Most active commenters
  • spogbiper(6)

←back to thread

95% of Companies See 'Zero Return' on $30B Generative AI Spend

(thedailyadda.com)
418 points speckx | 14 comments | 21 Aug 25 15:36 UTC | HN request time: 0.919s | source | bottom
Show context
IgorPartola ◴[21 Aug 25 16:28 UTC] No.44974700[source]▶
What are the actual use cases that can generate revenue or at least save costs today? I can think of:

1. Generate content to create online influence. This is at this point probably way oversaturated and I think more sophisticated models will not make it better.

2. Replace junior developers with Claude Code or similar. Only sort of works. After all, you can only babysit one of these at a time no matter how senior you are so realistically it will make you, what, 50% more productive?

3. Replace your customer service staff. This may work in the long run but it saves money instead of making money so its impact has a hard ceiling (of spending just the cost of electricity).

4. Assistive tools. Someone to do basic analysis, double check your writing to make it better, generate secondary graphic assets. Can save a bit of money but can’t really make you a ton because you are still the limiting factor.

Aside: I have tried it for editing writing and it works pretty well but only if I have it do minimal actual writing. The more words it adds, the worse the essay. Having it point out awkward phrasing and finding missing parts of a theme is genuinely helpful.

5. AI for characters in video games, robot dogs, etc. Could be a brave new frontier for video games that don’t have such a rigid cause/effect quest based system.

6. AI girlfriends and boyfriends and other NSFW content. Probably a good money maker for a decade or so before authentic human connections swing back as a priority over anxiety over speaking to humans.

What use cases am I missing?

replies(12): >>44974774 #>>44974790 #>>44974807 #>>44974858 #>>44974923 #>>44975121 #>>44975152 #>>44975248 #>>44975302 #>>44975853 #>>44975875 #>>44976266 #
1. spogbiper ◴[21 Aug 25 16:40 UTC] No.44974858[source]▶
I am working on a project that uses LLM to pull certain pieces of information from semi-structured documents and then categorize/file them under the correct account. it's about 95% accurate and we haven't even begun to fine tune it. i expect it will require human in the loop checks for the foreseeable future, but even with a human approval of each item, its going to save the clerical staff hundreds of hours per year. There are a lot of opportunities in automating/semi-automating processes like this, basically just information extraction and categorization tasks.
replies(5): >>44975000 #>>44975079 #>>44975108 #>>44975215 #>>44975349 #
2. beepbooptheory ◴[21 Aug 25 16:49 UTC] No.44975000[source]▶
How will you know in practice which 5% is wrong?
replies(1): >>44975052 #
3. spogbiper ◴[21 Aug 25 16:52 UTC] No.44975052[source]▶
the system presents a summary that a human has to approve, with everything laid out to make that as easy as possible, links to all the sources etc
4. kjkjadksj ◴[21 Aug 25 16:54 UTC] No.44975079[source]▶
Isn’t that something you can do with non ai tooling to 100% accuracy?
replies(1): >>44975190 #
5. systemerror ◴[21 Aug 25 16:55 UTC] No.44975108[source]▶
The big issue with LLMs is that they’re usually right — like 90% of the time — but that last 10% is tough to fix. A 10% failure rate might sound small, but at scale, it's significant — especially when it includes false positives. You end up either having to live with some bad results, build something to automatically catch mistakes, or have a person double-check everything if you want to bring that error rate down.
replies(2): >>44975162 #>>44976001 #
6. spogbiper ◴[21 Aug 25 16:59 UTC] No.44975162[source]▶
yes, the entire design relies on a human to check everything. basically it presents what it thinks should be done, and why. the human then agrees or does not. much work is put into streamlining this but ultimately its still human controlled
replies(1): >>44975256 #
7. spogbiper ◴[21 Aug 25 17:01 UTC] No.44975190[source]▶
in some similar cases yes, and this client has tried to accomplish that for literally decades without success. i don't want to be too detailed for reasons, but basically they cannot standardize the input to the point where anything non AI has been able to parse it very well.
8. whatever1 ◴[21 Aug 25 17:03 UTC] No.44975215[source]▶
All of the AI projects promise that they just need some fine tuning to go from poc to actual workable product. Nobody was able to fine tune them.

Sorry this is some bull. Either it works or it doesn’t.

9. wredcoll ◴[21 Aug 25 17:05 UTC] No.44975256{3}[source]▶
At the risk of being obvious, this seems set up for failure in the same way expecting a human to catch an automated car's mistakes is. Although I assume mistakes here probably don't matter very much.
replies(2): >>44975394 #>>44975953 #
10. LPisGood ◴[21 Aug 25 17:11 UTC] No.44975349[source]▶
> its going to save the clerical staff hundreds of hours per year

How many hundreds of hours is your team spending to get there? What is the ROI on this vs investing that money elsewhere?

replies(1): >>44976036 #
11. LPisGood ◴[21 Aug 25 17:15 UTC] No.44975394{4}[source]▶
This reminds me the issue with the old windows access control system.

If those prompts pop up constantly asking for elevated privileges, this is actually worse because it trains people to just reflexively allow elevation.

12. spogbiper ◴[21 Aug 25 17:58 UTC] No.44975953{4}[source]▶
yes, mistakes are not a huge problem. they will become evident farther down the process and they happen now with the human only system. worst case is the LLM fails and they just have to do the manual work that they are doing now
13. f3b5 ◴[21 Aug 25 18:01 UTC] No.44976001[source]▶
Depending on the use case, a 10% failure rate can be quite acceptable. This is of course for non-critical applications, like e.g. top-of-funnel sales automation. In practice, for simple uses like labeling data at scale, I'm actually reaching 95-99% accuracy in my startup.
14. spogbiper ◴[21 Aug 25 18:03 UTC] No.44976036[source]▶
Can't speak to the financial benefit over other investment. Total dev/testing time looks to be fairly small in comparison to time saved in even one year, although with different salaries etc I cannot be too certain on the money ratio. Ultimately not my direct concern, but those making decisions are very happy with results so far and looking for additional processes to apply this type of system to.