Both of those sound kinda dystopian. Because you just know the first one will start getting gamed by every company from the grocery companies trying to SEO the AI, to the big fossil fuel companies trying to get you to drive your car more.
How is making a picture of me as an astronaut “dystopian”?
The same technology can be used for widespread, low-cost, highly convincing misinformation and propaganda campaigns
The moon landing wasn’t faked, but I was there instead of Neil Armstrong. See these pics?
deleted by creator
You think the world will be better when literally anyone can create convincing misinformation and propaganda? Personally I prefer when that power is limited, even if there are still powerful entities that can do it
deleted by creator
I think everything gets better when it’s less centralized
Would it be better if everyone had access to nuclear weapons? Or biohazards?
Some things in this world, the fewer people that have access to them, the better. In a perfect world, we might have nobody have access to those things, but I’ll settle for few rather than many.
deleted by creator
Uses tons of energy which could ironically be used to get you to space for real (a lot more energy but at least you get a real experience).
I can’t wait for the technology to get basic enough where I can roll my own self hosted instance of it without it taking months. Because I can see a way it’s doable without a centralized service to get around that. But for mass consumer level, I can see that becoming true. But this can be applied to every bit of software currently. All of it can be ran by you, if you have time. Hell I’ve got my own cloud (hosted at my home ) music streaming service.
A lot of that is doable now - like, how many grocery stores are even nearby to someone, so writing a custom bit of code to check the website of each, one by one, and looking for previously manually-identified items could be automated.
One major downside is prioritization of large chain stores at the expense of smaller mom & pop ones that don’t maintain a constant inventory system accessible via the web. Someone could even volunteer their time to build them a database backend, but still they’d have to see the value in actually scanning the items every time or else it would quickly fall behind.
Yeah, it wouldn’t be a huge lift if you’re familiar with python.
That’s precisely what I was thinking, but reflecting more on it, I don’t know how well it would handle the webpages, so maybe some other languages mixed in too (I’m out of date, maybe PHP?). If AI writing code worked it would lower the barrier, but I’m not certain we’re quite there yet to trust anything it would create.
Python web scraping is just fine, with the llms you.have the option of either extracting the html and having the LLM read.over that, or having a vision ai OCR the page and make its own decision of what to extract.
Yeah I was going to say, I’ve done similar for clients with regards to competitor pricing
In other words, we need to recognize that the real problem is that companies will always try to game the system for product differentiation/market segmentation purposes, so the real solution is for the government to create and enforce standards.
You don’t need “AI” for that. All you would need is some standardized APIs for the various shops, and you could easily solve this with computer technology from 20 years ago.
The reality is, though, that there are no such APIs. LLMs on the other hand could be a valid tool for the use case.
It’s not that there’s no API. It’s that there’s probably a different API for every single grocery store. And they make random changes and don’t have public documentation. That’s why we need the AI.
Yup, exactly, no standardized APIs.
The stores don’t want you to have easy comparable access to their prices.
They’d quite like it if you just came in, saw that the item you wanted is out of stock, and then just buy some shit you didn’t need.
Yeah, we’re not going to make technology that drives prices down
But they’ll happily give you full access to everything they have if you’re another corpo and you promise to marginally improve their sales anyhow. That’s, sadly, how businesses work.
Indeed. LLMs read with the same sort of comprehension that humans have, so if a supermarket makes their website compatible with humans then it’s also compatible with LLMs. We have the same “API”, as it were.
Can LLMs interpret structured input like html?
Yup. And those that can’t can have a parser pull just the human-readable text out, like a blind person’s screen-reader would do.
deleted by creator
That sounds like an issue with your system prompt. If you’re using an LLM to interpret web pages for price information then you’d want to include instructions about what to do if the information simply isn’t in the web page to begin with. If you don’t tell the AI what to do under those circumstances you can’t expect any specific behaviour because it wouldn’t know what it’s supposed to do.
I suspect from this comment that you haven’t actually worked with LLMs much, and are just going off the general “lol they hallucinate” perception they have right now? I’ve worked with LLMs a fair bit and they very rarely have trouble interpreting what’s in their provided context (as would be the case here with web page content). Hallucinations come from relying on their own “trained” information, which they recall imperfectly and often gets a bit jumbled. To continue using a human analogy, it’s like asking someone to rely on their own memory rather than reading information from a piece of paper.
Or you could just prompt it to not guess prices for articles that don’t exist. Those models are pretty good at following instructions.
deleted by creator
No, that’s why we need regulations to enforce standards.
You just need someone to do it. Here in Austria someone did it: https://heisse-preise.io
It’s only in German and most of the prices aren’t from a public API but crawled from different sources.
It’s open source. Nothing except greed is stopping them from providing something like this.Imagine if instead of building their own bespoke systems, grocery stores (and other places) created an open source software foundation and worked together to produce the software they needed.
I sometimes dream of such things. Less waste, better inventory, customers get to choose inventory based on their wishlist, better prices, then I wake up.
We actually have a small liquor store nearby that really puts stuff on the shelves if you casually mention something you like. But that’s more the exception than the rule.
That’s impressive, and honestly looks like it was quite a bit of work. I wonder how the author finances himself? There doesn’t even seem to be a donation button on the site. I found a lengthy article on Wired but it doesn’t appear to mention how he can afford to do all of this for free.
It’s open source. Nothing except greed is stopping them from providing something like this.
Nothing is stopping anyone from doing this except the amount of work it takes to write and maintain all those data import scripts. I think greed is the wrong word here. It’s not unreasonable to expect some sort of monetary reward for providing a useful public service that actually helps people save money. Everyone’s gotta eat, right?
Actually, you’d be surprised. Instacart has up-to-date price and product data for TONS of grocery stores. And while their API likely isn’t public, they MUST have one in order for their smartphone apps to work.
LLMs are not a good tool for processing data like this. They would be good for presenting that data though.
Make an LLM convert the data into a standardized format for your traditional algorithm.
There’s no way to ensure that data will stay in that standardized format though. A custom model could but they are expensive to train.
Llms are excellent at consuming web data.
Not if you want to ensure the validity of the compiled coupons/discounts. A custom algorithm would be best but data standardization would be the main issue, regardless of how you process it.
What does validity mean in this case? A functionary LLM can follow links and make actions. I’m not saying it’s not “work” to develop your personal bot framework, but this is all doable from the home PC, with a self hosted llm
Edit and of course you’ll need non LLM code to handle parts of the processing, not discounting that
The LLM doesn’t do that though, that the software built around it that does that which is what I’m saying. Its definitely possible to do, but the bulk of the work wouldn’t be the task of the LLM.
Edit: forgot to address validity. By that I mean keeping a standard format and ensuring that the output is actually true given the input. Its not impossible, but its something that requires careful data duration and a really good system prompt.
Llms are great for scraping data
LLMs don’t scrape data, scrapers scrape data. LLMs predict text.
Puppeteer and gpt-vision are decidedly not LLMs
👍
there are no such APIs
Yes there are. You can obtain access to the Kroger API, the Meijer API, the Walmart API, and I’m sure others that I didn’t bother to Google. Failing getting access to the actual APIs, there are tons of web scraper projects that just parse those stores’ websites for product information, and web scrapers are still orders of magnitude more efficient than LLMs.
Instacart has prices for all of these stores and more. Obviously they’re not updating them by hand…
At the cost of huge amounts of wasted energy and the whole litany of concerns that are always co-morbid with AI, but technically yes they could work for this lol. Ideally we’d have standardized APIs and mandated pricing transparency, but unfortunately we live in a capitalist society where that will literally never happen ever.
All you would need is some standardized APIs for the various shops
Stores: “I’m going to stop you right there”
We need somebody to wear a 360 camera and go walk every aisle every day. Use image recognition to get the SKU and price from the labels + estimate stock level. Upload the data to an API that’s accessible to all for like $5/month.
Kind of like the Streetview cameras but for spying on actual in store prices.
Calling it now, some tech bro trust fund kid is going to make a start up for this and call it something markety like fresh4u or some shit. Then when everyone is using it they’ll sell your data to China.
And it’s a service because AI
And the service costs a subscription fee
And the service quality drops once it saturates the market
And the service now contains ads
And the grocery stores can pay to promote their store when it is not the most affordable option
And now it’s not economically feasible to not use their service
This is disgusting, but true.
Upvoted
I was working on this with a friend over 10 years ago but the only grocery store that made a decent effort at organizing their website to be scrapeable was Loblaws and all the others had APIs that cost $100,000
Which is one area ML models might (with the right investment) actually be useful. A model trained to look at web pages and relay information from the content visually like we do would be very powerful. The newer ChatGPT models have visual capabilities, I wonder if you could give it a website screen capture and ask it for prices.
Why would you want a model trained on outdated prices? This is not really something LLMs are particularly suited for.
Maybe to crunch historical data, but not for daily comparisons.Why would the model be trained on outdated prices? I’m not talking about LLMs, but separate model designed to parse visual information - specifically websites - and extract particular elements like prices. My comment about ChataGPT was in reference to the newer models which can relay visual information, I’m not suggesting that would be the right approach for training a new model.
The applications would be broader than just prices - this would allow you to scrape any human-readable website without needing to do bespoke development.
I am not sure, that would work. You could train a model that analyzes data and then feed it the data you want to transform. The data wouldn’t be the training data then but part of your request.
Like you can feed a book into GPT4/5 and then ask questions about it.For what you describe you wouldn’t really need AI just a more or less fuzzy parser (like the scan a receipt, get the prices ocr things). Unless I didn’t get it.
Flipp allows for some of this desired capability now through digital flyer scraping and online feeds, APIs. Maybe things have gotten better on the API side over time.
Pretty sure it’s a Canadian app, coincidentally.
The cheapest way to get groceries in the States has always been do all your grocery shopping in the same store, preferably a discount store like an Aldi, instead of cutting coupons and going to multiple different stores due to the simple fact that the gasoline used for driving around is most likely going to cancel out any saving from shopping around, an unfortunate side effect of America’s car centric infrastructure.
You don’t really need an AI to make this list, plus, I think there are apps that already trying to do exactly that.
However, getting a computer to draw yourself in ridiculous situations (usually with an equally ridiculous number of fingers) is great entertainment.
This kind of small scale optimization is not really the best use case for AI anyway. Considering the actual cost of running that kind of code at a large scale… I’m not convinced the savings are worth it even setting aside the petrol issue.
AI doesn’t need to be in the hands of consumers. It should be a step removed, working behind the scenes to make all those basic foods cheaper before you even go shopping. It should be optimizing supply chains, reducing production costs, and otherwise making us more efficient at a societal level.
Which, well, in some cases it already is. Sadly many companies just use it to optimise their marketing 🙄
going to multiple different stores due to the simple fact that the gasoline used for driving around is most likely going to cancel out any saving from shopping around
I wouldn’t jump to that conclusion. Here in suburbia, there are different stores every couple miles. Figure even a 5-mile detour to go to another store, and that “simple fact” of gasoline used turns out to cost less than a dollar. I save that much on a pair of salad kits by going to one store over another, and it’s really more of a one-mile detour anyway. Plus, there are simply things that one store does better than the other and I like to take advantage of that too.
Seriously. Sale items are often several dollars cheaper per item. It is well worth the time and gas driving to several stores unless they are very far apart, then just roll that into another trip. Some big “what could it cost, 10 dollars?” vibes off that comment.
You also need to factor in opportunity cost or concede that your free time doesn’t have value.
If you value your free time at the same rate that you work hourly, then suddenly it’s very hard to save money by spending more time. If you value free time as overtime equivalent, it gets even worse.
Dude, we all waste more than enough time on any given day that we don’t need to worry about the value of losing a half hour to save tens of dollars on our grocery bill. I can’t imagine anyone using a site like this one is particularly worried about lost productivity during their free time.
It’s not about “lost productivity”. It’s about what you enjoy doing. If you don’t enjoy shopping for food, it’s the same as if it were part of your job.
There are only two logical situations:
-
You dislike shopping - you should go to one store maximum because your time is valuable. Get everything else delivered online. Do something you like in your free time.
-
You like shopping - you should work for a shopping delivery service in your spare time. You can make hundreds of extra dollars and get your own groceries at the same time
Getting value for time is productivity. Up to you if value is in money or enjoyment. Your “logic” seems extreme. I’d have to have some irrational hatred for shopping before I’d spend even more on groceries to get someone else to do it. Similarly, I’d have to have some pretty strong feelings to love it so much I’d take a minimum wage job to do it in my spare time. I think the average person is going to fall firmly in the “if shopping for an extra 30 minutes saves me 20 dollars, I’m doing it” camp.
-
30 minutes for 10 hours and all the unnecessary waste of gasoline? Hard, hard pass. In fact, I’d work so that this was punished, what a waste of a limited resource that harms the environment.
Standard IRS reimbursement rate per mile driven is 67¢ per mile this year, which is essentially the per-mile average cost for driving a car. But like, with this sort of thing everyone has their own personal calculus for what they want to optimize for. Do they want to save as much money as possible? Do they want to have fun while shopping? Do they want to shop as quickly as they can? A lot of people will balance these priorities against each other and come up with a solution that isn’t optimal in any one specific area.
Any generative AI that was trained using the entirety of the Internet is gonna suck as an information tool, since it will have more bad information in it than correct information and its goal isn’t to make sure the info is accurate; its goal is to output text that looks intelligent and isn’t obviously generated by a computer.
Even if you fed it nothing but correct information, it will still end up blending multiple things into a single output, generating inaccurate information.
I don’t want AI that just generates shit anywhere but in a video game. I want a tool that can go through real data and give me the relevant stuff I am asking for. Which was handled better with whatever Google was doing 20 years ago than whatever the fuck AI shit they got going on now.
I don’t want AI that just generates shit
You vastly underestimate the demand for mediocre crap that exists in the world.
Then why not train an AI on the entirety of Wikipedia? I know it’s not all correct, but that should ensure most of the information is decently accurate. Would make for a great tool if it allowed to get the same info but explained in a more casual manner.
I know it’s not all correct, but that should ensure most of the information is decently accurate
The problem is that a generative AI does not generate correct content, it generates associated content. It looks at words/term/tokens that are frequently used together to generate a context, and will extrapolate on that, continuing to provide content that looks the teaching content.
The problem is that this will generate materials that LOOKS LIKE CORRECT material, but it doesn’t generate material that IS CORRECT. Thankfully for AI, those things overlap a lot, but they don’t always.
You need an absolutely insane amount of data to train LLMs. Hundreds of billions to tens of trillions of tokens. (A token isn’t the same as a word, but with numbers this massive it doesn’t even matter for the point.)
Wikipedia just doesn’t have enough data to make an LLM off of, and even if you could do it and get okay results, it’ll only know how to write text in the style of Wikipedia. While it might be able to tell you all about the how different cultures most commonly cook eggs, I doubt you’ll get any recipe out of it that makes sense.
If you were to take some base model (such as llama or gpt) and tune it in Wikipedia data, you’ll probably get a “llama in the style of Wikipedia” result, and that may be what you want, but more likely not.
Would make for a great tool if it allowed to get the same info but explained in a more casual manner.
There’s a simple English Wikipedia.
deleted by creator
Yeah, llms are a really great advancement in language processing, and the ability to let them hook into other systems after sussing out what the user means is legitimately pretty cool.
The issue is that people keep mistaking articulate mimicry of confidence and knowledge as actual knowledge and capability.It’s doubly frustrating at the moment because people keep thinking that llms are what AI is, and not just a type of AI. It’s like how now people hear “crypto” and assume you’re talking about the currency scheme, which is needlessly frustrating if you work in the security sector.
Making a system that looked at your purchase history (no real other way to get that data reliably otherwise), identified the staple goods you bought often and then tried to predict the cadence that you buy them at would be a totally feasible AI problem. Wouldn’t be even remotely appropriate for an llm until the system found the price by (probably) crudely scraping grocery store websites and then wanted to tell you where to go, because they’re good at things like "turn this data into a friendly shopping list message "
To be completely fair, the confusion is because of the marketing. You and I both know that Tesla cars can’t really drive themselves for the same reasons you outlined but the typical person sees “autonomous mode” or “self-driving” applied to what they are buying.
People treat llms like something out of a super hero movie because they’re led to believe it to be the case. The people shoveling in the money based on promises and projections are the root cause.
I would even say llms is an important part of what eventually will become an AI and not a type of AI in itself.
There’s a conflation of terms.
One sense of AI is as artificial intelligence: a huge swath of computer algorithms, techniques and study relating to machines measuring inputs, pulling information from them, and making decisions based on what they deduce. Sometimes it’s little more than a handful of equations that capture how to group things together by similarity. What matters is that it’s demonstrating demonstrating intelligence or some manner of operating on knowledge.
The other sense of AI is as a synonym for “a general purpose intelligent system of at least human level”.
Your phones auto complete is an example of the first sense of AI. The second sense doesn’t exist.
There’s a tendency for people to want to remove the AI label from anything they’re used to, or that isn’t like that second sense.
People are just really bad at prompt engineering and so they aren’t good at getting LLM’s like gemeni and GPT to do what they want
You can train it, within conversations to get good at specific tasks. They’re very useful, you just gotta know how to talk to them
The issue is that it’s a language model. You can go a long way by manipulating language to get useful results but it’s still fundamentally limited by languages inability to perform reason, only to mimic it.
Syntax can only take you so far, and it won’t always take you to the right place. Eventually you need something that can reason about the underlying meaning.
It’s still a computer at the end of the day, just use logic. It responds well to it, you remove it’s ability to be creative and tell it what you want to accomplish
I want people to use the right acronym, its a LLM not an AI.
LLMs are a form of AI.
AI covers so many things that it’s inappropriate to hate “AI” when what you really hate is corporate applications of machine learning.
shut up nerd! /j
You wait till I get home and run your comment through my LLM for a witty response, then you’ll see… You’ll all see!
Initialism.
Ahhh looknat me, living the gimmick.
What if I pronounce it lim?
Who am I to dictate how you cull your friendship circles. You do you.
deleted by creator
You have to understand that normal people have no idea what the difference is, nor do they care. To them, AI is just a term that loosely describes “computer does magic that appears somewhat intelligent”. This type of person would just as likely consider Google Maps to be AI as it would ChatGPT, Alexa, or Siri.
TBF an LLM would not be a solution to their problem.
Yeah, but I want AI to make a picture of you as an astronaut but with two extra domes on the suit so I can see ur sweet tits at the same time.
Ladies and gentlemen…THE INTERNET
If this isn’t the purpose of mass communication, then I don’t want it.
No sexualization for the gimp.
~hes been naughty~
Now THIS is pod racing.
Yeah electronics lately instead of helping people squeezes money from us
Android smartphones are just elaborate ad viewing devices with extra stuff to not throw it out. Apple are subscription milking devices where you effectively pay monthly tribute to be less of a walking data stick. pick ur poison
Crazy thing is… I’d pay for a service like the one mentioned in the post. As long as it was still a net savings of course. Even if I broke even it would still be taking the mental load of doing that off my wife and I.
It’s called a Costco membership
/S
deleted by creator
Name 1 Apple subscription that improves privacy as you are describing.
What I meant is that apps on the App Store are sub based usually. At least the good ones. It’s even rare to have ad revenue, pay to remove ads model. Effectively you have to pay subs to use device to the full on top of hefty apple tax.
On the play store way more apps are free but shitty quality or riddled with ads from top to bottom. data harvest fest + google selling u as product. Truly dystopic OS.
Subscription company vs ad company. Pay a lot to not be enslaved by google and I mean a lot like 4000$ (MBP+iPhone+pods+iPad+tv) + 50+$ monthly
AIs use will never see it reach its full potential because companies are liars and deplorable entities that have historically demonstrated they will screw over everyone and everything for profits.
In another universe, AI would help people. It would tell you that you are eating too much bread. It would alert you that you do not eat enough foods in vitamin A. It would tell you that your late night habits of staying up lead to poor health. It would tell you that going to be before 12am leads to you having much better restorative sleep. It would tell you you’re sitting in one position too long. That if you left now, you’d make it 5m early. The list is endless. A machine always calculating and monitoring your status; in an effort to improve your life.
In our universe it’s going to tell you to buy Anamin’s sleep aid, now 50% off. Then track how much you take it so the company that sell you more. Or pass the data to their “partners” so they can sell you more crap.
Your proposal sounds bad too. Not malicious like AI in this universe, but unbelievably annoying.
Then don’t listen to it. You’re trying to catch cases like this: https://www.reddit.com/r/todayilearned/comments/1d7zbvv/til_that_steve_jobs_had_tendency_to_eat_only_one/
Best we can do is sell you a $200 piece of plastic that promises to but doesn’t actually do these things, then automate your job away.
deleted by creator
I want self organising files and things so bad. I need an algorithm to look through my digital library and fix the metadata.
Look up llama fs
It is, sadly, all very poorly focused on the things that won’t benefit society as a whole, but once again, the ruling class. I really wish AI had not been developed with the intent to make white collar jobs obsolete. If only these same brilliant minds had been focused on robotics and processes that humans don’t want to participate in.
They are, look at modem factory or mine they’re full of robotics, now injuries are incredibly low and many of the most dangerous jobs are as safe as a well run office
We’re in a big transition as new technologies are developed, it’s going to take time but there are are some huge things coming soon - llms and cv are enabling fsctory toolarms to leave the factory so expect cooking arms, repair arms, construction arms, micro factories bringing manufacturing back to local markets… sure it sucks we don’t have it now but don’t hate the early stages of it just because it’s not finished yet.
What does creating pictures of astronaut kittens or videos of perpetual zooming have to do with any of that, let alone creating poems about robots and ninjas?
Because those are great ways of exploring and working with things like image analysis and item identification - they seem like the only thing that exists from a consumer point of view but the reality is they’re off shoots of developing more significant tools and mathematical understanding.
Think of it like Newton, he invented a desktop toy and a window decoration for hippys- his life in terms of products to market is laughablely insignificant, totally reduculous. However by laying the groundwork of modern science in optics and physics he demonstrated methods that could be used to totally revolutionize pretty much every part of life.
Natural language computing is hugely significant, it removes so many class and economic barriers while also benefitting us in the developed world almost as much. We’re going to start seeing ai integrated into sites like Amazon soon alowing you to describe the product you want and actually get search results that match, sounds insignificant but it’s a hugely difficult problem and could have a big effect on things. ‘Answering a A vase suitable for this flower’ or ‘a power adaptor to work with this device’ is a couple of orders more complex than creating a poem about robot ninja.
