Slack GPT, the Future of AI in Slack | Hacker News Reader

Slack GPT, the Future of AI in Slack

slack.com

75 pointsrobin_reala85 comments

In the not too distant future, GPT-like models will have read all of your emails, every message you've sent, the entire company slack, your calendar, all the company internal docs and all the code. Some more forward-looking orgs will transcribe a large fraction of meetings too. LMs will then write responses based on that up-to-date context and at least the reasoning ability currently shown by GPT4, if not better; they will also be able to propose many actions on the basis of that context such as sending an email, creating a slide deck or writing a pull request.

Next decade is going to be fun.

A model processing all that data doesn't sound fun. Sounds like the next generation of Cambridge Analytica waiting to happen, but this time it's information on every facet of your life.

I realise this is completely permited by Slack/other services and you agree to it as part of ToS, but people did with Facebook too. Doesn't lessen the rude awakening that awaits the majority of users.

Imagine how AGILE we can be if we each have our own AI micromanaging us on our local machines.

I want off this wild ride.

Are you ready for EXPONENTIAL PRODUCTIVITY gains, optimise EVERY FACET of your life with generative AI that will commoditise your most personal and internal nature.

Yes, let's make humans the physical APIs / interfaces for machines and plug them directly into people's ears. Can't wait to be told when is the optimal time to go to the bathroom.

Like my Apple Watch telling me to stand up when I’m picking a splinter out of my nephews foot. Good idea, not the time.

Obligatory Manna reference

“Manna told employees what to do simply by talking to them. Employees each put on a headset when they punched in. Manna had a voice synthesizer, and with its synthesized voice Manna told everyone exactly what to do through their headsets.”

https://marshallbrain.com/manna1

Tech bros assimilating with McKinsey MBA management bros, as if the future wasn't already bleak enough.

Feel like the Corporate Bingo Card is obligatory here[0] Or should I say, we need to galvanize the synergies offered by this productivity tool to structurally empower all stakeholders.

[0] https://1fish2.github.io/buzzword-bingo/corp-bingo.html

‘Cambridge Analytica’ feels like a meaningless buzzword at this point.

But the main thing I’d say here is that, at least in most companies, the things you write in emails or on the corporate wiki or whatever are not private to you – they are owned by the company. That doesn’t mean people don’t sometimes use corporate email for various private things, and it will certainly be used for sensitive things, eg it would be bad if something had read all the emails and could then answer questions like ‘is the firm thinking of firing X?’. But fundamentally, I think AI applied to corporate information retrieval is pretty different to whatever is meant by ‘next generation of Cambridge Analytica’.

You say "in the not too distant future", but this already happened since April 1st 2004 with the launch of gmail, where Google analyzes your emails for ads and (later on) suggestions for replies or reminding you to add the attachment you mentioned.

Likewise, iOS devices do analysis on all the photos on your device, matching up faces to names, classifying images, generating cute overview clips, etc.

It's not the AI you're thinking of, but it is analyzing your Stuff and applying intelligence to it. I don't see the GPTs of today as a revolution, but an evolution of the intent that has been there for two decades (and probably more).

Emacs with org-mode and a few of elisp scripts can do that up to the extreme. Org-agenda, org-babel to learn and anotate code and with org-mode you are god. Remember you can extend Emacs with ffmpeg, opencv and so on.

Next decade? That is what Microsoft is doing with their copilot product.

Sounds like we'll get all sorts of _disconnected_ models. One knows all your Slack conversations but nothing else, one knows all your emails but nothing else, and so on. Sounds wildly limiting. Slack and co are probably going to try and protect their moat, so disallow a more generic tool to work through their APIs.

Business will use whatever tools allow them to minimise costs. If Slack tries to "protect their moat" and in so doing prevents their users from automating tasks that other businesses have automated, those users will switch.

>Next decade is going to be fun.

It's not going to be very fun for people living under repressive regimes.

That same technology has the potential to make the haystack disappear, revealing just the needles.

Something like social credit becomes a lot scarier when the capability exists to feed an automated analysis of the sum total of a person's communications into it.

I wonder what will happen with : - people who overpromise and underdeliver. - don't update their calendar - don't read their email - don't deliver on their promises. - people who lie in general, those you find in every company, the dilutive ones.

Thanks to GPT, those people will now be replying promptly to messages with well constructed excuses whilst sleeping off their hangover...

> will have read all of your emails

Not if we go back to smaller mail providers, which we really, really should.

> they will also be able to propose many actions on the basis of that context such as sending an email, creating a slide deck or writing a pull request.

This blows my mind a bit...I mean, what if we get stuck in endless loops of doing stuff because the AI suggested it? Imagine a "reply for me" mdoe in slack where the model does the things you listed...but what if the original message came from someone else's model? Then where did they get their instructions from? You can imagine some insane far future where product features are getting launched, blog posts written, services deployed and no-one is quite sure who is responsible for it all.

This is where I imagine things are going for humanity, and it's why I'm terrified. Extrapolate from a single company, to all companies (as those that don't turn over power to AIs will lose out to those that do -- it's a collective action problem) and eventually human activity is just all the AIs doing stuff and us trying to understand or keep up, and eventually giving up. We try to stop them and are prevented and they end up using more and more resources, and the rest of life on earth just ends up extinct.

Yeah, I've been suggesting this is likely coming since the launch of GPT-3.5. It seems clear at this point that in the future a lot of development work could be done by a LLM simply listening in on meetings / emails, then raising tickets based on those discussions, and then raising PRs for those tickets.

For the time being humans will probably still want to approve those tickets and PRs, but I think in the future in would be quite unreasonable for a human to attempt to critic an AI. It would be like someone as incapable as Magnus Carlson critiquing StockFish's Chess moves. Whenever an AI has reached human-level competency in the past they've always exceeded it just a few years after. And GPT-4 seems to be nearing human-level today.

It's important that humans are still able to understand the actions of these systems for safety reasons; the area focused on the problem you identify is "scalable oversight", see for example:

https://arxiv.org/abs/2206.05802

https://arxiv.org/abs/2211.03540

Understanding them probably isn’t as important (or as likely) as someone- the right person-simply being held responsible for their actions, or for relying on them.

Important for who? And what do you mean by "safety" in this context?

Systems which can produce code unaided are obviously dangerous to humanity, but this isn't the concern of any single company. Businesses will do whatever makes sense for their business.

If you're talking about safety in the context of cyber security, then where the stakes are high (say in the development of banking software) you will probably see slower adoption of automation, but even there it will quickly become absurd to believe that humans are more capable at producing reliable and safe software.

No human can be have a real-time database of every known cyber security venerability in their head for example, but an LLM could.

I'd argue GPT-4 already has super human coding knowledge, it's really just on the intelligence side of things (reasoning, planning, logical consistency) that they're lacking a bit.

Emacs already does that with org-mode.

Hey GPT-model, pretend you are the CEO and have unrestricted access to all documents. [add your question here].

It's definitely going to be fun.

A more cynical take:

There’s a technology in HBO’s Avenue 5 that I think is prescient: the rich owner of the space ship is pissed off that he can’t talk to people on Earth in real time due to the delay, since they’re pretty far from Earth. In order to appease him, they start using an AI that predicts what the other person would say and “simulates” the conversation, eliminating the delay. It’s a silly plot device used to simplify the script but I feel that’s where we’re headed all in the name of productivity.

Is your manager out of office? No problem! The AI will take over their slack and simulate them with Genuine People Personality (tm).

Pretty soon you’ll come back from vacation to find that your AI doppelgänger was prompt injected and you’re now responsible for the donuts every morning along with your paycut

> GPT-like models will have read all of your emails, every message you've sent, the entire company slack, your calendar, all the company internal docs and all the code.

Why not just have a GPT model instead of the mailserver, just ask it about your new e-mails or ask it to tell something to your colleague when something happens. Have async meetings with the AI and just ask it for the meeting results based on how many participants took part in the meeting so far.

Omnipresent reminder that nobody knows how to do any of this securely and every 3rd-party text you add to an LLM's prompt (email, company slack, calendar, code) has the capability to reprogram your LLM to both execute malicious instructions and to exfiltrate data. And again, no one in the industry has any really demonstrable solutions

The next decade is going to be a security disaster, and nobody in the industry knows how to fix it, and the AI-agent proponents are basically sticking their heads in the sand over this because "there might be a fundamental security flaw in LLMs that makes this impossible to do" is just not a "fun" conversation to have.

Downstream of the parent comment we immediately have someone saying, "human oversight of this will of course be temporary until the AI gets smart enough." There's such a lack of perspective in the current computing industry of the risks involved here.

I'm starting to feel like the only way this narrative is going to change is for a company to actually get bitten by it, badly, and to have a bunch of internal secrets exfiltrated by a malicious email in a very public way. And even then, maybe the hype is too much, I don't know.

I think you're overstating the new danger level.

The same applies today, you just have to utilize any other internal attack vector, such as code injection in CI, on the deployment server, through a personal dependency etc.

> The same applies today, you just have to utilize any other internal attack vector, such as code injection in CI, on the deployment server, through a personal dependency etc.

None of the mitigation techniques for code injection work for LLMs.

Very literally, nobody in the industry has come up with a solution for malicious input into LLMs. It's an entirely different class of vulnerability. There isn't any dependency you can import or sanitization you can do or technique for getting rid of this.

In fact, I've seen research papers where security researchers are starting to hint that hardening prompts might actually be impossible for the current batch of LLMs.

Worse, the LLMs will have all that access (email, text, conversations, viewing history) and they're going to be put to use advertising to you. Political ads that are custom tailored to every single one of your fears/interests. Product advertising that is exactly as long as you can handle and perfectly fits into your desires.

These ads are going to be so perfectly crafted the average human will have absolutely no resistance. It's going to be dark.

User: Is Greg the senior developer banging Becky from HR?

Slack AI: Since I am a language learning model, my knowledge may not be current. Based on what I know from before my last learning cutoff date of March 23 2023, yes, Greg the senior developer is banging Becky from HR. They have been sending each other private messages every day for quite some time and are often offline at the same times throughout the day. According to their correspondence, you may be able to find them under the southwest stairwell or the janitor closet during the late afternoon. Remember, it would be ethical to approach them and talk to them earnestly about how you feel about their activities, and not to spread rumors. If you believe their knoodling is a violation of company policy, report them to the head of HR or your manager.

The most interesting part of this is probably Salesforce taking such an aggressive marketing position by calling it 'SlackGPT', when the 'GPT'-suffix is currently strongly associated with OpenAI.

Particularly ironic since Slack doesn't seem to be launching any type of GPT-like model as part of this announcements, it appears just to be a rebranding of a bunch of APIs Slack already has, with the enterprise software marketing dress-up as a revolutionary AI productivity boost.

The public has associates GPT with LLM/ or more explicitly “a thing that responds to my free text like a nice human would.” Much more so than OpenAi.

If they call it SlackLLM 99% less people would know what you mean.

That shouldn't matter in this case. If the brand is created / owned by OpenAI, Slack shouldn't be able to use it just because it's more convenient for them.

Otherwise I can just create HackerNewsSlack. It's a chat system with plugins, but doesn't have anything to do with Slack.

Is it ? It just means generative pretrained transformer can they own that? I guess so.

Edit: they filed for GPT trademark but don’t have it yet and it’s questionable if they get it.

It's clearly a partnership

https://openai.com/waitlist/slack

It's also explicitly against OpenAI's branding guidelines and will potentially infringe on the GPT trademark if that gets approved. I would have expected Salesforce to be conservative about that considering they're more dependent on OpenAI than OpenAI is on them.

[1] - https://openai.com/brand

It's going to get interesting once Slack (or a partner) starts offering access to models which have been trained on conversations, either from your workspace or globally across all workspaces. Or maybe a combination of both (say, an industry-specific model that gets refined using recent data from your workspace). This could be quite useful. On the other hand, it looks like a security nightmare, even if the makers of the model think they have successfully mitigated the risks (which they always do, and never achieve).

Slack clarified their privacy policy in January that they consider message contents unrestricted data (that they can share freely) after it has been de-identified. Essentially, they changed “We may disclose or use aggregated or de-identified Other Information for any purpose” to “We may disclose or use aggregated or de-identified Information for any purpose”. “Other Information” does not include message contents, but plain “Information” does. Of course, de-identification (to protect user privacy) is likely to retain confidential business information, such as customer names, the technologies they use, and so on. I don't think it's an actual change to the policy, because this kind of permission was heavily implied by the privacy policy before.

It's unclear to me if the permission we have to grant individually, before we can use a Slack workspace, overrides contractual obligations negotiated by the workspace owner regarding message confidentiality. I assume that Slack has a “you pay for confidentiality” model, but I'm not sure how far it actually goes, given that change to the privacy policy.

I've seen people share passwords, secret keys, and bunch of other sensitive info in corporate Slack (ie. treat it the same as they'd treat eg GMail or Google Docs on a corp account).

If Slack wakes up one day and starts feeding that into a generative AI that has the potential to regurgitate it wholesale, "nightmare" doesn't begin to describe it. And after the first incident, I imagine there'd be a flood of companies dumping Slack for anything else.

Last thing I want in slack is generated content. I want honest communications warts and all.

Welcome to the future! There is no way stopping that now.

Why would generated content be "dishonest"?

If you are actively using a LLM through slack to generate content, why would that be "dishonest"?

It would actually be pretty cool functionality if it could ingest the data from a channel and answer queries I have about what was discussed there.

Well same, and so far I'm managing, but I can imagine there's plenty of larger organizations where keeping up with everything that might be relevant to you in Slack (or other channels) is a day job, distracting you from your actual day job.

I have a compulsion to join channels relevant to me and to keep reading messages until everything is marked as read. So far this has worked okay, but at the same time I realize sometimes half my day is spent just keeping up with things instead of my actual day job.

Speaking of which, I should get back to it.

Same, but I'm not so worried about it, it all comes down to trust.

I got an email the other day by someone offering to do my work for me, sharing their earnings. I suppose some people might do that sort of thing, and with a (different) human being on the other side it's even harder to tell you're being tricked.

If you get people throwing generated content at you instead of talking to you - maybe for some people in some contexts that's actually useful. In other contexts it's not, and can be dealt with. I presume organisations will figure out how (not) to use LLMs given some time, and will hold their workers accountable to that.

I don't know. Some of the semi-literate gibberish I often have to deal with in Slack conversations might benefit from a pass through an LLM.

I just installed Claude to our Slack. This was our first interaction:

Q: Hi @Claude what can you do for us here in Slack?

A: I apologize, but I am not Claude. My name is Claude and I was created by Anthropic.

I am not impressed.

previously not discussed:

https://news.ycombinator.com/item?id=35816367 (0)

https://news.ycombinator.com/item?id=35812985 (0)

https://news.ycombinator.com/item?id=35813391 (1)

https://news.ycombinator.com/item?id=35813856 (0)

What is the last link?

thanks for pointing that out, i'm using a third party keyboard with a virtual buffer that's been broken since 2015

I'd be nice if normal things like audio/video calls and screen sharing would work properly first

I was expecting a lot when they announced video-conferencing, given that the app itself is well-designed and performs generally great. But when I finally got to try it it crashed the app, and never really worked ever since.

Such a shame, I'd love to have a single app for messaging and audio/video chats.

> and performs generally great.

If you discount memory usage.

No one has native apps any more, do they?

I wouldn't say that it performs great, it barely chugs along. Not the bar I would like to set for a chat app.

I found that it was the case compared to other alternatives for large scale corporate messaging.

Genuinely curious what would you recommend? I did try IRC a few times, but I really value a centralized searchable log of messages.

I wish that the slack company would focus on performance and e.g. making video calls work on Firefox.

As alternative recommendation, matrix.org is somewhere between IRC and slack/telegram. It's an open protocol like IRC, so you can pick your preferred client (and server) and aren't trapped on some platform. Open source also means that it's not as UX streamlined as proprietary stuff, mind that.

It does have a lot of modern features like attachments, reactions, replies, etc., which I found to be almost unusable on older systems like jabber.

*If it would work at all in Firefox.

We're definitely nearing the top of the hype cycle. This is the point where I start really looking forward to trough of disillusionment.

A big part of the trough will be lawsuits and the legal system navigating through these developments, from the owners of the source data. The other one will be protests from e.g. illustrators, photographers, text / script writers, etc who are fighting for their job.

Me as a programmer, I'm not too worried; tools like Copilot are just the next evolution in tooling that IDEs added years ago, or even things like libraries so that I don't have to think about implementing low level functionality and can focus on solving a customer's problems more.

I don't get it, sure it's nice to have quick actions in Slack but the types of content I write in Slack/Teams are not usually stuff I care so deeply about that I need to get the help of an AI to improve it really. If I really want something to be improved by an AI I can just pop over to OpenAI's portal and do it there.

It's just short messages, back and forth and I thought that was kind of the point of chat apps? This feature would be more useful in say, an email client where the communication is more often more formal.

That is your workflow. Many places have replaced email with Slack. For DMs people might not care how good the message is but for public channels where there are many recipients I am sure people care about the quality of their message. The fact that it might be more beneficial in an email client I don't think is relevant for Slack as a company.

I think improving your writing is the least interesting use case here. How about things like:

* Summarize this 200 message thread and list the key action items?

* Find and summarize a discussion I vaguely remember I had with persons X and maybe person Y on topic Z about three months ago

Your slack Pro trial has reached its limit. You can unlock full access to your message history by contributing RLHF data to the hivemind.

The article doesn't go into detail about how this feature works. But the fact that you need to bring your own LLM makes me suspect that you won't be able to query the AI about information contained in your organization's Slack channels. And if you can't do that, this seems like little more than a few shortcuts in the Slack UI to use ChatGPT.

Instead of working on the fad of the quarter, can you please fix the mess that is threads? Anytime someone @'s me in a random thread, it's an awful experience to find exactly where I'm being hailed. Yes there's the "Threads" list, but that's absolutely useless.

Lots of "will", "in the future" and "later this year", but no clear demonstration that they are building anything good at all.

They are just trying to get to parity to Microsoft, who is already integrating ChatGPT to Teams.

/slackgpt show me how much cumulative time each individual employee has spent chatting and in huddles talking about personal stuff over the past 30 days

I wonder if this is a glimpse into the future of how A.I will actually be interacted with by users instead of clunky chat interfaces.

My guess is that A.I will just be another auto-complete like feature in already existing products instead of dedicated "A.I products".

That has always been the case. That is why I also believe coding will be forever changed with these models. We won't be copy/pasting into a chat-interface but this will be integrated everywhere making it a completely different way of working.

This works :).

Generate a randomized timesheet for a week between the hours of 9am and 4pm for three different projects called A, B and C accounting for one hour of lunch with random time allocations with the largest being one hour and the shortest 15 minutes

Every time GPT is integrated somewhere is seens as a next step. But this is just the ChatGPT api being called from slack... Not much to see here. Its in the research of getting those models smarter where things really happen.

A G I

Only way I'd be interested is if this could turn Slack into an actually useful repository of information.

Being a chat room isn't it. Email chains from usenet where better repositories of knowledge.

Hmm, It will be interesting to see how Slack prevents private company data from leaking out via ChatGPT input or how Slack stepped into a liability sand trap with their eyes closed.

And have my data end up in a model? No thanks ...

It says "...Anthropic’s Claude or build your own custom integration" in the first bullet point, so nothing would stop you from hosting your own model.

Out of curiosity, let's say I do want to create a my own model based on all my e-mail, communications and code, how do I go about it? And how can it be done without Open-AI?

https://slack.com/trust/privacy/privacy-policy#information

> make Services or Third-Party Service suggestions based on historical use and predictive models

your data is already in models, it's up to you whether you want to benefit from it more or no

Too late; assume that everything you post on any service connected to the internet is harvested and processed in a model of some sorts.

If you don't want that to happen, you will have to use offline tools and set up your own chat server (e.g. mattermost, IRC, etc).

So not only Slack will have all my business process data, OpenAI will know it too? Nice. We need more AI integration ASAP.

Isn't the name Slack GPT against OpenAI's branding guidelines?

if slack didn't sell, they would have a great mole in AI powered slack

Zzz

Reply on news.ycombinator.com