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

341–350 of 426 posts

Re: Slack AI Training with Customer Data

#341

Post by WhiteNoiz3 »

JackC wrote: post #331
WhiteNoiz3 wrote: post #287

To add some nuance to this conversation, what they are using this for is Channel recommendations, Search results, Autocomplete, and Emoji suggestion and the model(s) they train are specific to your workspace (not shared between workspaces). All of which seem like they could be handled fairly privately using some sort of vector (embeddings) search. I am not defending Slack, and I can think of number of cases where tra…

> the model(s) they train are specific to your workspace (not shared between workspaces) That's incorrect -- they're stating that they use your "messages, content, and files" to train "global models" that are used across workspaces. They're also stating that they ensure no private information can leak from workspace to workspace in this way. It's up to you if you're comfortable with that.

From the wording, it sounds like they are conscious of the potential for data leakage and have taken steps to avoid it. It really depends on how they are applying AI/ML. It can be done in a private way if you are thoughtful about how you do it. For example:

Their channel recommendations: "We use external models (not trained on Slack messages) to evaluate topic similarity, outputting numerical scores. Our global model only makes recommendations based on these numerical scores and non-Customer Data"

Meaning they use a non-slack trained model to generate embeddings for search. Then they apply a recommender system (which is mostly ML not an LLM). This sounds like it can be kept private.

Search results: "We do this based on historical search results and previous engagements without learning from the underlying text of the search query, result, or proxy" Again, this is probably a combination of non-slack trained embeddings with machine learning algos based on engagement. This sounds like it can be kept private and team specific.

autocomplete: "These suggestions are local and sourced from common public message phrases in the user’s workspace." I would be concerned about private messages being leaked via autocomplete, but if it's based on public messages specific to your team, that should be ok?

Emoji suggestions: "using the content and sentiment of the message, the historic usage of the emoji [in your team]" Again, it sounds like they are using models for sentiment analysis (which they probably didn't train themselves and even if they did, don't really leak any training data) and some ML or other algos to pick common emojis specific to your team.

To me these are all standard applications of NLP / ML that have been around for a long time.

Re: Slack AI Training with Customer Data

#342

Post by WhiteNoiz3 »

wolfwyrd wrote: post #301
WhiteNoiz3 wrote: post #287

To add some nuance to this conversation, what they are using this for is Channel recommendations, Search results, Autocomplete, and Emoji suggestion and the model(s) they train are specific to your workspace (not shared between workspaces). All of which seem like they could be handled fairly privately using some sort of vector (embeddings) search. I am not defending Slack, and I can think of number of cases where tra…

The way it's written means this just isn't the case. They _MAY_ use it for what you have mentioned above. They explicitly say "...here are a few examples of improvements..." and "How Slack may use Customer Data" (emph mine). They also... may not? And use it for completely different things that can expose who knows what via prompt hacking.

Agreed, and that is my concern as well that if people get too comfortable with it then companies will keep pushing the bounds of what is acceptable. We will need companies to be transparent about ALL the things they are using our data for.

Re: Slack AI Training with Customer Data

#343

Post by mvkel »

How does one technically opt-out after model training is completed? You can't exactly go into the model and "erase" parts of the corpus post-hoc.

Like when you send an email to feedback@slack.com with that perfect subject like (jeez, really?) what exactly does the customer support rep do on their end to opt you out?

Now is definitely the time to get/stay loud. If it dies down, the precedent has been set.

Re: Slack AI Training with Customer Data

#344

Post by lars_francke »

I contacted support to opt out.. Here is the answer.

"Hi there,

Thank you for reaching out to Slack support. Your opt-out request has been completed.

For clarity, Slack has platform-level machine learning models for things like channel and emoji recommendations and search results. 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. Our published policies cover those here (https://slack.com/trust/data-management/privacy-principles), and as shared above your opt out request has been processed.

Slack AI is a separately purchased add-on that uses Large Language Models (LLMs) but does not train those LLMs on customer data. Slack AI uses LLMs hosted directly within Slack’s AWS infrastructure, so that customer data remains in-house and is not shared with any LLM provider. This ensures that Customer Data stays in that organization’s control and exclusively for that organization’s use. You can read more about how we’ve built Slack AI to be secure and private here: https://slack.engineering/how-we-built-slack-ai-to-be-secure....

Kind regards, Best regards,"

Re: Slack AI Training with Customer Data

#345

Post by skydhash »

codingdave wrote: post #289
metadat wrote: post #190

Earlier quoted context omitted.

But the business will suffer by most likely being less successful due to less cohesive communication.. it's "The Ick" either way.

I'm not sure chat apps improve business communications. They are ephemeral, with differing expectations on different teams. Hardly what I'd label as "cohesive" Async communications are critical to business success, to be sure -- I'm just not convinced that chat apps are the right tool.

From what I’ve seen (not much actually) Most channels can be replaced by a forum style discussion board. Chat can be great for 1:1 and small team interactions. And for tool interactions.

Re: Slack AI Training with Customer Data

#346

Post by mvkel »

zeckalpha wrote: post #141
mvkel wrote: post #121

Earlier quoted context omitted.

Most, if not all SaaS software is multi-tenant, so we've been living in the "will not" world for decades now.

In your experience.

The SaaS business model breaks if you go single tenant except in Fortune 500 enterprise

Re: Slack AI Training with Customer Data

#347

Post by 8372049 »

LelouBil wrote: post #35
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

Yeah, at least if the client is open source you could verify.

"If you can read assembly, all programs are open source."

Sure, it's easier and less effort if the program is actually open source, but it's absolutely still possible to verify on bytecode, decompiled or disassembled programs, too.

Re: Slack AI Training with Customer Data

#348

Post by lxgr »

kepano wrote: post #149
mvkel wrote: post #121

Earlier quoted context omitted.

Most, if not all SaaS software is multi-tenant, so we've been living in the "will not" world for decades now.

That's exactly my point. "File over app"[1] is just as relevant for businesses as it is for individuals — if you don't want your data to be used for training, then take sovereignty of it. [1] https://stephango.com/file-over-app

"File over app" is a good way of putting it!

Something strange is happening on your blog, fwiw: Bookmarking it via command + D flips the color scheme to "night mode" – is that intentional?

Re: Slack AI Training with Customer Data

#349

Post by 8372049 »

whimsicalism wrote: post #25
blhack wrote: post #21

How could this possibly comply with European "right to be forgotten" legislation? In fact, how could any of these AI models comply with that? If a user requests to be forgotten, is the entire model retrained (I don't think so).

> how could any of these AI models comply with that? If a user requests to be forgotten, is the entire model retrained (I don't think so). I don't believe that is the current interpretation of GDPR, etc. - if the model is trained, it doesn't have to be deleted due to a RTBF request afaik. there is significant legal uncertainty here Recent GDPR court decisions mean that this is probably still non-compliant due to the…

> Likely they are just filtering out all data produced in the EEA.

Likely they are just hoping to not get caught and/or consider it cost of doing business. GDPR has truly shown us (as if we didn't already know) that compliance must be enforced.

Re: Slack AI Training with Customer Data

#350

Post by hehdhdjehehegwv »

hyping9 wrote: post #330
hehdhdjehehegwv wrote: post #154

I long ago replaced Slack with Signal chat rooms. You can set an auto delete and it’s all secure to start with. Also, free.

Cool. Now do it for a 3,000 person org with varied tech skills and patience.

There is no way a chat room with 3,000 people is remotely useful to anybody.
Post reply on HN