FaceDeer

The "environmental destruction" angle is likely to cause trouble because it's objectively debatable, and often presented in overblown or deceptive ways.

Except that is also a subjective and emotionally-charged argument.

and also alarming enough for them to take action.

Is this really an intent to explain in good faith? Sounds like you're trying to manipulate their opinion and actions rather than simply explaining yourself.

If someone was to tell me that they simply don't want to use generative AI, that they prefer to do writing or drawing by hand and don't want suggestions about how to use various AI tools for it, then I just shrug and say "okay, suit yourself."

It's so that the readers can tell who's being talked about at a glance, not weird at all.

I have to admit, as an Albertan, to the tiniest bit of relief to finally see a different province's name in a headline like this.

Also dread that it's spreading, though.

It comes down to whether you can demonstrate this flaw. If you have a way to show it actually working then credentials shouldn't matter.

If your attempts at disclosure are being ignored then check:

• Am I presenting this in a way that makes me seem like a deranged crazy person?
• Am I a deranged crazy person?

Try to resolve those. If the company you're trying to contact is still send your emails to the spam bin, maybe try contacting other people who have done disclosure on issues like this before. If you can convince them then they can use their own credibility to advance the issue.

If that doesn't work then I guess check the "deranged crazy person" things one more time and move on to disclosing it publicly yourself.

The Coordinated Vulnerability Disclosure (CVD) process:

1. Discovery: The researcher finds the problem.

2. Private Notification: The researcher contacts the vendor/owner directly and privately. No public information is released yet.

3. The Embargo Period: The researcher and vendor agree on a timeframe for the fix (industry standard is often 90 days, popularized by Google Project Zero).

4. Remediation: The vendor develops and deploys a patch.

5. Public Disclosure: Once the patch is live (or the deadline expires), the researcher publishes their findings, often assigned a CVE (Common Vulnerabilities and Exposures) ID.

6. Proof of Concept (PoC): Technical details or code showing exactly how to exploit the flaw may be released to help defenders understand the risk, usually after users have had time to patch.

You say the flaw is "fundamental", suggesting you don't think it can be patched? I guess I'd inform my investment manager during the "private notification" phase as well, then. It's possible you're wrong about its patchability, of course, so I'd recommend carrying on with CVD regardless.

I'm sure this thread will have more than just knee-jerk scary "feels" or inaccurate pop culture references in it, and we'll be able to have a nice discussion about what the technology in the linked article is actually about.

If you believe that Google's just going to brazenly lie about what they're doing, what's the point of changing the settings at all then?

In fact, Google is subject to various laws and they're subject to concerns by big corporate customers, both of which could result in big trouble if they end up flagrantly and wilfully misusing data that's supposed to be private. So yes, I would tend to believe that if the feature doesn't say the data is being used for training I tend to believe that. It at least behooves those who claim otherwise to come up with actual evidence of their claims.

You are being sarcastic but this is indeed the case. Especially for companies like Google, which are concerned about being sued or dumped by major corporations that very much don't want their data to be used for training without permission.

There's a bit of a free-for-all with published data these days, but private data is another matter.

Yes, they are. Not sure why you are bringing that up.

I am bringing it up because the setting Google is presenting only describes using AI on your data, not training AI on your data.

Yes, exactly. Training an AI is a completely different process from prompting it, it takes orders of magnitude more work and can't be done on a model that's currently in use.