this post was submitted on 20 Mar 2026
206 points (98.6% liked)
Fuck AI
6318 readers
742 users here now
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Can you come up with better ways to quickly search and summarize massive amounts of data?
Thats what I find their best use case is, and theres no better solution for it, so I use it for that heavily.
But can you actually trust what it outputs?
Hallucinations are a known thing that LLMs struggle with. If you're trusting the output of your LLM summary without validating the data, can you be sure there are no errors in it?
And if you're having to validate the data every time because the LLM can make errors, why not skip the extra step?
Hallucinations aren't relevant as an issue when it comes to fuzzy searching.
Im not talking about the LLM generating answers, Im talking about sifting through vector databases to find answers in large datasets.
Which means hallucinations arent a problem now.
Can you give an example of a task and the industry where you could handle such a high level of fault tolerance? I believe there are some out there, but curious as to yours.
What fault tolerance?
I tell it to find me the info, it searches for it via provided tools, locates it, and presents it to me.
Ive very very rarely seen it fail at this task even on large sets.
Usually if theres a fail point its in the tools it uses, not the LLM itself.
But LLMs often are able to handle searching via multiple methods, if they have the tools for it. So if one tool fails they'll try another.
How do you know it found the right info? How do you know it didn't miss some? Who is verifying the output? This is why I asked for a specific example, to understand your point better.
For instance, if you needed to find a book in a library, and there were an LLM that you asked to locate the section it's in, you would be the one verifying the output by going to that section and finding the book (because presumably that's why you asked). Maybe there is more than one copy of that book, or maybe the LLM tells you the wrong place to look--that's not a big deal, and would have the fault tolerance I'm talking about.
The same way.
The result the LLM produces is a link to the relevant information directly I can click and go to it.
Example would be a giant collection of files, think like, 10gb+ of many pdfs, or more. I want any relevant sections on a topic, it quickly can aggregate on it and give me links I click to open up straight to relevant sections in specific files, and then read rhem.
This stuff comes up a lot in my industry (software dev) as we often inherit huge data pools of legacy documentation on massive codebases.
When I am tasked with fixing 1 small specific piece of the system, it could take me hours to find the specific stuff Im looking for on the (often poorly maintained) docs.
But also vector db setup to map to that data, and an LLM wired up to it, can search it in milliseconds and pull up relevant sections asap, and I can click and dig deeper from there as much as I need.
This sort of "fuzzy searching" vectorization of tokens is what an LLM does very well. Its part of how it produces its output, but you can reverse the process to create search indexes (effectively reversing the data through the LLM to turn the data into deterministic vectors)
Amd its important to note, the fault tolerance you perceive here doesnt apply. Thos specific type of searching with vector DBs will always produce the same results from the same input, everytime. Its deterministic.
Except, imo, AI searching is literally a regression vs other search methods.
I work as a field operations supervisor for an ISP, and we use a GPS system to keep track of our fleet. They've been cramming AI into it, and I decided to give it a shot.
I had a report of a van running a stop sign. The report only had a license plate, so I asked the AI which of the vehicles in my fleet had that plate. And it thought about it and returned a vehicle. So I follow the link to that vehicle's status page, and the license plate doesn't match. Isn't even close.
It's only in recent time that searching has turned into such a fuzzy concept, and somehow AI turned up and made everything worse.
So you can trust AI if you want. I'll keep doing things manually and getting them right the first time.
That sounds like a tooling problem.
Either your tooling was outright broken, or not present.
It should be a very trivial task to provide an agent with a MCP tool it can invoke to search for stuff like that.
Searching for a known specific value is trivial, now you are back to just basic sql db operations.
These types of issues arise when either:
A: the tool itself just gave the LLM bad info, so thats not the LLMs fault. It accurately reported on wrong data it got handed.
B: the LLM just wasnt given a tool at all and you prompted it poorly to give room for hallucinating. You just asked it "who has this license plate" instead of "use your search tool to look up who has this license plate", the latter would result in it reporting the lack of a tool to search with, the former will heavily encourage it to hallucinate an answer.
You don't think AI hallucinations affect your work? What company do you work for? I'm asking so that I can stay as far away from it as possible.
They dont impact it at all, its not relevant to using MCP vector searching for info.
That’s not what LLMs are for. You’re looking for LibreOffice Calc or a SQL query. If you need to process large amounts of data, you could train an ML model for it, but LLMs are specifically for generating text.
RNNoise is excellent at filtering noise from audio. LLMs couldn’t do that.
By 'data' I'm guessing they mean natural text, where something like SQL wouldn't work.
But yeah, most legit use cases are basically MLs trained for a specific purpose.
Well, given that LLMs have been shown to be shit at accurately summarising, I would say that my own, human parsing is a better way to summarise large amounts of information, slow as it may be.
I have not had this experience tbh, Ive found summarizing to be one of the few things they are good at out of the box.
If your LLM summarizes something poorly you probably just fucked something up and got a "shit in, shit out" problem.
Can you conjure up some compelling proof AI is actually any good at this? Because my experience with literally anything I know well enough to provide my own summary of is that it's just about certain to be hilariously incorrect.
What Model Context Protocols have you tried that you had issues with?
Ive found most vector db search MCPs are pretty solid.
Sounds like a legitimate use case, as long as you have lots of fault tolerance (for example, fine if you want a general impression of something, but not great for deciding on medication dosage). The fault tolerance is the kicker here though--I see people using these tools when they can't afford the faults they produce, and sometimes it's fine until it isn't.
There are a handful of other legit use cases for "AI", which often come down to niche ML applications. Generating age-advanced images for missing persons, for example, is a very valuable tool that avoids artistic bias. But like lots of other technical buzzwords (remember blockchain?) the actual usefulness is usually reserved to a handful of use cases. And I don't happen to have any of those in my life.
It's become more efficient then a Google search these days. But that might be Google just getting so bad.