8 ms·

> Those services have some sort of legal bounds that they're not going to use user data for "improving services" I don't think this is true. Quite the opposite

> Those services have some sort of legal bounds that they're not going to use user data for "improving services" I don't think this is true. Quite the opposite, they can do what they want with your data including training models with data you've given them in the past.

summerlight 3y ago
Any evidence, more than "they will do because they can do"? Technically, they can do that, but only if they think it's a great idea to play with a relentless stream of lawsuits where some of them may put their business into existential threats.
- bluefishinit 3y ago
  Does Google have any evidence that OpenAI is using their data nefariously? It's only prudent to assume that data you put on the internet will at some point in the future be used to train an AI model. We already know that state level actors are hoovering up civilian data en masse, so this is probably already happening.
  - summerlight 3y ago
    > Does Google have any evidence that OpenAI is using their data nefariously? OpenAI has clearly stated that they will use user input for training so it's now Google's responsibility to keep their confidential information from ChatGPT inputs, assuming that OpenAI can make unintentional mistakes. What's wrong in here? And your claim is a completely different story, Google and other cloud services clearly state that they will handle user data separately so it's their liability if something goes wrong. And you're trying to insist that they will secretly use user data for whatever they want, breaking their public promises and putting their own business at severe legal risks. So tell me, why would they do that? > We already know that state level actors are hoovering up civilian data en masse, so this is probably already happening. Please don't put your own precious conspiracy unless you have plausible evidence. That only harms the credibility of your claim and deteriorate signal to noise ratio in this forum.
    - bluefishinit 3y ago
      > Please don't put your own precious conspiracy unless you have plausible evidence. This isn't speculative. We know that the government uses Google (and others) to access pretty much all data on the internet: https://en.wikipedia.org/wiki/PRISM https://en.wikipedia.org/wiki/PRISM From the article: > Internal NSA presentation slides included in the various media disclosures show that the NSA could unilaterally access data and perform "extensive, in-depth surveillance on live communications and stored information" with examples including email, video and voice chat, videos, photos, voice-over-IP chats (such as Skype), file transfers, and social networking details. ^ all of that is going to be going into models.
  - vineyardmike 3y ago
    > Does Google have any evidence that OpenAI is using their data nefariously? Considering their chat product is less than a year old, and has had multiple bugs that expose chats to other customers, and has been shown to not be honoring “delete” requests… I think it’s safe to say that there’s evidence that OpenAI is sufficiently negligent even if not nefarious. Oh that’s to say nothing of the actual data being used in training. Imagine if a bing engineer can just ask a GPT for info on what google is planning and it spits out data it was trained on? Big risk.
vineyardmike 3y ago
Hmm have you ever actually looked at any of these legal agreements? Because it’s certainly not happening at google. Generally companies don’t let other companies play with their data. This may be happening on personal gmail accounts but it’s not happening on corporate accounts. Source: https://workspace.google.com/learn-more/security/security-whitepaper/page-6.html https://workspace.google.com/learn-more/security/security-wh...
- throwuwu 3y ago
  Same distinction for ChatGPT. The personal account using the web interface has a disclaimer stating data may be used. The API and the private instance have policies that state data will not be used or retained.
neuronexmachina 3y ago
Were any of those models generative?

summerlight 3y ago

Any evidence, more than "they will do because they can do"? Technically, they can do that, but only if they think it's a great idea to play with a relentless stream of lawsuits where some of them may put their business into existential threats.

bluefishinit 3y ago
Does Google have any evidence that OpenAI is using their data nefariously? It's only prudent to assume that data you put on the internet will at some point in the future be used to train an AI model. We already know that state level actors are hoovering up civilian data en masse, so this is probably already happening.
- summerlight 3y ago
  > Does Google have any evidence that OpenAI is using their data nefariously? OpenAI has clearly stated that they will use user input for training so it's now Google's responsibility to keep their confidential information from ChatGPT inputs, assuming that OpenAI can make unintentional mistakes. What's wrong in here? And your claim is a completely different story, Google and other cloud services clearly state that they will handle user data separately so it's their liability if something goes wrong. And you're trying to insist that they will secretly use user data for whatever they want, breaking their public promises and putting their own business at severe legal risks. So tell me, why would they do that? > We already know that state level actors are hoovering up civilian data en masse, so this is probably already happening. Please don't put your own precious conspiracy unless you have plausible evidence. That only harms the credibility of your claim and deteriorate signal to noise ratio in this forum.
  - bluefishinit 3y ago
    > Please don't put your own precious conspiracy unless you have plausible evidence. This isn't speculative. We know that the government uses Google (and others) to access pretty much all data on the internet: https://en.wikipedia.org/wiki/PRISM https://en.wikipedia.org/wiki/PRISM From the article: > Internal NSA presentation slides included in the various media disclosures show that the NSA could unilaterally access data and perform "extensive, in-depth surveillance on live communications and stored information" with examples including email, video and voice chat, videos, photos, voice-over-IP chats (such as Skype), file transfers, and social networking details. ^ all of that is going to be going into models.
- vineyardmike 3y ago
  > Does Google have any evidence that OpenAI is using their data nefariously? Considering their chat product is less than a year old, and has had multiple bugs that expose chats to other customers, and has been shown to not be honoring “delete” requests… I think it’s safe to say that there’s evidence that OpenAI is sufficiently negligent even if not nefarious. Oh that’s to say nothing of the actual data being used in training. Imagine if a bing engineer can just ask a GPT for info on what google is planning and it spits out data it was trained on? Big risk.

vineyardmike 3y ago

Hmm have you ever actually looked at any of these legal agreements? Because it’s certainly not happening at google. Generally companies don’t let other companies play with their data. This may be happening on personal gmail accounts but it’s not happening on corporate accounts. Source: https://workspace.google.com/learn-more/security/security-whitepaper/page-6.html https://workspace.google.com/learn-more/security/security-wh...

throwuwu 3y ago
Same distinction for ChatGPT. The personal account using the web interface has a disclaimer stating data may be used. The API and the private instance have policies that state data will not be used or retained.

neuronexmachina 3y ago

Were any of those models generative?