?
We only collect what’s needed to provide you great service—like your name, address, email, and payment details when you sign up or order. We also gather some technical information (like IP address or service performance data) to keep your connection fast and reliable.
I wonder how much one can analyze with timings and sizes. If a news article on a known hostname has 5 Xitter embeds and 4 Instagram ones, the Creep in the Middle can count how many bytes are transferred in the HTTPS requests to Xitter/Instagram (there won't be 9 separate TCP connects due to connection reuse), and compare that to its own scrape of news articles of that host...
I was on the fence with switching to starlink from cellular data, but I realy cant stand the idea of a wrestling match with another giant corporation, weaponising "may", might", "occasionaly", "certain terms and conditions", "reserve the right to"
better the perv I know then some new more ambitios one.
It's gross, but of course we think many ISPs are abusive in similar ways? I haven't ever used a Starlink connection, I wonder if their latency a problem for VPNs or tunneling.
I've recently spent a few hundred hours playing Arc Raiders on Starlink from a small island in the Pacific. ~80ms ping to Australian servers is pretty mindblowing. Jitter and packet loss also tend to be insignificant in the absence of obstructions.
My home server back in New Zealand is behind CGNAT and I had issues with high birate (>25Mbps) HTTPS streaming over Tailscale. I suspect MTU size causing packet fragmentation combined with DERP relay fallback under CGNAT was the culprit, but that's outside my expertise to diagnose fully.
Reverse proxying to a VPS in Sydney with Pangolin achieved much better performance, almost 100Mbps over WebDAV. Somewhere inbetween I tried SMB over Cloudflare WARP (~70Mbps), but allowing Cloudflare to terminate TLS seemed incongruent with self-hosting everything else.
Indeed that is the case - all the other major consumer ISPs have ties with media companies and have much more incentive to collect and abuse user data than Starlink does.
Consistent 25-35ms to 1.1.1.1. You won't see >60ms unless you're deprioritized, and that's about your choice of plan not about what the network can do.
i have my starlink plugged directly into a gl.inet router that runs a wireguard client. all of the clients connect to the gl.inet wifi. absolutely nothing unencrypted goes out the starlink wan.
it works fine.
i have the exact same setup at home with cox cable, because they will data mine the shit out of your traffic (and they have your phone number and home address to link it to). most terrestrial residential isps do this shit. vpn everything.
It was impossible to do this through the app (a website opened from the app XD) or the website.
When creating the ticket,a chat opens and a warning appeared: "By continuing, you agree that your data will be used to train AI models. See the Privacy Policy for how to opt out."
Does this mean it's impossible to get support if I don't want AI?
After disgustingly evading Grok's support AI, was it possible to open the ticket to disable data sharing for AI training and therefore cancel future support? XD
https://x.com/scottbuscemi/status/2013397238568099843?s=20
last item here seems to indicate that the OP tweet is a (bad faith?) overly broad reading. The "in accordance with our privacy policy" link part is showing right there in that original tweet, and doesn't seem to have been checked.
I know a lot of people think Musk is simply a mustachio-twirling comic book villain so there's not a lot of critical thinking ever applied to this sort of story, but even still, he'd have to be an order of magnitude dumber than buying Twitter, to think they could just train on all browsing data and that people would be chill about it.
The UK's BT Broadband did this around 2007 via Phorm. Actually they did worse - used the data to inject custom advertising. Not only were their customers chill with it, so was the Information Commissioner's Office, the government arm that ostensibly protects our privacy.
https://en.wikipedia.org/wiki/Phorm
Pretty hilarious that they call enabled by default "opt in".
> , to think they could just train on all browsing data and that people would be chill about it.
I mean, he does think he can do what he wants with impunity and track record is that yes, he can. He thinks that asking his bot to not produce child porn is an imposition, for example.
Right? What an insane statement. Of course Musk thinks he can do this with impunity. He thinks he has the absolute right to do anything he wants.
How is ai-produced "porn" different than hentai? We're getting very divorced from the reason CSAM is illegal in the first place.
The previous status quo was that skilled digital artists could easily produce fake porn with real people's faces pasted on using Photohop. Now, lots of people can cheaply produce the same. So? Eventually, the dumber people will come around to where most of us are, which is understanding that obviously the existence of an image of a thing is not proof of the existence of the thing itself, and we can stop pearl-clutching about what weirdos make fake AI pictures of, just like we don't have to care what some weirdo doodles in the privacy of their own home.
why do you think buying twitter was dumb?
Because it was a massive money pit that was already failing. If he wanted to have a microblogging site at x.com he could have built one at x.com and paid everyone interesting to use it for a year, for far less than he paid for Twitter.
One curious thing is they SOLD it to him.
They sold it for a lot more than it was actually worth, got actual cash from it and got a return on their investment, why wouldn't they have sold? Twitter was barely profitable when he bought it, not selling would've been crazy.
We heard this story a million times, but who's to say what's Twitter really worth? Musk bought much more than a "barely profitable website" that day. And the liberal leaders should have seen this coming. Yet they were busy laughing out loud, and mocking, and making predictions on how he'll run the site into the ground, and humiliate himself, etc.
How's that going by the way?
So, as most of the web is HTTPS now they have DNS requests (if users haven't used a third party DNS like 1.1.1.1), and IP addresses. Maybe the SNI domain name if they are doing packet inspection.
Not really sure how useful this would be on model training?
Maybe ranking which sites it should give as answers based on popularity?
This x thread may not be the best source of clarity on what is actually being default opted-into. Sorry. I looked into it and it seems that Starlink denies browsing history would be shared [0]. Seems I can't edit the title any more.
> Do you share my personal information for AI training? We are committed to protecting your privacy. In some instances, we may share personal information with trusted third-party partners who, among other activities, help us develop AI-enabled tools that improve your customer experience, although you can always opt out. Rest assured that we take reasonable safeguards to protect and secure your information whenever it is used or shared.
> Will these AI models see my Internet history? No, your internet history will never be shared with AI models, including individual browsing habits or geolocation tracking, and we comply with laws prohibiting unauthorized surveillance.
> What personal information does Starlink collect from me? We only collect what’s needed to provide you great service—like your name, address, email, and payment details when you sign up or order. We also gather some technical information (like IP address or service performance data) to keep your connection fast and reliable.
[0] https://starlink.com/support/article/b82cf54a-8e57-917a-bd06...
I was thinking the same. It's not like there is vast amounts of unencrypted HTTP just running around everything uses TLS nowadays.
They could find URLs to scrape maybe, but whats the point of that when certificate transparency lists exist?
I wonder how much one can analyze with timings and sizes. If a news article on a known hostname has 5 Xitter embeds and 4 Instagram ones, the Creep in the Middle can count how many bytes are transferred in the HTTPS requests to Xitter/Instagram (there won't be 9 separate TCP connects due to connection reuse), and compare that to its own scrape of news articles of that host...
> cryps1s
Post by CISO @OpenAI. Must be a slow day.
glass houses
I was on the fence with switching to starlink from cellular data, but I realy cant stand the idea of a wrestling match with another giant corporation, weaponising "may", might", "occasionaly", "certain terms and conditions", "reserve the right to" better the perv I know then some new more ambitios one.
It's gross, but of course we think many ISPs are abusive in similar ways? I haven't ever used a Starlink connection, I wonder if their latency a problem for VPNs or tunneling.
I've recently spent a few hundred hours playing Arc Raiders on Starlink from a small island in the Pacific. ~80ms ping to Australian servers is pretty mindblowing. Jitter and packet loss also tend to be insignificant in the absence of obstructions.
My home server back in New Zealand is behind CGNAT and I had issues with high birate (>25Mbps) HTTPS streaming over Tailscale. I suspect MTU size causing packet fragmentation combined with DERP relay fallback under CGNAT was the culprit, but that's outside my expertise to diagnose fully.
Reverse proxying to a VPS in Sydney with Pangolin achieved much better performance, almost 100Mbps over WebDAV. Somewhere inbetween I tried SMB over Cloudflare WARP (~70Mbps), but allowing Cloudflare to terminate TLS seemed incongruent with self-hosting everything else.
> many ISPs are abusive in similar ways?
Indeed that is the case - all the other major consumer ISPs have ties with media companies and have much more incentive to collect and abuse user data than Starlink does.
They say 20-60ms, users seem to report 30-200ms. Seems very dependent on where you're at.
Consistent 25-35ms to 1.1.1.1. You won't see >60ms unless you're deprioritized, and that's about your choice of plan not about what the network can do.
I would be surprised there are not some amount of oversubscribed geo areas, or remote spots with longer paths, etc.
It does seem good for many, just perhaps not all.
I average 20-25ms at my lake home in Central MN on Starlink Residential lite; very similar to my Comcast Business connection at my house.
It’s crazy to me, frankly.
i have my starlink plugged directly into a gl.inet router that runs a wireguard client. all of the clients connect to the gl.inet wifi. absolutely nothing unencrypted goes out the starlink wan.
it works fine.
i have the exact same setup at home with cox cable, because they will data mine the shit out of your traffic (and they have your phone number and home address to link it to). most terrestrial residential isps do this shit. vpn everything.
It was impossible to do this through the app (a website opened from the app XD) or the website.
When creating the ticket,a chat opens and a warning appeared: "By continuing, you agree that your data will be used to train AI models. See the Privacy Policy for how to opt out."
Does this mean it's impossible to get support if I don't want AI?
After disgustingly evading Grok's support AI, was it possible to open the ticket to disable data sharing for AI training and therefore cancel future support? XD
Like what data, my DNS queries and TCP connects? Ok I guess