Hacker News Reader

Comment by paxys | Hacker News Reader

> Data will not leak across workspaces.

> If you want to exclude your Customer Data from helping train Slack global models, you can opt out.

I don't understand how both these statements can be true. If they are using your data to train models used across workspaces then it WILL leak. If they aren't then why do they need an opt out?

Edit: reading through the examples of AI use at the bottom of the page (search results, emoji suggestions, autocomplete), my guess is this policy was put in place a decade ago and doesn't have anything to do with LLMs.

Another edit: From https://slack.com/help/articles/28310650165907-Security-for-...

> Customer data is never used to train large language models (LLMs).

So yeah, sounds like a nothingburger.

View on HN

They're saying they won't train generative models that will literally regurgitate your text, my guess is classifiers are fair game in their interpretation

You are assuming they're saying that, because it's one charitable interpretation of what they're saying.

But they haven't actually said that. It also happens that people say things based on faulty or disputed beliefs of their own, or people willfully misrprepresent things, etc

Until they actually do say something as explicit as what you suggest, they haven't said anything of the sort.

> Data will not leak across workspaces. 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, memorize, or be able to reproduce some part of Customer Data.

I feel like that is explicitly what this is saying.

The problem is, it's really really hard to guarantee that.

Yes if they only train say, classifiers, then the only thing that can leak is the classification outcome. But these things can be super subtle. Even a classifier could leak things if you can hack the context fed into it. They are really playing with fire here.

yes, i certainly agree with you. i think oftentimes these policies are written by non-technical people

i'm not entirely convinced that classifiers and LLMs are disjoint to begin with

If is also hard to guarantee that, in a multi-tenant application, users will never see other users' data due to causes like mistakes AuthZ logic, caching gone awry, or other unpredictable situations that come up in distributed systems—yet even before the AI craze we were all happy to use these SaaS products anyway. Maybe this class of vulnerability is indeed harder to tame than most, but third-party software has never been without risks.

The OP privacy policy explicitly states that autocompletion algorithms are part of the scope. "Our algorithm that picks from potential suggestions is trained globally on previously suggested and accepted completions."

And this can leak: for instance, typing "a good business partner for foobars is" might not send that text upstream per se, but would be consulting a local model whose training data would have contained conversations that other Slack users are having about brands that provide foobars. How can Slack guarantee that the model won't incorporate proprietary insights on sourcing the best foobar producers into its choice of the next token? And sure, one could build an adversarial model that attempts to minimize this kind of leakage, but is Slack incentivized to create such a thing vs. just building an optimal autocomplete as quickly as possible?

Even if it were just creating classifiers, similar leakages could occur there, albeit requiring more effort and time from attackers to extract actionable data.

I can't blame Slack for wanting to improve their product, but I'd also encourage any users with proprietary conversations to encourage their admins to opt out as soon as possible.

yeah i absolutely agree that even classifiers can leak, and the autocorrect thing sounds like i was wrong about generative (it sounds like an n-gram setup?)... although they also say they don't train LLMs (what is an n-gram? still an LM, not large... i guess?)

> How can Slack guarantee that the model won't incorporate proprietary insights on sourcing the best foobar producers into its choice of the next token?

This is explained literally in the next sentence after the one you quoted: "We protect data privacy by using rules to score the similarity between the typed text and suggestion in various ways, including only using the numerical scores and counts of past interactions in the algorithm."

If all the global model sees is {similarity: 0.93, past_interactions: 6, recommendation_accepted: true} then there is no way to leak tokens, because not only are the tokens not part of the output, they're not even part of the input. But such a simple model could still be very useful for sorting the best autocomplete result to the top.

There are no data leaks in Ba Sing Se.

This reminds me of a company called C3.ai which claims in its advertising to eliminate hallucations using any LLM. OpenAI, Mistral, and others at the forefront of this field can't manage this, but a wrapper can?? Hmm...

Ah yes, the stock everyone believed in and thought it would reach the moon during 2020.

Right? I take Slack and Salesforce at their word. They’re good companies and look out for the best interests of their customers. They have my complete trust.

Reply on news.ycombinator.com