Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Slack AI Training with Customer Data

slack.com

Post reply on HN
Print view View original

191–200 of 426 posts

Re: Slack AI Training with Customer Data

#191

Post by FuckButtons »

chefandy wrote: post #136
zmmmmm wrote: post #43

> For any model that will be used broadly across all of our customers, we do not build or train these models in such a way that they could learn, memorise, or be able to reproduce some part of Customer Data This feels so full of subtle qualifiers and weasel words that it generates far more distrust than trust. It only refers to models used "broadly across all" customers - so if it's (a) not used "broadly" or (b) only…

Nah. Whoever decided to create the reality their counsel is dancing around with this disclaimer is the actual problem, though it's mostly a problem for us, rather than them.

It’s a problem for them if it looses customer trust / customers.

Re: Slack AI Training with Customer Data

#192

Post by __loam »

m463 wrote: post #71
koolba wrote: post #61

> We offer Customers a choice around these practices. If you want to exclude your Customer Data from helping train Slack global models, you can opt out. If you opt out, Customer Data on your workspace will only be used to improve the experience on your own workspace and you will still enjoy all of the benefits of our globally trained AI/ML models without contributing to the underlying models. Why would anyone not opt…

> Why would anyone not opt-out? This is basically like all privacy on the internet. Everyone WOULD opt-out, if it was easy, and it becomes a whack-a-optput game. note how you opt-out (generic contact us), and what happens when you do opt-out (they still train anyway)

Opt out should be the default by law

Re: Slack AI Training with Customer Data

#193

Post by ugh123 »

Whatever the models used, or type of data within accounts this operates on, this clause would be red lined in most of the big customer accounts that have leverage during the sales/renewal process. Small to medium accounts will be supplying most of this data.

Re: Slack AI Training with Customer Data

#194

Post by EGreg »

int_19h wrote: post #166
EGreg wrote: post #8

Earlier quoted context omitted.

Every company that promises "end-to-end encryption" is just pinky-swearing to you also. Like Telegram or WhatsApp

Telegram client is open source, so you can see what exactly happens there when you enable E2EE.

Reproducible builds somehow ?

Re: Slack AI Training with Customer Data

#195

Post by bongodongobob »

nyc_data_geek wrote: post #46

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…

I think what you're missing is assuming that what an LLM "reads" thinks is a true statement. Shitposting is almost like meta slang. I feel like that's a necessary thing for it to train on to truly understand language. I feel like people underestimate the depth LLMs can pick up on.

Re: Slack AI Training with Customer Data

#196

Post by bongodongobob »

nyc_data_geek wrote: post #55
leoh wrote: post #51

Earlier quoted context omitted.

Yes, consider an existing LLM being given “shitpost-y” messages and asking it if there is anything interesting in there. It could probably summarize it well and that could then be used for training another LLM. etc etc

This assumes everything in the training data set is accurate. Sometimes people are wrong, obtuse, sarcastic, etc. LLM's don't have any way of detecting or accounting for this, do they? That output, then being used to train other LLM's, just creates an ouroboros of AI generated dogshit.

The training data doesn't need to be strictly accurate. If it was, you'd just be programming a deterministic robot. The whole point is the feed it actual human language. Giving it shitposts and sarcasm is literally what makes it good. Think of it like 100 people guessing the number of marbles in a jar. Average their guesses and it will be very close. The training data is the guesses, the inference is the average.

Re: Slack AI Training with Customer Data

#197

Post by hackernewds »

afc wrote: post #147
zmmmmm wrote: post #43

> For any model that will be used broadly across all of our customers, we do not build or train these models in such a way that they could learn, memorise, or be able to reproduce some part of Customer Data This feels so full of subtle qualifiers and weasel words that it generates far more distrust than trust. It only refers to models used "broadly across all" customers - so if it's (a) not used "broadly" or (b) only…

Especially when a few paragraphs below they say: > If you want to exclude your Customer Data from helping train Slack global models, you can opt out. So Customer Data is not used to train models "used broadly across all of our customers [in such a way that ...]", but... it is used to help train global models. Uh.

so if I don't want slack to train on _anything_ what do I do? I still suspect everything now

Re: Slack AI Training with Customer Data

#198

Post by hackernewds »

FuckButtons wrote: post #191
chefandy wrote: post #136

Earlier quoted context omitted.

Nah. Whoever decided to create the reality their counsel is dancing around with this disclaimer is the actual problem, though it's mostly a problem for us, rather than them.

It’s a problem for them if it looses customer trust / customers.

if they lose enough, they will "sorry we got caught"

if they don't, they will not do anything

Re: Slack AI Training with Customer Data

#199

Post by hackernewds »

j45 wrote: post #186
zmmmmm wrote: post #43

> For any model that will be used broadly across all of our customers, we do not build or train these models in such a way that they could learn, memorise, or be able to reproduce some part of Customer Data This feels so full of subtle qualifiers and weasel words that it generates far more distrust than trust. It only refers to models used "broadly across all" customers - so if it's (a) not used "broadly" or (b) only…

I'm imagining a corporate slack, with information discussed in channels or private chats that exists nowhere else on the internet.. gets rolled into a model. Then, someone asks a very specific question.. conversationally.. about such a very specific scenario.. Seems plausible confidential data would get out, even if it wasn't attributed to the client. Not that it’s possible to ask an llm how a specific or random comp…

exactly. a fun game to see why it is so hard to prevent this

https://gandalf.lakera.ai/

Re: Slack AI Training with Customer Data

#200

Post by hackernewds »

__loam wrote: post #192
m463 wrote: post #71

Earlier quoted context omitted.

> Why would anyone not opt-out? This is basically like all privacy on the internet. Everyone WOULD opt-out, if it was easy, and it becomes a whack-a-optput game. note how you opt-out (generic contact us), and what happens when you do opt-out (they still train anyway)

Opt out should be the default by law

so upgrades and customer approves everything? slippery slope to over regulation
Post reply on HN