Like the direction of thought here, don’t think Gmail or any major email for that matter could be leveraged to train a public model, far too risky because of the nature of all the private and confidential information in there.
That said, what G and M are both doing is using it at the user level, which will continue to give them an edge. Google has your search history, your GPS, your browsing history, your app history (G apps) , among other things. Add emails to that list and that training dataset is hard to replicate for anyone else as a leverage for their LLM. The delta is too big to travel.
I agree here Nick. AI is moving so quickly that it's hard to look far down the road sometimes, but this fundamental shift is going to upend the future of the web. A decade ago, we had an unspoken agreement with Google that we would give them information in exchange for traffic, and now that agreement is null and void. Companies everywhere are going to find ways to protect their IP while still scouring for prospects.
OK, walk with me for a minute. What's stopping AI from subscribing to the paid sites to get around the paywalls? They don't need 10,000 licenses... just 1 license will do.
Or, for a slower, but still plausible, approach... a Tesla robot with AI, sitting behind a Macbook Pro, with a paid subscription, browsing behind the paywalls?
So happy to see you here William. My first thought was "Hey, at least at that point, they are paying something to access the data." Maybe, just maybe, future subscriptions even include their own TOS explicitly stating no repurposing of content?
Like the direction of thought here, don’t think Gmail or any major email for that matter could be leveraged to train a public model, far too risky because of the nature of all the private and confidential information in there.
That said, what G and M are both doing is using it at the user level, which will continue to give them an edge. Google has your search history, your GPS, your browsing history, your app history (G apps) , among other things. Add emails to that list and that training dataset is hard to replicate for anyone else as a leverage for their LLM. The delta is too big to travel.
I think you are right. I hope it truly doesn’t become a LLM vs the world… we need to find a happy medium.
The day Google switches primary default experience to AI Mode is the day of reckoning.
I agree here Nick. AI is moving so quickly that it's hard to look far down the road sometimes, but this fundamental shift is going to upend the future of the web. A decade ago, we had an unspoken agreement with Google that we would give them information in exchange for traffic, and now that agreement is null and void. Companies everywhere are going to find ways to protect their IP while still scouring for prospects.
It will be very interesting for sure. I think we'll see a ton more changes before we land on what will be "closer" to the new norm.
Truly excellent thought experiment here, Nick.
OK, walk with me for a minute. What's stopping AI from subscribing to the paid sites to get around the paywalls? They don't need 10,000 licenses... just 1 license will do.
Or, for a slower, but still plausible, approach... a Tesla robot with AI, sitting behind a Macbook Pro, with a paid subscription, browsing behind the paywalls?
So happy to see you here William. My first thought was "Hey, at least at that point, they are paying something to access the data." Maybe, just maybe, future subscriptions even include their own TOS explicitly stating no repurposing of content?