LLMs: using statistics to generate reasonable-sounding wrong answers from bad data.
Often the answers are pretty good. But you never know if you got a good answer or a bad answer.
And the system doesn’t know either.
For me this is the major issue. A human is capable of saying “I don’t know”. LLMs don’t seem able to.
Accurate.
No matter what question you ask them, they have an answer. Even when you point out their answer was wrong, they just have a different answer. There’s no concept of not knowing the answer, because they don’t know anything in the first place.
The worst for me was a fairly simple programming question. The class it used didn’t exist.
“You are correct, that class was removed in OLD version. Try this updated code instead.”
Gave another made up class name.
Repeated with a newer version number.
It knows what answers smell like, and the same with excuses. Unfortunately there’s no way of knowing whether it’s actually bullshit until you take a whiff of it yourself.
So instead of Prompt Engineer, the more accurate term should be AI Taste Tester?
From what I’ve seen you’ll need an iron stomach.
They really aren’t. Go ask about something in your area of expertise. At first glance, everything will look correct and in order, but the more you read the more it turns out to be complete bullshit. It’s good at getting broad strokes but the details are very often wrong.
Now imagine someone that doesn’t have your expertise reading that answer. They won’t recognize those details are wrong until it’s too late.
That is about the experience I have. I asked it for factual information in the field I work at. It didn’t gave correct answers. Or, it gave working protocols which were strange and would not be successful.
With proper framework, decent assertions are possible.
- It must cite the source and provide the quote, not just a summary.
- An adversarial review must be conducted
If that is done, the work on the human is very low.
That said, it’s STILL imperfect, but this is leagues better than one shot question and answer
Except LLMs don’t store sources.
They don’t even store sentences.
It’s all a stack of massive N-dimensional probability spaces roughly encoding the probabilities of certain tokens (which are mostly but not always words) appearing after groups of tokens in a certain order.
And all of that to just figure out “what’s the most likely next token”, an output which is then added to the input and fed into it again to get the next word and so on, producing sentences one word at a time.
Now, if you feed it as input a long, very precise sentence taken from a unique piece, maybe you’re luck and it will output the correct next word, but if you already have all that you don’t really need an LLM to give you the rest.
Maybe the “framework” you seek - which is quite akin to a indexer with a natural language interface - can be made with AI, but it’s not something you can do with LLMs because their structure is entirely unsuited for it.
The proper framework does, with data store, indexing and access functions.
The cutting edge work is absolutely using LLMs in post-rag pipelines.
Consumer grade chat interfaces def do not do this.
Edit if you worry about topics like context window, sentence splitting or source extraction, you aren’t using a best in class framework any more.
As I mentioned in another post, about the same topic:
Slapping the words “artificial intelligence” onto your product makes you look like those shady used cars salesmen: in the best hypothesis it’s misleading, in the worst it’s actually true but poorly done.
I think AI has mostly been about luring investors into pumping up share prices rather than offering something of genuine value to consumers.
Some people are gonna lose a lot of other people’s money over it.
I tried to find the advert but I see this on YouTube a lot - an Adobe AI ad which depicts, without shame, AI writing out a newsletter/promo for a business owner’s new product (cookies or ice cream or something), showing the owner putting no effort into their personal product and a customer happily consuming because they were attracted by the thoughtless promo.
How are producers/consumers okay with everything being so mediocre??
How are producers/consumers okay with everything being so mediocre??
“You’re always trying to make everything just a little bit worse so that you can feel good about having a lot more of it. I love it. It’s so human!” - The Good Place
Yes, I’m getting some serious dot-com bubble vibes from the whole AI thing. But the dot-com boom produced Amazon, and every company is basically going all-in in the hope they are the new Amazon while in the end most will end up like pets.com but it’s a risk they’re willing to take.
OpenAI will fail. StabilityAI will fail. CivitAI will prevail, mark my words.
“You might lose all your money, but that is a risk I’m willing to take”
- visionairy AI techbro talking to investors
Investors pump money in a bunch of companies so the chances of at least one of them making it big and paying them back for all the failed investments is almost guaranteed. That’s what taking risks is all about.
Sure, but it SEEMS, that some investors are relying on buzzword and hype, without research and ignoring the fundamentals of investing, i.e. besides the ever evolving claims of the CEO, is the company well managed? What is their cash flow and where is it going a year from now? Do the upper level managers have coke habits?
You’re right, but these fundamentals don’t really matter anymore, investors are buying hype and hoping to sell a bigger hype for more money later.
My doorbell camera manufacturer now advertises their products as using, “Local AI” meaning, they’re not relying on a cloud service to look at your video in order to detect humans/faces/etc. Honestly, it seems like a good (marketing) move.
A lot of it is follow the leader type bullshit. For companies in areas where AI is actually beneficial they have already been implementing it for years, quietly because it isn’t something new or exceptional. It is just the tool you use for solving certain problems.
Investors going to bubble though.
Yeah, can make some products better but most of the products these days that use AI, it doesn’t actually need them. It’s annoying to use products that actively shovel AI when it doesn’t even need it.
Ya know what pfoduct MIGHT be better with AI?
Toasters. They have ONE JOB, and everybody agrees their toaster is crap. But you’re not going to buy another toaster, because that too will be crap.
How about a toaster, that accurately, and evenly toasts your bread, and then DOESN’T give you a heart attack at 5am when you’re still half asleep???
IS THAT TOO MUCH TO ASK???
This is the visionary we need. Take my venture capital millions on a magic carpet ride, time traveler!
Nah. We already have AI toasters, and they’re ambitious, but rubbish.
Adding AI is just serious overkill for a toaster, especially when it wouldn’t add anything meaningful, not compared to just designing the toaster better.
It only needs one string of conditions that it can understand: don’t catch on fire. Turn yourself off IF smoke.
Sweet, I’m the one who gets to link the obligatory Technology Connections toaster video!
Aw man, now I want this toaster.
I said the exact same thing months ago when I saw that video. I don’t even use a toaster.
Definitely. Many companies have implemented AI without thinking with 3 brain cells.
Great and useful implementation of AI exists, but it’s like 1/100 right now in products.
My old company before they laid me off laid off our entire HR and Comms teams in exchange for ChatGPT Enterprise.
“We can just have an AI chatbot for HR and pay inquiries and ask Dall-e to create icons and other content”.
A friend who still works there told me they’re hiring a bunch of “prompt engineers” to improve the quality of the AI outputs haha
That’s an even worse ‘use case’ than I could imagine.
HR should be one of the most protected fields against AI, because you actually need a human resource.
And “prompt engineer” is so stupid. The “job” is only necessary because the AI doesn’t understand what you want to do well enough. The only productive guy you could hire would be a programmer or something, that could actually tinker with the AI.
I’m sorry. Hope you find a better job, on the inevitable downswing of the hype, when someone realizes that a prompt can’t replace a person in customer service. Customers will invest more time, i.e., even wait in a purposely engineered holding music hell, to have a real person listen to them.
God that sounds like hell.
If my employer is anything to go by, much of it is just unimaginative businesspeople who are afraid of missing out on what everyone else is selling.
At work we were instructed to shove ChatGPT into our systems about a month after it became a thing. It makes no sense in our system and many of us advised management it was irresponsible since it’s giving people advice of very sensitive matters without any guarantee that advice is any good. But no matter, we had to shove it in there, with small print to cover our asses. I bet no one even uses it, but sales can tell customers the product is “AI-driven”.
No shit, because we all see that AI is just technospeak for “harvest all your info”.
More like “instead of making something that gets the job done, expect pur unfinished product to complain and not do whatever it’s supposed to”. Or just plain false advertising.
Either way, not a good look and I’m glad it’s not just us lemmings who care.
Not to mention it’s usually dog shit out put
I refuse to use Facebook anymore, but my wife and others do. Apparently the search box is now a Meta AI box, and it pisses them every time. They want the original search back.
That’s another thing companies don’t seem to understand. A lot of them aren’t creating new products and services that use ai, but are removing the existing ones, that people use daily and enjoy, and forcing some ai alternative. Of course people are going to be pissed off!
Market shows that investors are actively turned on by products that use AI
Market shows that the market buys into hype, not value.
Market shows that hype is a cycle and the AI hype is nearing its end.
How can you tell when the cycle is ending?
When one of two things happens:
- A new hype starts to replace it (can happen fast though!)
- The hype starts to specialize into subcategories of the hype (e.g. AI images, AI videos, AI text generation)
When “AI” hype dies down we are likely to see “AI” removed from various topics because enough people know and understand the hyped parent topic. It’ll just be “image generation”, “video generation”, “generated text”, etc.
There are different types of people in the market. The informed ones hate AI, and the uninformed love it. The informed ones tend to be the cornerstones of businesses, and the uninformed ones tend to be in charge.
So we have… All this. All this nonsense. All because of stupid managers.
It’s the new block chain or NFT hype, they think it’s magic.
But what if it actually is magic this time? Just this once!? And we miss the hype train?! (This is a sarcastic impression of real conversations I have had.)
Customers worry about what they can do with it, while investors and spectators and vendors worry about buzzwords. Customers determine demand.
Sadly what some of those customers want to do is to somehow improve their own business without thinking, and then they too care about buzzwords, that’s how the hype comes.
Hi, I’m annoying and want to be helpful. Am I helpful? If I repeat the same options again when you’ve told me I’m not helpful, will that be helpful? I won’t remember this conversation once it’s ended.
Hi, which option have you told me you already don’t want would you like?
Sorry, I didn’t quite catch that, please rage again.
For the love of god, defund MBAs.
They’ve overhyped the hell out of it and slapped those letters on everything including a lot of half baked ideas. Of course people are tired of it and beginning to associate ai with bad marketing.
This whole situation really does feel dotcommish. I suspect we will soon see an ai crash, then a decade or so later it will be ubiquitous but far less hyped.
Thing is, it already was ubiquitous before the AI “boom”. That’s why everything got an AI label added so quickly, because everything was already using machine learning! LLMs are new, but they’re just one form of AI and tbh they don’t do 90% of the stuff they’re marketed as and most things would be better off without them.
I get AI has its uses but I don’t need my mouse to have any thing AI related (looking at you Logitech).
LLM based AI was a fun toy when it first broke. Everyone was curious and wanted to play with it, which made it seem super popular. Now that the novelty has worn off, most people are bored and unimpressed with it. The problem is that the tech bros invested so much money in it and they are unwilling to take the loss. They are trying to force it so that they can say they didn’t waste their money.
Honestly they’re still impressive and useful it’s just the hype train overload and trying to implement them in areas they either don’t fit or don’t work well enough yet.
Even in areas where they would fit it’s really annoying how some companies are trying to push it down our throats.
It’s always some obnoxious UI element, screaming at me their 3 example questions, and I always sigh and think, “I have to assume you can only answer these 3 particular questions, and why would I ask those questions, and when I ask UI questions I expect precise answers so would I want to use AI for that.”
I have no doubt that LLM’s have more uses than I can think of, but come on…
I’m happy for studies like this. People who are trying to smear their AI all over our faces need to calm, the f…k, down.
AI does a good job of generating character portraits for my TTRPG games. But, really, beyond that I haven’t found a good use for it.
…also TTRPH, TTRPI, TTRPJ, TTRPK, TTRPL, TTRPM, TTRPN, TTRPO, TTRPP, TTRPQ, TTRPR, TTRPS, TTRPT, TTRPU, TTRPV, TTRPW, TTRPX, TTRPY and TTRPZ games.
But beyond that, no good use, no siree.
PS: spoiler
that was WAY harder to type than I expected.
Many of us who are old enough saw it as an advanced version of ELIZA and used it with the same level of amusement until that amusement faded (pretty quick) because it got old.
If anything, they are less impressive because tricking people into thinking a computer is actually having a conversation with them has been around for a long time.
So you want to tell me they all spent billions and made huge data centres that suck more power than small country so we can all play with it, generate some cringy smut and then toss it away?
This is kinda insane if that’s how it will play out
I have no qualms about AI being used in products. But when you have to tell me that something is “powered by AI” as if that’s your main selling point, then you do not have a good product. Tell me what it does, not how it does it.
I literally uninstalled and disabled every AI process and app in that latest galaxy AI update, which was the whole update btw. my reasons are:
1- privacy and data sharing.
2- the battery, cpu, ram of AI bloatware running in the background 247.
3- it was chaging and doing things which I didn’t want especially in the galary photo albums and camera AI modes.
I was considering a new Samsung phone - is that baked into it? (Assuming you’re talking Samsung anyway, based on the galaxy name)
Samsung is a nightmare, don’t purchase their products.
For example: I used to have a Samsung phone. If I plugged it into the USB port on my computer Windows Explorer would not be able to see it to transfer files. My phone would tell me I need to download Samsung’s drivers to transfer files. I could only get them by downloading Samsung’s software. Once I installed the software Windows Explorer was able to see the device and transfer files. Once I uninstalled the software Windows Explorer couldn’t see the device again.
Anything Samsung can do in your region to insert themselves between you and what you are trying to do they will do.
There are even companies slapping AI labels onto old tech with timers to trick people into buying it.
That one DankPods video of the “AI Rice cooker” comes to mind
For what it’s worth, rice cookers have been touting “fuzzy logic” for like 30 years. The term “AI” is pretty much the same, it just wasn’t as buzzy back then.
Take the hint, MBAs.
They don’t care. At the moment AI is cheap for them (because some other investor is paying for it). As long as they believe AI reduces their operating costs*, and as long as they’re convinced every other company will follow suit, it doesn’t matter if consumers like it less. Modern history is a long string of companies making things worse and selling them to us anyway because there’s no alternatives. Because every competitor is doing it, too, except the ones that are prohibitively expensive.
[*] Lol, it doesn’t do that either
I mean, pretty obvious if they advertise the technology instead of the capabilities it could provide.
Still waiting for that first good use case for LLMs.
Haven’t you been watching the Olympics and seen Google’s ad for Gemini?
Premise: your daughter wants to write a letter to an athlete she admires. Instead of helping her as a parent, Gemini can magic-up a draft for her!
On the plus side for them, they can probably use Gemini to write their apology blog about how they missed the mark with that ad.
I think the LLM could be decent at the task of being a fairly dumb personal assistant. An LLM interface to a robot that could go get the mail or get you a cup of coffee would be nice in an “unnecessary luxury” sort of way. Of course, that would eliminate the “unpaid intern to add experience to a resume” jobs. I’m not sure if that’s good or bad,l. I’m also not sure why anyone would want it, since unpaid interns are cheaper and probably more satisfying to abuse.
I can imagine an LLM being useful to simulate social interaction for people who would otherwise be completely alone. For example: elderly, childless people who have already had all their friends die or assholes that no human can stand being around.
Is that really an LLM? Cause using ML to be a part of future AGI is not new and actually was very promising and the cutting edge before chatGPT.
So like using ML for vision recognition to know a video of a dog contains a dog. Or just speech to text. I don’t think that’s what people mean these days when they say LLM. Those are more for storing data and giving you data in forms of accurate guesses when prompted.
ML has a huge future, regardless of LLMs.
Llm’s are ML…or did I miss something here?
Yes. But not all Machine Learning (ML) is LLM. Machine learning refer to the general uses of neural networks while Large Language Models (LLM) refer more to the ability for an application, or a bot, to understand natural language and deduct context from it, and act accordingly.
ML in general as a much more usages than only power LLM.
Wrote my last application with chat gpt. Changed small stuff and got the job
Please write a full page cover letter that no human will read.
