Hacker News Reader

Comment by cj | Hacker News Reader

I'm the "IT guy" at our 30 person startup.

My 2024 stance is "buy every AI add-on and decide whether to keep it next year"

So our team has access to Enterprise ChatGPT, Gemini, Notion AI, Slack AI, and basically every AI add-on in every SaaS platform that offers it as an upsell (Github Copilot, ReadMe AI, etc).

2024 is the year of "I don't know what the hell these AI tools are going to be useful for, so let's buy them all"

2025 will be the year of "Ok, we spent $xxx on all these AI tools last year, is anyone actually using them?"

I predict we'll be canceling a lot of those subscriptions. Which all in cost us over $100/mo per employee.

View on HN

That's a nice idea but it sounds like you must be spending far above $100/m. Back of the envelope suggests you might be spending the equivalent of an additional salary, and in many companies that's a hard thing to justify without any proven value.

100/mo per employee, it says. Roughly 3k/mo, or 36k annualized.

It says "over $100/m", but these addons are all $10-20 USD each, and it implies more than the list. With just a few of the common ones it's easy to reach $125/m and the approach seems to be to get all of them. Up to $200 is quite likely.

That would be 72k USD which is a salary. Admittedly not a software engineer salary, but a significantly above average US salary. Selling the idea of getting another employee is a big deal for most businesses of 30 people.

I think most 30-person businesses would spend $72k in “R&D” if the promise of such investment is deemed worthy

36k in a small company is a non-trivial operating expense, esp. for hype-based tools that won't live there tomorrow.

like, it's a rounding error as IT budgets go, but throwing a bunch of money at something that's already considered dubious is silly. That's 36k in bonuses for Juniors, or a X-Mas party, licenses for software that you'll actually use.

Exactly their point. 2024 is the experiment (which costs money)

2025 is the pick the winners and drop the losers.

And use money for a different experiment.

Yes, and my point is that this is an expensive process that is not accessible to many 30 person companies because it's close to hiring another person, for the benefit of maybe 1-2 tool choices in a year's time.

if ai can make 25 do the work of 30...then it would be worth it.

It's never let 30 do the work of 50. The corporate fantasy is always about firing people.

You can get some AI services and fire some people straight away. Increasing the work from 30 to 50 needs more sales which may or may not happen one day.

Investor pitches are the other way round - all we are going to grow 50x, no we're going to have layoffs.

Someone in the company also might start to ask where the productivity explosion is.

This is exactly how Ive seen things occur. It makes a lot of sense, given some tools are great new additions (coding especially), but others fall flat (IMO, search)

This makes sense for the current add-on wave but expect the launch of independent AI platforms to increase in the coming year.

I don't think this will happen whilst GPUs are so expensive and constrained.

You need to be extremely well capitalised and have a good answer about how to compete against the cloud providers as well as OpenAI, Anthropic etc.

That sounds great. Where I work, the CIO has blocked access to all AI for "security reasons". Any advice on how to counter that claim? I don't even know how to approach it. It feels like a freight company refusing to use the new "train" for shipping because they are loud and big and scary.

Why? That seems like a sensible take tbh. You don’t want some other company having chat logs with all your internal code and strategic documents inside.

Well we already use all sorts of third parties that have all of our data. Do you think Microsoft should be trusted more than OpenAI? Honest question.

OpenAI by default includes transcripts in its training data. You have to explicitly opt out (and trust that the op out actually does anything).

I wouldn’t trust that everyone correctly opts out. If they don’t, the X months from now “tell me about Foo company’s strategic plans” could regurgitate their internal docs.

Even if we don't opt out, it's not plaintext data right? Further, I feel like it's likely to be the tokenized version right?

And as far as strategic plans, I don't think it regurgitates novel, one source information does it? Isn't the regurgitation more like boilerplate, seen many times type stuff?

Has anyone actually been able to prompt a password out of an LLM? That kind of thing should be ignored because it's high entropy and rare afaik. For example, my name is bongodongobob and my password=hunter17. Do you really think someone will be able to pull this fact out of an LLM at some point? That doesn't seem to be the way they work as I understand it.

Yes, it’s the full transcript including whatever data you upload. It’s pretty trivial to prompt an LLM to regurgitate its training data, especially the largest models. Even if it is rare, single dicument instance. You might get some word substitutions, but certainly the gist of the original will come through. Not high entropy passwords, but full documents? Yes.

Give me a prompt that regurgitates a valid windows server license and I'll believe this.

I explicitly said that high entropy things like passwords and license keys are not likely to work due to how information is compressed in the training process. But if you meant the windows server license agreement text:

https://chatgpt.com/share/673431d5-b278-8004-ad16-10529788f5...

Ok well that isn't an example of something rare. It's a public document by a very large and public company that intended that document to be visible.

>all your internal code and strategic documents

are already stored on someone else's computer already, at 99% of organizations. Every cloud offering comes with business data privacy guarantees.

And is that also true for the various AI tool offerings? If they train their AI on your data, how could they guarantee that the AI is not reproducing parts of it for requests of a different customer?

All of the business data agreements stipulate data is not used for training. Even consumer ChatGPT provides an opt-out for training.

Most training data is synthetic now.

Ask them the definition of using it securely. Ask what they need to see/know to allow a service to be added.

Echoing what another commenter said, depending on your industry, the data you deal with, and competitive landscape, what your CIO said may be sensible. Note especially that parent commenter works at a small startup, where security posture is typically more lax, for good business reasons.

I just don't understand the nature of whatever compromise or exposure is. What is the new threat? I don't see it.

One very common risk is simply giving another company your clients data. Your company may have confidentiality contracts with its corporate clients that prevent this. Alternatively, consider for example the situation where you are trying to integrate AI into a customer service process. Your customers may have some legally protected expectation of how their personal data is processed by your company.

Locally hosted AI?

that's a nice strategy for a 39 person startup ! using the tools is indeed the only way to see what works and what doesn't.

Reply on news.ycombinator.com