Live data from Hacker News

Slack AI Training with Customer Data

slack.com

241–250 of 426 posts

Re: Slack AI Training with Customer Data

#241
post #234
post #227

Well I really hope this massively blows up in their face when all of Europe goes to work just about now, and then North America in 5-8 hours. Let's see if we have another Helldivers 2 event that makes them do a hard backpedal after losing thousands of large customers that will not under any circumstances take the chance. I have a friend with a law firm who just called me yesterday for advice as he's thinking about sw…

But you can opt out, right? So what’s the problem? Also, is Teams (and other messengers) any different?

and how does that even work? Slack is a chat app. Does everyone involved in the chat need to opt out for it to be meaningful? What about bots?

Re: Slack AI Training with Customer Data

#242
post #239
post #234

Earlier quoted context omitted.

But you can opt out, right? So what’s the problem? Also, is Teams (and other messengers) any different?

> But you can opt out, right? So what’s the problem? This thinking is the problem. "Oh, we just added your entire private/privileged/NDA/corporate information to our training set without your consent. What's the problem?" Opt-out must be the default . Edit: By "Opt-out must be the default ." I mean: no one's data must be included until they explicitly give consent via an opt-in :)

Especially since once it has been trained, it is in the model, and I am not aware of any way anyone has discovered to later remove from the model single or selected training data points, except for re-training/re-learning the model. So basically the crime might already be done.

But I also know that so many businesses are too sluggish to make a switch and employees incapable of understanding the risk. So unfortunately not all of Europe will switch away. But I hope a significant number gives them the middle finger.

Re: Slack AI Training with Customer Data

#243
post #239
post #234

Earlier quoted context omitted.

But you can opt out, right? So what’s the problem? Also, is Teams (and other messengers) any different?

> But you can opt out, right? So what’s the problem? This thinking is the problem. "Oh, we just added your entire private/privileged/NDA/corporate information to our training set without your consent. What's the problem?" Opt-out must be the default . Edit: By "Opt-out must be the default ." I mean: no one's data must be included until they explicitly give consent via an opt-in :)

> Opt-out must be the default.

Don't you mean opt-in must be the default?

Or am I misunderstanding the concept of opt-ins :P

Re: Slack AI Training with Customer Data

#244
post #239
post #234

Earlier quoted context omitted.

But you can opt out, right? So what’s the problem? Also, is Teams (and other messengers) any different?

> But you can opt out, right? So what’s the problem? This thinking is the problem. "Oh, we just added your entire private/privileged/NDA/corporate information to our training set without your consent. What's the problem?" Opt-out must be the default . Edit: By "Opt-out must be the default ." I mean: no one's data must be included until they explicitly give consent via an opt-in :)

Exactly! Allowing access to your data should only be opt-in

Re: Slack AI Training with Customer Data

#245
post #239

Earlier quoted context omitted.

> But you can opt out, right? So what’s the problem? This thinking is the problem. "Oh, we just added your entire private/privileged/NDA/corporate information to our training set without your consent. What's the problem?" Opt-out must be the default . Edit: By "Opt-out must be the default ." I mean: no one's data must be included until they explicitly give consent via an opt-in :)

> Opt-out must be the default. Don't you mean opt-in must be the default? Or am I misunderstanding the concept of opt-ins :P

Opt-in is "I agree to have my data included"

Opt-out is "I don't agree to have my data included"

Re: Slack AI Training with Customer Data

#247
post #234
post #227

Well I really hope this massively blows up in their face when all of Europe goes to work just about now, and then North America in 5-8 hours. Let's see if we have another Helldivers 2 event that makes them do a hard backpedal after losing thousands of large customers that will not under any circumstances take the chance. I have a friend with a law firm who just called me yesterday for advice as he's thinking about sw…

But you can opt out, right? So what’s the problem? Also, is Teams (and other messengers) any different?

Defaults matter!

Just look at how much Apple and Mozilla get from Google by having their browser as a default (ca. $20,000,000,000 and $400,000,000 IIRC per annum).

Or look at how many people rejected the tracking prompt displayed for FB when it was added to iOS (+70%).

Re: Slack AI Training with Customer Data

#248
post #239

Earlier quoted context omitted.

> But you can opt out, right? So what’s the problem? This thinking is the problem. "Oh, we just added your entire private/privileged/NDA/corporate information to our training set without your consent. What's the problem?" Opt-out must be the default . Edit: By "Opt-out must be the default ." I mean: no one's data must be included until they explicitly give consent via an opt-in :)

> Opt-out must be the default. Don't you mean opt-in must be the default? Or am I misunderstanding the concept of opt-ins :P

Opt-out by default = Opt-in.

Opt-in by default = Opt-out

Re: Slack AI Training with Customer Data

#249

Story time. I was at a VC conference last year and if I learned nothing else there, I learned how to spell "AI". Every single exhibitor just about had their signage proudly proclaiming their capabilities in this area, but one in particular struck me. They were touting the API integrations they could offer to train their "Enterprise AI"/LLM, and among those integrations were things like M365, Slack, etc. It struck me…

The more obvious things are that it's not training llms fully on all channels.

Some quick ideas:

Search and summarize other messages. No new llms training and about mostly linking to existing answers.

Fine tune on your messages, but only customer support messages in the public channel, not "eng-shitpost"

Natural language requests over your company data.

Post reply on HN